CN113486989A

CN113486989A - Knowledge graph-based object recognition method and device, readable medium and equipment

Info

Publication number: CN113486989A
Application number: CN202110892035.0A
Authority: CN
Inventors: 刘江冬; 钟黎; 易坤
Original assignee: Beijing ByteDance Network Technology Co Ltd
Current assignee: Beijing ByteDance Network Technology Co Ltd
Priority date: 2021-08-04
Filing date: 2021-08-04
Publication date: 2021-10-08
Anticipated expiration: 2041-08-04
Also published as: CN113486989B

Abstract

The disclosure relates to a method, a device, a readable medium and equipment for identifying an object based on a knowledge graph, relating to the technical field of electronic information processing, wherein the method comprises the following steps: determining a target game vector for representing a target game and an object vector to be recognized for representing an object to be recognized according to a pre-established game knowledge graph, determining the correlation degree of the object to be recognized and the target game according to the object vector to be recognized, the target game vector and a pre-trained recognition model, wherein the recognition model is obtained by training a seed object vector for representing a seed object of the target game and the target game vector, the seed object vector is determined according to the game knowledge graph, and if the correlation degree of the object to be recognized and the target game meets a preset condition, determining that the object to be recognized is the target object of the target game. According to the method and the device, the target object can be effectively identified without interaction between the object to be identified and the target game, and the efficiency and accuracy of object identification are improved.

Description

Knowledge graph-based object recognition method and device, readable medium and equipment

Technical Field

The present disclosure relates to the field of electronic information processing technologies, and in particular, to a method, an apparatus, a readable medium, and a device for object recognition based on a knowledge graph.

Background

With the continuous development of electronic information technology, various game-like applications are appearing in the application market. In the game operation process, the exposure of the game is increased by means of putting multimedia contents, so that the active amount of the game is increased. In order to improve the accuracy of the putting, a target putting platform suitable for the game is usually selected from a plurality of putting platforms, and the putting is oriented on the target putting platform. In the prior art, a method for identifying a target object and accurately delivering the target object based on a knowledge graph exists, however, in this way, an interactive behavior exists between the object and a game, and for a new game on shelf, the object and the game often do not have the interactive behavior or have few interactive behaviors, and the knowledge graph has the problem of cold start, so that the target object cannot be identified.

Disclosure of Invention

This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.

In a first aspect, the present disclosure provides a method for knowledge-graph based object recognition, the method comprising:

determining a target game vector for representing a target game and an object vector to be recognized for representing an object to be recognized according to a game knowledge graph established in advance;

determining the correlation degree of the object to be recognized and the target game according to the object vector to be recognized, the target game vector and a pre-trained recognition model, wherein the recognition model is obtained by training a seed object vector used for representing the target game seed object and the target game vector, and the seed object vector is determined according to the game knowledge graph;

and if the correlation degree of the object to be recognized and the target game meets a preset condition, determining that the object to be recognized is the target object of the target game.

In a second aspect, the present disclosure provides a knowledge-graph based object recognition apparatus, the apparatus comprising:

the vector determination module is used for determining a target game vector for representing a target game and an object vector to be identified for representing an object to be identified according to a game knowledge graph established in advance;

the relevancy determining module is used for determining the relevancy of the object to be recognized and the target game according to the object vector to be recognized, the target game vector and a pre-trained recognition model, wherein the recognition model is obtained by training a seed object vector of a seed object used for representing the target game and the target game vector, and the seed object vector is determined according to the game knowledge graph;

and the identification module is used for determining that the object to be identified is the target object of the target game if the correlation degree of the object to be identified and the target game meets a preset condition.

In a third aspect, the present disclosure provides a computer readable medium having stored thereon a computer program which, when executed by a processing apparatus, performs the steps of the method of the first aspect of the present disclosure.

In a fourth aspect, the present disclosure provides an electronic device comprising:

a storage device having a computer program stored thereon;

processing means for executing the computer program in the storage means to implement the steps of the method of the first aspect of the present disclosure.

According to the technical scheme, the method comprises the step of determining the object vector to be identified and the target game vector according to the game knowledge graph established in advance. And then determining the correlation degree of the object to be recognized and the target game according to the vector of the object to be recognized, the vector of the target game and a pre-trained recognition model, wherein the recognition model is obtained by training according to the vector of the seed object used for representing the seed object of the target game and the vector of the target game. And finally, determining the object to be recognized as the target object of the target game under the condition that the correlation degree of the object to be recognized and the target game meets the preset condition. According to the method, the vectors for representing the object to be recognized and the target game are obtained through the game knowledge graph, and the vectors are recognized through the recognition model, so that whether the object to be recognized is the target object or not is determined. The target object can be effectively identified without interaction between the object to be identified and the target game, the cold start problem of object identification is solved, and the efficiency and accuracy of object identification are improved.

Additional features and advantages of the disclosure will be set forth in the detailed description which follows.

Drawings

The above and other features, advantages and aspects of various embodiments of the present disclosure will become more apparent by referring to the following detailed description when taken in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numbers refer to the same or similar elements. It should be understood that the drawings are schematic and that elements and features are not necessarily drawn to scale. In the drawings:

FIG. 1 is a flow diagram illustrating a method of knowledge-graph based object recognition in accordance with an exemplary embodiment;

FIG. 2 is a schematic diagram illustrating a game knowledge-graph according to an exemplary embodiment;

FIG. 3 is a flow diagram illustrating training a recognition model in accordance with an exemplary embodiment;

FIG. 4 is a flow diagram illustrating another method of training a recognition model in accordance with an illustrative embodiment;

FIG. 5 is a diagram illustrating a recognition model in accordance with an exemplary embodiment;

FIG. 6 is a flow diagram illustrating the establishment of a game knowledge graph in accordance with an exemplary embodiment;

FIG. 7 is another flow diagram illustrating the establishment of a game knowledge graph in accordance with an exemplary embodiment;

FIG. 8 is a flow diagram illustrating another method of knowledge-graph based object recognition in accordance with an exemplary embodiment;

FIG. 9 is a schematic diagram illustrating an object game atlas according to an example embodiment;

FIG. 10 is a block diagram illustrating a knowledge-graph based object recognition apparatus in accordance with an exemplary embodiment;

FIG. 11 is a block diagram illustrating another knowledge-graph based object recognition apparatus in accordance with an exemplary embodiment;

FIG. 12 is a block diagram illustrating another knowledge-graph based object recognition apparatus in accordance with an exemplary embodiment;

FIG. 13 is a block diagram illustrating an electronic device in accordance with an example embodiment.

Detailed Description

Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While certain embodiments of the present disclosure are shown in the drawings, it is to be understood that the present disclosure may be embodied in various forms and should not be construed as limited to the embodiments set forth herein, but rather are provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the disclosure are for illustration purposes only and are not intended to limit the scope of the disclosure.

It should be understood that the various steps recited in the method embodiments of the present disclosure may be performed in a different order, and/or performed in parallel. Moreover, method embodiments may include additional steps and/or omit performing the illustrated steps. The scope of the present disclosure is not limited in this respect.

The term "include" and variations thereof as used herein are open-ended, i.e., "including but not limited to". The term "based on" is "based, at least in part, on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Relevant definitions for other terms will be given in the following description.

It should be noted that the terms "first", "second", and the like in the present disclosure are only used for distinguishing different devices, modules or units, and are not used for limiting the order or interdependence relationship of the functions performed by the devices, modules or units.

It is noted that references to "a", "an", and "the" modifications in this disclosure are intended to be illustrative rather than limiting, and that those skilled in the art will recognize that "one or more" may be used unless the context clearly dictates otherwise.

The names of messages or information exchanged between devices in the embodiments of the present disclosure are for illustrative purposes only, and are not intended to limit the scope of the messages or information.

FIG. 1 is a flow chart illustrating a method of knowledge-graph based object recognition, as shown in FIG. 1, according to an exemplary embodiment, the method comprising the steps of:

step 101, according to a game knowledge graph established in advance, a target game vector for representing a target game and an object vector to be identified for representing an object to be identified are determined.

Wherein, the game knowledge graph comprises a plurality of nodes and at least one edge, and the plurality of nodes comprise: the object node, the game node and the content node, wherein each edge is used for representing that two nodes at two ends of the edge have correlation.

For example, to identify a target object of a target game from a large number of objects, a game knowledge graph may be first established. It is understood that a game ecology comprises a plurality of objects, games and contents, and the game knowledge graph can represent the game ecology and reflect the association among the objects, games and contents. The game knowledge graph can comprise a plurality of nodes, and the nodes are divided into three types: an object node for representing an object, a game node for representing a game, and a content node for representing content. The object may be understood as a delivery platform for delivering multimedia content, and the user may download a game, enter a game, and the like through the delivery platform. The delivery platform may be, for example, one application program, a group of application programs corresponding to the same server, one page in the application program, or the like. An object may also be understood as a drop zone within which a user may download a game, enter a game, etc. The delivery area may be, for example, an area covered by a local area network, an area covered by a base station, an area provided by an operator for providing services, and the like. An object may also be understood as a terminal device through which a user may download a game, enter a game, etc. The object may also be understood as a user, and the present disclosure is not particularly limited to the specific meaning of the object. Content is to be understood as any content exposed by an object, such as any of text (e.g. news, novels, blogs, etc.), audio (e.g. music, radio, audio books, etc.), video (e.g. movies, television shows, short videos, etc.). The game knowledge graph further comprises at least one edge, if any two nodes have an association, an edge exists between the two nodes, namely each edge can represent that the two nodes at the two ends of the edge have the association, and further, the width or the value of each edge can also represent the attribute of the association between the two nodes at the two ends of the edge. For example, there is an edge between the content node representing content a and the game node representing game a, which is used to characterize content a to which game a is referred. Further, if the content a is a blog where 70% of the text is describing the game a, then the value of the edge may be 0.7. In one implementation, each node may further include an attribute of the node (e.g., an object node includes a representation of a corresponding object), and in another implementation, the attribute of each node may be used as an attribute node and an edge may be established between the attribute node and the corresponding node. Taking the game knowledge map shown in fig. 2 as an example, the game knowledge map includes: a node A representing an object A, a node B representing an object B, a node A representing a game A, a node B representing a game B, a node a representing a content a, a node B representing a content B, a node A1 representing an image of an object A, a node A1 representing a label of a game A, and a node a1 representing a label of a content a. The one-way arrow is arranged between the node a1 and the node a, which indicates that the node a1 is an attribute node of the node a, the two-way arrow is arranged between the node b and the node a, which indicates that the node a and the node b have an association (for example, in a scenario in which an object is an application program, the object a and the object b may use the same server). A unidirectional arrow is between node A and node A1, indicating that node A1 is an attribute node of node A, and a bidirectional arrow is between node B and node A, indicating that node A and node B have similar associations (e.g., game A and game B belong to the same game developer). Between node a and node a1 is a unidirectional arrow indicating that node a1 is an attribute node of node a, and between node b and node a is a bidirectional arrow indicating that there is a similar association between node a and node b (e.g., content a and content b describe the same game). It should be noted that, in the scenario where the object is a user, the information required for establishing the game knowledge graph is obtained under the authorization of the user, or is actively submitted after the user reads the relevant description, or is inevitably sent to the server by the terminal device when the user uses the terminal device.

To identify an object to be identified, an object vector to be identified for characterizing the object to be identified and a target game vector for characterizing a target game may be determined by a game knowledge graph. The object to be recognized can be an object represented by any object node in the game knowledge graph. In one implementation, a vector corresponding to each node in the game knowledge Graph may be determined by using a Graph Neural Network (GNN), a Graph Convolutional Neural Network (GCN), a Graph space Network (GCN), or a Graph sage (Graph simple and aggreGatE), so as to obtain an object vector to be recognized corresponding to an object node representing an object to be recognized and a target game vector corresponding to a game node representing a target game. In another implementation mode, a plurality of sub-graph spectrums capable of reflecting the association of multiple dimensions such as objects, contents, games and the like among the game nodes can be established according to the game knowledge graph, and a vector corresponding to each game node is determined by using a preset graph representing algorithm and the plurality of sub-graph spectrums. And establishing a plurality of sub-graph spectrums capable of reflecting the association of a plurality of dimensions such as games, contents and the like between each object node and the game nodes according to the game knowledge graph, and finally determining a vector corresponding to each object node by combining the vector corresponding to each game node and the plurality of sub-graph spectrums, thereby obtaining an object vector to be identified corresponding to the object node representing the object to be identified and a target game vector corresponding to the game node representing the target game.

And 102, determining the correlation degree of the object to be recognized and the target game according to the object vector to be recognized, the target game vector and a pre-trained recognition model, wherein the recognition model is obtained by training a seed object vector for representing the target game seed object and the target game vector, and the seed object vector is determined according to a game knowledge graph.

And 103, if the correlation degree of the object to be recognized and the target game meets a preset condition, determining that the object to be recognized is the target object of the target game.

For example, a seed object vector for characterizing a seed object of a target game may be determined based on a game knowledge graph, and then a recognition model may be pre-trained using the seed object vector and the target game vector, where a recognition model may be understood as a model that is capable of recognizing the relevance of any one object to the target game. The recognition model may be a classification model in the lookelike mode, and may be, for example: random Forest (RF), Adaboost model, Xgboost model, etc., and the recognition model may also be a Convolutional Neural Network (CNN), which is not limited in this disclosure. The plurality of seed objects can be understood as a launching platform for launching the multimedia content of the target game, a launching platform for providing a target game access interface, a terminal device provided with the target game, or a user who downloads and uses the target game. For example, before the target game is put on shelf, some users may try to play the target game by using internal testing and the like, and the some users are the seed objects. For another example, before the target game is put on shelf, the multimedia content of the target game may be played on a part of the playing platforms, and the playing platforms may be used as seed objects. The recognition model may be trained with the seed object vector as a positive sample and the object vector of the non-seed object as a negative sample. After the recognition model is obtained through training, the object vector to be recognized and the target game vector can be input into the recognition model, so that the correlation degree of the object to be recognized and the target game can be determined according to the output of the recognition model. It is understood that the recognition model can output a degree of correlation score of the object to be recognized with the target game as a degree of correlation of the object to be recognized with the target game.

And finally, if the correlation degree of the object to be recognized and the target game meets a preset condition, determining that the object to be recognized is the target object of the target game. The preset condition may be greater than or equal to a preset threshold, and if the correlation between the object to be recognized and the target game is greater than or equal to the preset threshold, it may be determined that the object to be recognized is the target object. The preset condition may also be a maximum specified number (for example, 1000) of degrees of correlation, and the degrees of correlation between the plurality of objects to be recognized and the target game may be determined respectively, and then the specified number of objects to be recognized with the maximum degree of correlation may be set as the target objects. After the target object is determined, the multimedia content of the target game may be targeted for the target object.

The game knowledge graph represents the ecology of the whole game and can reflect the association among objects, games and contents, so that the interaction behavior between the object to be identified and the target game is not needed, and meanwhile, the vectors capable of representing the object to be identified and the target game are utilized, so that the target object is effectively identified, the cold start problem of object identification is solved, and the efficiency of object identification is improved. Meanwhile, the game knowledge graph can reflect the association among the objects, games and contents, so that the object vector to be recognized and the target game vector represent the object to be recognized and the target game from three dimensions of the objects, the games and the contents, for the recognition model, more dimensional information is obtained for recognition, the correlation degree of the object to be recognized and the target game can be recognized more accurately, and the accuracy of object recognition is improved.

In summary, the present disclosure first determines the object vector to be identified and the target game vector according to the game knowledge graph established in advance. And then determining the correlation degree of the object to be recognized and the target game according to the vector of the object to be recognized, the vector of the target game and a pre-trained recognition model, wherein the recognition model is obtained by training according to the vector of the seed object used for representing the seed object of the target game and the vector of the target game. And finally, determining the object to be recognized as the target object of the target game under the condition that the correlation degree of the object to be recognized and the target game meets the preset condition. According to the method, the vectors for representing the object to be recognized and the target game are obtained through the game knowledge graph, and the vectors are recognized through the recognition model, so that whether the object to be recognized is the target object or not is determined. The target object can be effectively identified without interaction between the object to be identified and the target game, the cold start problem of object identification is solved, and the efficiency and accuracy of object identification are improved.

In one implementation, the implementation of step 102 may be:

and inputting the vector of the object to be recognized, the vector of the target game and the object characteristics of the object to be recognized into the recognition model to obtain the correlation degree between the object to be recognized and the target game, which are output by the recognition model, wherein the object characteristics are determined according to the object information of the object to be recognized.

For example, to identify the object to be identified, the object feature may be determined according to the object information of the object to be identified. In the scene that the object is the user, the object information is obtained under the condition that the authorization of the user is obtained, or is actively submitted after the user reads the relevant description, or the terminal equipment is inevitably sent to the server when the user uses the terminal equipment. Furthermore, data related to personal attributes in the object information are all subjected to desensitization processing, for example, a certain type of data may be partially hidden, or a certain type of data may be segmented, and the like. The object characteristics may be understood as a statistical indicator or a trend of change that can reflect the activity of the object to be identified, and may include one or more characteristics. The object features may include multiple features of the object at different stages. For example, in a scene in which the object is a user, the time length of browsing the content related to the target game by the object to be identified, which is obtained according to the statistics of the object information, may be used as the object feature, the frequency of browsing the content related to the target game by the object to be identified, which is obtained according to the statistics of the object information, may be used as the object feature, and the mode (e.g., screen-on mode, information flow) and channel (e.g., different APPs) of browsing the content related to the target game by the object to be identified may be used as the object feature. In a scene in which the object is a launching platform, the duration of the content related to the target game, which is shown by the object to be identified and obtained according to the object information statistics, can be used as the object characteristic, and the frequency of the content related to the target game, which is shown by the object to be identified and obtained according to the object information statistics, can also be used as the object characteristic. The present disclosure does not specifically limit the types of specific object features and the manner of acquiring the object features. After the object features are obtained, the object features, the object vector to be recognized and the target game vector can be spliced and input into the recognition model together, so that the correlation degree between the object to be recognized and the target game output by the recognition model can be obtained. For the recognition model, besides obtaining the object vector to be recognized and the target game vector, the object characteristics are combined, the degree of correlation between the object to be recognized and the target game can be recognized more accurately, and the accuracy of object recognition is further improved.

Fig. 3 is a flowchart illustrating training a recognition model according to an exemplary embodiment, where object nodes include object nodes corresponding to a plurality of sample objects, the sample objects include positive sample objects and negative sample objects, and the positive sample objects include seed objects, as shown in fig. 3. The method further comprises the following steps:

and 104, training the recognition model aiming at the target game.

Specifically, the implementation manner of step 104 may be:

step 1041, obtaining a sample input set, the sample input set comprising: a sample input corresponding to each sample object, the sample input comprising: an object vector and a target game vector for characterizing the sample object, as determined from the game knowledge graph.

Step 1042, obtaining a sample output set, where the sample output set includes a sample output corresponding to each sample input, and each sample output includes a real recognition result of a corresponding sample object.

And 1043, taking the sample input set as the input of the recognition model, and taking the sample output set as the output of the recognition model, so as to train the recognition model.

For example, when training the recognition model in the above embodiments, a sample input set and a sample output set need to be obtained first. The sample input set comprises sample input corresponding to each sample object in a plurality of sample objects, and the sample input corresponding to the sample object is an object vector and a target game vector of the sample object. It should be noted that the plurality of sample objects include a plurality of positive sample objects and a plurality of negative sample objects, the positive sample objects may include seed objects, and further, a ratio (for example, may be 1: 1) between the number of positive sample objects and the number of negative sample objects may be controlled. The sample output set includes a sample output corresponding to each sample input, each sample output including a true recognition result of the corresponding sample object. Wherein, the true recognition result of the positive sample object is correct (can be represented as 1), that is, the positive sample object is the target object of the target game, and the true recognition result of the negative sample object is wrong (can be represented as 0), that is, the negative sample object is not the target object of the target game. The sample input set may then be trained as an input to the recognition model and the sample output set as an output to the recognition model, such that when the sample input set is input, the output of the recognition model matches the sample output set. For example, the parameters of the neurons in the recognition model, such as weights (in English: Weight) and offsets (in English: Bias) of the neurons, can be modified by a back propagation algorithm with the goal of reducing the loss function according to the output of the recognition model and the difference (or mean square error) from the sample output set as the loss function of the recognition model. And repeating the steps until the loss function meets a preset condition, for example, the loss function is smaller than a preset loss threshold.

In one application scenario, the positive sample objects may further include an extension object, which is determined by:

step 1) determining the correlation degree of other games and the target game according to other game vectors and target game vectors for representing other games, wherein the other games are games except the target game, and the other game vectors are determined according to the game knowledge graph.

And 2) taking other games of which the correlation degree with the target game is greater than or equal to a preset correlation degree threshold value as the related games corresponding to the target game.

And 3) taking the active object of the related game as an extension object.

For example, in general, the seed objects need to be obtained by internal measurement, and when the number of the seed objects is too small or the seed objects cannot be obtained, a cold start problem of the recognition model may occur. Therefore, the correlation degree between the other games and the target game can be calculated by each game vector determined in the game knowledge graph, and then the other games with the correlation degree larger than or equal to the preset correlation threshold value with the target game are used as the corresponding correlation games of the target game, wherein the correlation games can be one or more. The correlation between the other game and the target game may be, for example, a game vector corresponding to the other game and a cosine similarity between the game vector and the target game, or a game vector corresponding to the other game and a Jaccard similarity between the game vector and the target game, which is not limited in this disclosure. Finally, the active object of the related game can be used as an extension object, and the extension object is used as a positive sample object for training the recognition model, so that the cold start problem of the recognition model caused by too few seed objects can be solved. Specifically, the active objects of the related games can be determined directly according to history data acquired by the related games, or the objects meeting specified conditions (for example, the objects are continuously online for N days, or the use time exceeds M hours) can be determined as the active objects, or the active objects can be screened out from the game knowledge graph, wherein edges are formed between object nodes corresponding to the active objects and game nodes corresponding to the related games, and the numerical values of the edges meet the specified conditions.

Fig. 4 is a flowchart illustrating another method for training a recognition model according to an exemplary embodiment, and as shown in fig. 4, the step 1043 may be implemented by:

and step A, clustering the sample input set according to a preset clustering algorithm to obtain a plurality of groups of sample input subsets, wherein each group of sample input subsets corresponds to one group of sample output subsets.

And B, aiming at each group of sample input subsets, inputting the identifier models corresponding to the group of sample input subsets, which are included in the identification model, and using the sample output subsets corresponding to the group of sample input subsets as the output of the identifier models corresponding to the group of sample input subsets so as to train the identifier models corresponding to the group of sample input subsets.

And determining the output of the recognition model according to the output of the recognizer model corresponding to the input subset of each group of samples.

For example, when the sample object covers a plurality of object groups, the sample input set may be clustered according to a preset clustering algorithm to obtain a plurality of groups of sample input subsets, where each group of sample input subsets includes a part of sample inputs in the sample input set, and it may be understood that the part of sample inputs belong to the same object group. Accordingly, each set of sample input subsets corresponds to a set of sample output subsets, which includes the sample outputs corresponding to the portion of sample inputs. The Clustering algorithm may be, for example, DBSCAN (english: Density-Based Spatial Clustering of Applications with Noise, chinese: Density-Based Clustering method with Noise), which is not specifically limited by the present disclosure.

Further, the recognition model may include a recognition submodel corresponding to each set of sample input subsets, that is, the recognition submodel corresponds to the sample input subsets one to one, and the structure of the recognition model may be as shown in fig. 5. For each set of sample input subsets, the corresponding identifier model may be input to the set of sample input subsets, and the sample output subset corresponding to the set of sample input subsets is used as the output of the corresponding identifier model to train the corresponding identifier model, so that when the set of sample input subsets is input, the output of the corresponding identifier model may be matched with the sample output subset corresponding to the set of sample input subsets. For example, the parameters of the neurons in the corresponding recognizer model may be modified by a back propagation algorithm with the goal of reducing the loss function, based on the output of the corresponding recognizer model, and the difference (or mean square error) with the corresponding sample output subset as the loss function of the corresponding recognizer model. And repeating the steps until the loss function meets a preset condition, for example, the loss function is smaller than a preset loss threshold.

Finally, the output of the recognition model may be determined according to the output of the recognition submodel corresponding to each set of sample input subsets, i.e. the recognition model may be understood as a voting system comprising a plurality of recognition submodels. For example, the average value of the outputs of the identifier models corresponding to each group of sample input subsets may be used as the output of the identification model, the maximum value of the outputs of the identifier models corresponding to each group of sample input subsets may be used as the output of the identification model, and the outputs of the identifier models corresponding to each group of sample input subsets may be weighted and summed to be used as the output of the identification model. The present disclosure does not specifically limit this. Because the sample object covers various object groups, the correspondingly trained recognition model has stronger generalization capability, and the correlation degree between the object in the various object groups and the target game can be accurately recognized.

In an application scenario, the identifier model corresponding to each group of sample input subsets is a tree integration model and includes a plurality of tree models. The implementation manner of the step B may include:

first, for each tree model, the set of sample input subsets is randomly sampled to obtain a sample input subset, the sample input subset comprising a smaller number of sample inputs than the set of sample inputs.

Thereafter, each sample input comprised in the subset of sample inputs is randomly sampled to obtain a sample input corresponding to the sample input to which the sample input belongs.

And finally, taking the sample input corresponding to each sample input included in the sampling input subset as the input of the tree model, and taking the sample output corresponding to each sample input included in the sampling input subset as the output of the tree model, so as to train the tree model.

Wherein the output of the tree integration model is determined from the output of each tree model.

For example, each of the above identifier models may be a tree integration model (or referred to as an integration tree model), such as: which may be a random forest, Adaboost model, Xgboost model, etc., including a plurality of tree models, which may be, for example, classification trees. The process of training for each recognizer model may be a process of jointly training a plurality of tree models included in the recognizer model.

For each tree model, the set of sample input subsets may be randomly sampled to obtain a sample input subset, where the sample input subset includes a smaller number of sample inputs than the set of sample inputs, i.e., the sample input subset includes a portion of the sample inputs in the set of sample input subsets. For example, the set of sample inputs includes 100 sample inputs, and the random sampling may include 75 sample inputs in the resulting sample input subset. And then, randomly sampling each sample input included in the sampling input subset to obtain a sampling sample input corresponding to the sample input, wherein the sampling sample input belongs to the sample input, namely the dimension of the sampling sample input is smaller than the dimension of the sample input. For example, the subset of sample inputs includes 75 sample inputs, each sample input being a 256-dimensional vector. Each sample input is randomly sampled, and 75 sample inputs corresponding to the obtained 75 sample inputs are obtained, and each sample input can be a 200-dimensional vector. And then taking the sampling sample input corresponding to each sample input included in the sampling input subset as the input of the tree model, and taking the sample output corresponding to each sample input included in the sampling input subset as the output of the tree model so as to train the tree model, so that when the sampling sample input is input, the output of the corresponding tree model can be matched with the corresponding sample output. For example, the parameters of the neurons in the corresponding tree model may be modified by a back propagation algorithm with the goal of reducing the loss function, based on the output of the corresponding tree model and the difference (or mean square error) with the corresponding sample output as the loss function of the corresponding tree model. And repeating the steps until the loss function meets a preset condition, for example, the loss function is smaller than a preset loss threshold.

Finally, the output of the tree integration model may be determined from the output of each tree model, i.e. the tree integration model may be understood as a voting system comprising a plurality of tree models. For example, the average value of the outputs of the respective tree models may be used as the output of the tree integration model, the maximum value of the outputs of the respective tree models may be used as the output of the tree integration model, or the outputs of the respective tree models may be weighted and summed to be the output of the tree integration model. The present disclosure does not specifically limit this. Because each group of sample input subsets is randomly sampled and each sample input is randomly sampled, the generalization capability of the tree integration model can be improved in two dimensions of sample quantity and sample representation, and the correlation degree of the object and the target game can be accurately identified by the identification model.

FIG. 6 is a flow diagram illustrating the establishment of a game knowledge-graph, as shown in FIG. 6, by the steps of:

and step C, acquiring object information of a plurality of objects, game information of a plurality of games and content information of a plurality of contents. The content is any one of text, audio and video, the plurality of objects comprise objects to be identified and seed objects, and the plurality of games comprise target games.

For example, a game knowledge map may be created prior to identifying the object to be identified. Specifically, object information of a plurality of objects, game information of a plurality of games, and content information of a plurality of contents may be acquired first. Wherein the number of objects, the number of games and the number of contents are not related to each other. It is understood that the game ecology includes a plurality of objects, games and contents, wherein the plurality of objects includes the above-mentioned object to be recognized and the seed object, and may also include other objects. The plurality of games includes the above-mentioned target game and may also include other games (e.g., the related games mentioned above).

Specifically, the object information may include, for example, an object image, an object identifier, an object group type, and the like of the corresponding object. It should be noted that, in a scenario where the object is a user, the object information is obtained under the condition of obtaining the authorization of the corresponding user, or the corresponding user actively submits after reading the relevant description, or the terminal device inevitably sends the object information to the server when the user uses the terminal device. Furthermore, data related to personal attributes in the object information are all subjected to desensitization processing, for example, a certain type of data may be partially hidden, or a certain type of data may be segmented, and the like. The game information may include game tags for the corresponding game (e.g., poker, conversion, battle, basketball, turn, role play, timeliness, etc.), game categories (e.g., chess, cards, games, sports, etc.), game developers, and the like. The content information may comprise content tags, content categories, content providers, etc. of the corresponding content, wherein the content tags and content categories, and the game tags and game categories may be shared, i.e. the respective content tags and content categories may be determined according to the games involved in the content. For example, if game a is involved in content a, the content tag and content category of content a may be the same as game a game tag and game category.

After the object information, the game information, and the content information are obtained, the object information, the game information, and the content information may be subjected to data cleansing and then normalization processing. Specifically, the process of data cleaning may be to delete information with more data loss, and may also be to complement information with less data loss by using an interpolation method. The present disclosure does not specifically limit this.

And D, establishing an object node corresponding to each object, a game node corresponding to each game and a content node corresponding to each content.

And E, establishing edges among the nodes according to a preset association rule.

For example, corresponding nodes can be established in the game knowledge graph for each object, each game and each content, that is, at least three types of nodes are included in the game knowledge graph: an object node for representing an object, a game node for representing a game, and a content node for representing content. Further, the game knowledge graph may further include object attribute nodes for representing object information, game attribute nodes for representing game information, and content attribute nodes for representing content information. Then, according to a preset association rule, an edge can be established between two nodes with the association, so that the game knowledge graph is obtained.

FIG. 7 is another flow diagram illustrating the establishment of a game knowledge graph according to an exemplary embodiment, and as shown in FIG. 7, the implementation of step E may include:

step E1, according to the object information of each object, determining the game related object meeting the association rule with the object, wherein the game related object comprises: at least one of an object, a game and a content, and an edge is established between an object node corresponding to the object and a node corresponding to the game-related object.

Step E2, determining a content associated object satisfying the association rule with each content according to the content information of the content, the content associated object including: games and/or contents, and establishing edges between content nodes corresponding to the contents and nodes corresponding to the content-associated objects.

And E3, determining the associated games meeting the association rules with the games according to the game information of each game, and establishing edges between the game nodes corresponding to the games and the game nodes corresponding to the associated games.

The specific manner of establishing edges in the game knowledge graph may be to determine association rules first. The association rules may include associations of multiple dimensions: first association between objects: such as associations between objects, similarity of object information of objects to objects exceeding a first threshold (e.g., 75%), etc. Second association between object and game: such as object download, use, and consumption of the game. Third association between object and content: such as object browsing, commenting, praising, sharing content. Fourth association of content with content: for example, the similarity between the content and the content information of the content exceeds a second threshold (e.g., 60%), and the content information of the content include the same content tag (or content category, content provider). Fifth association between content and game: for example, the similarity of the content information of the content and the game information of the game exceeds a third threshold (e.g., 50%). Game to game sixth association: for example, the similarity of the game information of the game and the game exceeds a fourth threshold (e.g., 80%), and the same game tag (or game category, game developer) is included in the game information of the game and the game.

Then, the objects, games, and contents (i.e. game-related objects) satisfying the association rule (e.g. first association, second association, and third association) with each object can be determined according to the association rule, and corresponding edges can be established. Determine the game, content (i.e., content associated object) that satisfies the association rule (e.g., fourth association, fifth association) with each content, and establish the corresponding edge. Associated games that satisfy the association rule (i.e., the sixth association) with each game are determined and corresponding edges are established. In this way, the edges in the game knowledge graph can describe the association between objects, games and contents from multiple dimensions.

It should be noted that the execution sequence in the above embodiments is only for illustration, and the execution sequence between the step E1, the step E2, and the step E3 may be set according to specific requirements, and the disclosure is not limited in this regard.

Fig. 8 is a flowchart illustrating another method for knowledge-graph based object recognition according to an exemplary embodiment, and as shown in fig. 8, the implementation of step 101 may include:

at step 1011, game vectors characterizing each game are determined from the game knowledge graph.

For example, the game vectors used to characterize each game may be determined based on the game knowledge graph. Specifically, the method can be realized by the following steps:

and 4) determining an object game map according to the edges between the object nodes and the game nodes in the game knowledge map, determining an object content map according to the edges between the object nodes and the content nodes in the game knowledge map, and determining the game map according to the edges between the game nodes and the game nodes in the game knowledge map.

And step 5) determining each game vector according to a preset graph representation algorithm, an object game graph, an object content graph and a game graph.

The object game graph is used for representing the association of any two game nodes in the object dimension, the object content graph is used for representing the association of any two game nodes in the content dimension, and the game graph is used for representing the association of any two game nodes in the game dimension.

For example, an object game map for representing the association of any two game nodes in the object dimension, an object content map for representing the association of any two game nodes in the content dimension, and a game map for representing the association of any two game nodes in the game dimension may be extracted from the game knowledge map. It can be understood that the object game map, the object content map, and the game map each include a game node corresponding to each game in the plurality of games, and if there is an association between any two games in the object dimension, an edge may be established between two game nodes corresponding to the two games in the object game map. Similarly, if the two games are related in the content (or game) dimension, an edge may be established between two game nodes corresponding to the two games in the object content graph (or game graph).

Specifically, game a and game B have an association in the object dimension, which may be understood as that the degree of overlap of the objects associated with game a and game B is greater than a preset fifth threshold (e.g., 30%), that is, the number of object nodes in the game knowledge graph, where an edge exists between the game nodes corresponding to game a and an edge exists between the game nodes corresponding to game B, exceeds 30% of the number of objects. Then an edge may be established between the game node corresponding to game a and the game node corresponding to game B in the object game graph, and further, the coincidence degree of the object may be taken as the width of the edge, or the value of the edge.

Game a and game B are associated in a content dimension, and it is understood that the degree of overlap of objects associated with content a and content B is greater than a preset sixth threshold (for example, 20%), where game a is involved in content a and game B is involved in content B, that is, the number of object nodes in the game knowledge graph where there are edges between content nodes corresponding to content a and edges between content nodes corresponding to content B are more than 20% of the number of objects. Then an edge may be established between the content node corresponding to the content a and the content node corresponding to the content b in the object content graph, and further, the overlap ratio of the object may be used as the width of the edge, or the value of the edge.

Game A and game B are associated in a game dimension, which can be understood as that the similarity of game information of game A and game B exceeds a seventh threshold (for example: 50%), namely, the edge between the game node corresponding to game A and the game node corresponding to game B in the game knowledge graph represents that the similarity of game information of game A and game B exceeds 50%. Then an edge may be established between the game node corresponding to game a and the game node corresponding to game B in the game graph, and further, the similarity of the object may be used as the width of the edge or the value of the edge.

After the object game map, the object content map, and the game map are obtained, each game vector may be determined according to a preset map representation algorithm. The graph representation algorithm can represent the graph in an adjacent list or an adjacent linked list mode, and therefore a vector for representing each node in the graph is obtained. Specifically, the object game map input map may be represented by an algorithm to obtain a first game vector corresponding to each game, the object content map input map may be represented by an algorithm to obtain a second game vector corresponding to each game, and the game map input map may be represented by an algorithm to obtain a third game vector corresponding to each game. Finally, the first game vector, the second game vector, and the third game vector corresponding to each game may be averaged as the game vector corresponding to the game. The first game vector, the second game vector and the third game vector corresponding to each game can be spliced to be used as the game vector corresponding to the game. The maximum value of the first game vector, the second game vector and the third game vector corresponding to each game can be used as the game vector corresponding to the game. The manner in which the game vector is specifically determined is not particularly limited by this disclosure. In this way, the game vector corresponding to each game can characterize the game from multiple dimensions, and contains a larger amount of information than a mode of describing the game only by game information.

Based on the game knowledge map and each game vector, an object vector is determined that characterizes each object, step 1012.

For example, after each game vector is obtained, an object vector characterizing each object may be further determined in conjunction with the game knowledge graph. Specifically, the method can be realized by the following steps:

in one application scenario, step 1012 may be implemented by:

and 6) determining the object game map according to the edges between the object nodes and the game nodes in the game knowledge map, and determining the object content map according to the edges between the object nodes and the content nodes in the game knowledge map.

And 7) determining each object vector according to each game vector, the object game map and the object content map.

The object game map is used for representing the association of each object node and each game node in the game dimension, and the object content map is used for representing the association of each object node and each game node in the content dimension.

For example, an object game map for characterizing the association of each object node with each game node in the game dimension and an object content map for characterizing the association of each object node with each game node in the content dimension may be extracted from the game knowledge map. It can be understood that the object game graph and the object content graph each include an object node corresponding to each object and a game node corresponding to each game, and if there is an association between an object and a game in a game dimension, an edge may be established between the object node corresponding to the object and the game node corresponding to the game in the object game graph. Similarly, if the game and the game have an association in the content dimension, an edge may be established between the object node corresponding to the object and the game node corresponding to the game in the object content graph. The target game map in step 1011 shows a map different from the target game map in step 1012, and similarly, the target content map in step 1011 shows a map different from the target content map in step 1012.

Specifically, the object nail and the game a are associated in a game dimension, which can be understood as that the object nail downloads, uses or consumes the game a, that is, an edge exists between an object node corresponding to the object nail and a game node corresponding to the game a in the game knowledge graph. Then an edge may be established between the object node corresponding to the object a and the game node corresponding to the game a in the object game graph, and further, any kind of measure (for example, duration of use, frequency of use, etc.) of downloading, using, or consuming the game a by the object a may be used as the width of the edge, or the value of the edge.

The object A and the game A are associated in a content dimension, and the object A can be understood as browsing, commenting, agreeing or sharing content a, wherein the content a relates to the game A, namely, an edge exists between an object node corresponding to the object A and a content node corresponding to the content a in the game knowledge graph. Then, an edge may be established between the object node corresponding to the object a and the game node corresponding to the game a in the object content graph, and further, any kind of metric (for example, the number of forwarding times, the number of approval, and the like) of browsing, commenting, approving, or sharing the content a of the object a may be used as the width of the edge, or the value of the edge.

After the object game map and the object content map are obtained, each object vector may be determined based on the object game map, the object content map, and each game vector previously determined. Specifically, a first object vector corresponding to each object may be determined according to the object game map, and a second object vector corresponding to each object may be determined according to the object content map. Taking the object game map shown in fig. 9 as an example, if there are edges between the node a corresponding to the object a and the node B corresponding to the game a and between the node B corresponding to the object B and only between the node B and the node B corresponding to the object B in the object game map, the game vector corresponding to the game a and the game vector corresponding to the game B may be summed to be the first object vector corresponding to the object a, and the game vector corresponding to the game B may be the first object vector corresponding to the object B. Further, if the value of the edge existing between the node a and the node a is 0.8, and the value of the edge existing between the node a and the node B is 0.2, then the game vector corresponding to the game a and the game vector corresponding to the game B may be weighted and summed according to the weights of 0.8 and 0.2 to be used as the first object vector corresponding to the object a.

The manner of determining the second object vector based on the object content map and each game vector is the same as the manner of determining the first object vector based on the object game map and each game vector, and will not be described herein again. Finally, the first object vector and the second object vector corresponding to each object may be averaged to serve as the object vector corresponding to the object. Or the first object vector and the second object vector corresponding to each object may be spliced to serve as the object vector corresponding to the object. The maximum value of the first object vector and the second object vector corresponding to each object can also be used as the object vector corresponding to the object. The present disclosure does not specifically limit the manner in which the object vector is specifically determined. In this way, the object vector corresponding to each object can represent the object from multiple dimensions by using the game vector, reflects the association of each object and each game in different dimensions, and contains more information compared with the mode of describing the object only by using the object information.

Step 1013, an object vector for representing the object to be identified is taken as the object vector to be identified, and a game vector for representing the target game is taken as the target game vector.

For example, after obtaining each game vector and each object vector according to steps 1011 to 1012, an object vector for characterizing the object to be recognized may be used as the object vector to be recognized, and a game vector for characterizing the target game may be used as the target game vector. Further, an object vector for characterizing the seed object may be used as the seed object vector.

Fig. 10 is a block diagram illustrating a knowledge-graph based object recognition apparatus according to an exemplary embodiment, and as shown in fig. 10, the apparatus 200 includes:

the vector determination module 201 is configured to determine, according to a game knowledge graph established in advance, a target game vector for characterizing a target game and an object vector to be recognized for characterizing an object to be recognized.

The relevancy determining module 202 is configured to determine the relevancy of the object to be recognized and the target game according to the object vector to be recognized, the target game vector and a pre-trained recognition model, where the recognition model is obtained by training a seed object vector of a seed object used for representing the target game and the target game vector, and the seed object vector is determined according to the game knowledge graph.

The identifying module 203 is configured to determine that the object to be identified is the target object of the target game if the correlation between the object to be identified and the target game meets a preset condition.

In an application scenario, the relevancy determination module 202 may be configured to:

In another application scenario, the game knowledge-graph comprises a plurality of nodes and at least one edge, the plurality of nodes comprising: object nodes, game nodes, and content nodes. Each edge is used to characterize the association between two nodes at both ends of the edge.

Fig. 11 is a block diagram illustrating another knowledge-graph-based object recognition apparatus according to an exemplary embodiment, where object nodes include object nodes corresponding to a plurality of sample objects, the sample objects include positive sample objects and negative sample objects, and the positive sample objects include seed objects, as shown in fig. 11. The apparatus 200 may further comprise:

and the training module 204 is configured to train the recognition model for the target game.

The training module 204 may be configured to perform the following steps:

step a, obtaining a sample input set, wherein the sample input set comprises: a sample input corresponding to each sample object, the sample input comprising: an object vector and a target game vector for characterizing the sample object, as determined from the game knowledge graph.

And b, acquiring a sample output set, wherein the sample output set comprises a sample output corresponding to each sample input, and each sample output comprises a real identification result of a corresponding sample object.

And c, taking the sample input set as the input of the recognition model, and taking the sample output set as the output of the recognition model so as to train the recognition model.

In another application scenario, the positive sample objects may further include an extension object, which is determined in the following manner.

And 3) taking the active object of the related game as an extension object.

In yet another application scenario, step c may include:

and c1, clustering the sample input set according to a preset clustering algorithm to obtain a plurality of groups of sample input subsets, wherein each group of sample input subsets corresponds to one group of sample output subsets.

And c2, inputting the set of sample input subsets into the recognition model to form the corresponding recognizer models of the set of sample input subsets, and using the sample output subsets corresponding to the set of sample input subsets as the output of the corresponding recognizer models of the set of sample input subsets to train the corresponding recognizer models of the set of sample input subsets.

In another application scenario, the identifier model corresponding to each group of sample input subsets is a tree integration model and includes a plurality of tree models. The implementation manner of step c2 may include:

In one implementation, the game knowledge graph is created by:

and d, acquiring object information of a plurality of objects, game information of a plurality of games and content information of a plurality of contents. The content is any one of text, audio and video, the plurality of objects comprise objects to be identified and seed objects, and the plurality of games comprise target games.

And e, establishing an object node corresponding to each object, a game node corresponding to each game and a content node corresponding to each content.

And f, establishing edges among the nodes according to a preset association rule.

In another implementation, the implementation of step f may include:

step f1, according to the object information of each object, determining the game related object meeting the association rule with the object, wherein the game related object comprises: at least one of an object, a game and a content, and an edge is established between an object node corresponding to the object and a node corresponding to the game-related object.

Step f2, according to the content information of each content, determining a content associated object satisfying the association rule with the content, the content associated object comprising: games and/or contents, and establishing edges between content nodes corresponding to the contents and nodes corresponding to the content-associated objects.

Step f3, according to the game information of each game, determining the associated game meeting the association rule with the game, and establishing an edge between the game node corresponding to the game and the game node corresponding to the associated game.

Fig. 12 is a block diagram illustrating another knowledge-graph based object recognition apparatus according to an exemplary embodiment, and as shown in fig. 12, the vector determination module 201 may include:

a first determining sub-module 2011 is configured to determine game vectors for characterizing each game according to the game knowledge map.

A second determining submodule 2012 for determining an object vector characterizing each object based on the game knowledge base and each game vector.

And the third determining submodule 2013 is used for taking the object vector for representing the object to be identified as the object vector to be identified and taking the game vector for representing the target game as the target game vector.

In one application scenario, the first determining sub-module 2011 may be configured to perform the following steps:

In an application scenario, the second determining submodule 2012 may be configured to perform the following steps:

With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.

Referring now to fig. 13, a schematic structural diagram of an electronic device (e.g., an execution subject, which may be a terminal device or a server in the above-mentioned embodiments) 300 suitable for implementing an embodiment of the present disclosure is shown. The terminal device in the embodiments of the present disclosure may include, but is not limited to, a mobile terminal such as a mobile phone, a notebook computer, a digital broadcast receiver, a PDA (personal digital assistant), a PAD (tablet computer), a PMP (portable multimedia player), a vehicle terminal (e.g., a car navigation terminal), and the like, and a stationary terminal such as a digital TV, a desktop computer, and the like. The electronic device shown in fig. 13 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present disclosure.

As shown in fig. 13, the electronic device 300 may include a processing means (e.g., a central processing unit, a graphics processor, etc.) 301 that may perform various appropriate actions and processes in accordance with a program stored in a Read Only Memory (ROM)302 or a program loaded from a storage means 308 into a Random Access Memory (RAM) 303. In the RAM 303, various programs and data necessary for the operation of the electronic apparatus 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input/output (I/O) interface 305 is also connected to bus 304.

Generally, the following devices may be connected to the I/O interface 305: input devices 306 including, for example, a touch screen, touch pad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; an output device 307 including, for example, a Liquid Crystal Display (LCD), a speaker, a vibrator, and the like; storage devices 308 including, for example, magnetic tape, hard disk, etc.; and a communication device 309. The communication means 309 may allow the electronic device 300 to communicate wirelessly or by wire with other devices to exchange data. While fig. 13 illustrates an electronic device 300 having various means, it is to be understood that not all illustrated means are required to be implemented or provided. More or fewer devices may alternatively be implemented or provided.

In particular, according to an embodiment of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a non-transitory computer readable medium, the computer program containing program code for performing the method illustrated by the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network through the communication means 309, or installed from the storage means 308, or installed from the ROM 302. The computer program, when executed by the processing device 301, performs the above-described functions defined in the methods of the embodiments of the present disclosure.

It should be noted that the computer readable medium in the present disclosure can be a computer readable signal medium or a computer readable storage medium or any combination of the two. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples of the computer readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present disclosure, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In contrast, in the present disclosure, a computer readable signal medium may comprise a propagated data signal with computer readable program code embodied therein, either in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: electrical wires, optical cables, RF (radio frequency), etc., or any suitable combination of the foregoing.

In some embodiments, the terminal devices, servers may communicate using any currently known or future developed network Protocol, such as HTTP (HyperText Transfer Protocol), and may interconnect with any form or medium of digital data communication (e.g., a communications network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), the Internet (e.g., the Internet), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future developed network.

The computer readable medium may be embodied in the electronic device; or may exist separately without being assembled into the electronic device.

The computer readable medium carries one or more programs which, when executed by the electronic device, cause the electronic device to: determining a target game vector for representing a target game and an object vector to be recognized for representing an object to be recognized according to a game knowledge graph established in advance; determining the correlation degree of the object to be recognized and the target game according to the object vector to be recognized, the target game vector and a pre-trained recognition model, wherein the recognition model is obtained by training a seed object vector used for representing the target game seed object and the target game vector, and the seed object vector is determined according to the game knowledge graph; and if the correlation degree of the object to be recognized and the target game meets a preset condition, determining that the object to be recognized is the target object of the target game.

Computer program code for carrying out operations for the present disclosure may be written in any combination of one or more programming languages, including but not limited to an object oriented programming language such as Java, Smalltalk, C + +, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the subject computer, partly on the subject computer, as a stand-alone software package, partly on the subject computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the subject computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The modules described in the embodiments of the present disclosure may be implemented by software or hardware. Where the name of a module does not in some cases constitute a limitation on the module itself, for example, the vector determination module may also be described as a "module that determines an object vector to be recognized and a target game vector".

The functions described herein above may be performed, at least in part, by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), Application Specific Standard Products (ASSPs), systems on a chip (SOCs), Complex Programmable Logic Devices (CPLDs), and the like.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

Example 1 provides a method of knowledge-graph-based object recognition, in accordance with one or more embodiments of the present disclosure, comprising: determining a target game vector for representing a target game and an object vector to be recognized for representing an object to be recognized according to a game knowledge graph established in advance; determining the correlation degree of the object to be recognized and the target game according to the object vector to be recognized, the target game vector and a pre-trained recognition model, wherein the recognition model is obtained by training a seed object vector used for representing the target game seed object and the target game vector, and the seed object vector is determined according to the game knowledge graph; and if the correlation degree of the object to be recognized and the target game meets a preset condition, determining that the object to be recognized is the target object of the target game.

Example 2 provides the method of example 1, the determining a degree of correlation of the object to be recognized with the target game according to the object vector to be recognized, the target game vector, and a pre-trained recognition model, including: and inputting the object vector to be recognized, the target game vector and the object characteristics of the object to be recognized into the recognition model to obtain the correlation degree between the object to be recognized and the target game output by the recognition model, wherein the object characteristics are determined according to the object information of the object to be recognized.

Example 3 provides the method of example 1 or example 2, the game knowledge-graph comprising a plurality of nodes and at least one edge, the plurality of nodes comprising: object nodes, game nodes and content nodes; each edge is used to characterize the association between two nodes at both ends of the edge.

Example 4 provides the method of example 3, the object node comprising object nodes corresponding to a plurality of sample objects, the sample objects comprising positive and negative sample objects, the positive sample object comprising the seed object; the method further comprises the following steps: training the recognition model for the target game; the training the recognition model for the target game includes: obtaining a sample input set, the sample input set comprising: a sample input corresponding to each of the sample objects, the sample input comprising: an object vector and the target game vector determined from the game knowledge graph and used for representing the sample object; obtaining a sample output set comprising a sample output corresponding to each of the sample inputs, each of the sample outputs comprising a true identification result of the corresponding sample object; and taking the sample input set as the input of the recognition model, and taking the sample output set as the output of the recognition model so as to train the recognition model.

Example 5 provides the method of example 4, the positive sample object further including an extension object, the extension object determined by: determining the relevance of other games and the target game according to other game vectors used for representing other games and the target game vector, wherein the other games are games other than the target game, and the other game vectors are determined according to the game knowledge graph; taking the other games with the correlation degree greater than or equal to a preset correlation degree threshold value with the target game as the related games corresponding to the target game; and taking the active object of the related game as the extension object.

Example 6 provides the method of example 4, the taking the set of sample inputs as inputs to the recognition model and the set of sample outputs as outputs to the recognition model to train the recognition model, comprising: clustering the sample input set according to a preset clustering algorithm to obtain a plurality of groups of sample input subsets, wherein each group of sample input subsets corresponds to a group of sample output subsets; for each group of the sample input subsets, inputting the group of sample input subsets into the identifier models corresponding to the group of sample input subsets, and using the sample output subsets corresponding to the group of sample input subsets as the output of the identifier models corresponding to the group of sample input subsets to train the identifier models corresponding to the group of sample input subsets; and determining the output of the identification model according to the output of the identifier model corresponding to each group of the sample input subset.

Example 7 provides the method of example 6, wherein the identifier model corresponding to each set of the sample input subsets is a tree integration model comprising a plurality of tree models; the inputting the set of sample input subsets into the recognition model, wherein the recognition submodels corresponding to the set of sample input subsets are included, and the sample output subsets corresponding to the set of sample input subsets are used as the outputs of the recognition submodels corresponding to the set of sample input subsets to train the recognition submodels corresponding to the set of sample input subsets, includes: randomly sampling the set of sample input subsets for each of the tree models to obtain a sample input subset, the sample input subset comprising a smaller number of the sample inputs than the set of sample input subsets; randomly sampling each sample input included in the sampling input subset to obtain a sampling sample input corresponding to the sample input, wherein the sampling sample input belongs to the sample input; taking a sampling sample input corresponding to each sample input included in the sampling input subset as an input of the tree model, and taking a sample output corresponding to each sample input included in the sampling input subset as an output of the tree model, so as to train the tree model; the output of the tree integration model is determined from the output of each of the tree models.

Example 8 provides the method of example 3, the game knowledge-graph being established by: acquiring object information of a plurality of objects, game information of a plurality of games and content information of a plurality of contents; establishing an object node corresponding to each object, a game node corresponding to each game and a content node corresponding to each content; and establishing edges among the nodes according to a preset association rule.

Example 9 provides the method of example 8, according to one or more embodiments of the present disclosure, the establishing an edge between the plurality of nodes according to a preset association rule, including: according to the object information of each object, determining a game associated object meeting the association rule with the object, wherein the game associated object comprises: at least one of an object, a game and a content, and establishing an edge between an object node corresponding to the object and a node corresponding to the game-related object; according to the content information of each content, determining a content associated object meeting the association rule with the content, wherein the content associated object comprises: games and/or contents, and establishing edges between content nodes corresponding to the contents and nodes corresponding to the content-associated objects; and determining the associated games meeting the association rules with the games according to the game information of each game, and establishing edges between the game nodes corresponding to the games and the game nodes corresponding to the associated games.

Example 10 provides the method of example 9, the determining, from a pre-established game knowledge-graph, a target game vector for characterizing a target game and an object-to-be-identified vector for characterizing an object-to-be-identified, comprising: determining game vectors for characterizing each of the games based on the game knowledge graph; determining an object vector for characterizing each of the objects based on the game knowledge graph and each of the game vectors; and taking an object vector for representing the object to be identified as the object vector to be identified, and taking a game vector for representing a target game as the target game vector.

Example 11 provides the method of example 10, the determining, from the game knowledge graph, game vectors for characterizing each of the games, comprising: determining an object game map according to edges between object nodes and game nodes in the game knowledge map, determining an object content map according to edges between object nodes and content nodes in the game knowledge map, and determining a game map according to edges between game nodes and game nodes in the game knowledge map; determining each game vector according to a preset graph representation algorithm, the object game map, the object content map and the game map; the object game graph is used for representing the association of any two game nodes in an object dimension, the object content graph is used for representing the association of any two game nodes in a content dimension, and the game graph is used for representing the association of any two game nodes in a game dimension.

Example 12 provides the method of example 10, the determining, from the game knowledge-graph and each of the game vectors, an object vector for characterizing each of the objects, comprising: determining an object game map according to edges between object nodes and game nodes in the game knowledge map, and determining an object content map according to edges between object nodes and content nodes in the game knowledge map; determining each of the object vectors from each of the game vectors, the object game map and the object content map; wherein the object game graph is used for representing the association of each object node and each game node in a game dimension, and the object content graph is used for representing the association of each object node and each game node in a content dimension.

Example 13 provides, in accordance with one or more embodiments of the present disclosure, a knowledge-graph based object recognition apparatus, comprising: the vector determination module is used for determining a target game vector for representing a target game and an object vector to be identified for representing an object to be identified according to a game knowledge graph established in advance; the relevancy determining module is used for determining the relevancy of the object to be recognized and the target game according to the object vector to be recognized, the target game vector and a pre-trained recognition model, wherein the recognition model is obtained by training a seed object vector of a seed object used for representing the target game and the target game vector, and the seed object vector is determined according to the game knowledge graph; and the identification module is used for determining that the object to be identified is the target object of the target game if the correlation degree of the object to be identified and the target game meets a preset condition.

Example 14 provides a computer readable medium having stored thereon a computer program that, when executed by a processing device, implements the steps of the methods of examples 1-12, in accordance with one or more embodiments of the present disclosure.

Example 15 provides, in accordance with one or more embodiments of the present disclosure, an electronic device, comprising: a storage device having a computer program stored thereon; processing means for executing the computer program in the storage means to implement the steps of the methods of examples 1 to 12.

The foregoing description is only exemplary of the preferred embodiments of the disclosure and is illustrative of the principles of the technology employed. It will be appreciated by those skilled in the art that the scope of the disclosure herein is not limited to the particular combination of features described above, but also encompasses other embodiments in which any combination of the features described above or their equivalents does not depart from the spirit of the disclosure. For example, the above features and (but not limited to) the features disclosed in this disclosure having similar functions are replaced with each other to form the technical solution.

Further, while operations are depicted in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Likewise, while several specific implementation details are included in the above discussion, these should not be construed as limitations on the scope of the disclosure. Certain features that are described in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination.

Although the subject matter has been described in language specific to structural features and/or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims. With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.

Claims

1. A method for knowledge-graph based object recognition, the method comprising:

2. The method of claim 1, wherein the determining the degree of correlation between the object to be recognized and the target game according to the object vector to be recognized, the target game vector and a pre-trained recognition model comprises:

and inputting the object vector to be recognized, the target game vector and the object characteristics of the object to be recognized into the recognition model to obtain the correlation degree between the object to be recognized and the target game output by the recognition model, wherein the object characteristics are determined according to the object information of the object to be recognized.

3. The method of claim 1 or 2, wherein the game knowledge-graph comprises a plurality of nodes and at least one edge, the plurality of nodes comprising: object nodes, game nodes and content nodes; each edge is used to characterize the association between two nodes at both ends of the edge.

4. The method of claim 3, wherein the object nodes comprise object nodes corresponding to a plurality of sample objects, wherein the sample objects comprise positive sample objects and negative sample objects, and wherein the positive sample objects comprise the seed objects;

the method further comprises the following steps: training the recognition model for the target game;

the training the recognition model for the target game includes:

obtaining a sample input set, the sample input set comprising: a sample input corresponding to each of the sample objects, the sample input comprising: an object vector and the target game vector determined from the game knowledge graph and used for representing the sample object;

obtaining a sample output set comprising a sample output corresponding to each of the sample inputs, each of the sample outputs comprising a true identification result of the corresponding sample object;

and taking the sample input set as the input of the recognition model, and taking the sample output set as the output of the recognition model so as to train the recognition model.

5. The method of claim 4, wherein the positive sample object further comprises an extension object, the extension object being determined by:

determining the relevance of other games and the target game according to other game vectors used for representing other games and the target game vector, wherein the other games are games other than the target game, and the other game vectors are determined according to the game knowledge graph;

taking the other games with the correlation degree greater than or equal to a preset correlation degree threshold value with the target game as the related games corresponding to the target game;

and taking the active object of the related game as the extension object.

6. The method of claim 4, wherein the training the recognition model by using the sample input set as an input of the recognition model and the sample output set as an output of the recognition model comprises:

clustering the sample input set according to a preset clustering algorithm to obtain a plurality of groups of sample input subsets, wherein each group of sample input subsets corresponds to a group of sample output subsets;

for each group of the sample input subsets, inputting the group of sample input subsets into the identifier models corresponding to the group of sample input subsets, and using the sample output subsets corresponding to the group of sample input subsets as the output of the identifier models corresponding to the group of sample input subsets to train the identifier models corresponding to the group of sample input subsets;

and determining the output of the identification model according to the output of the identifier model corresponding to each group of the sample input subset.

7. The method of claim 6, wherein the identifier model corresponding to each set of the sample input subsets is a tree integration model comprising a plurality of tree models; the inputting the set of sample input subsets into the recognition model, wherein the recognition submodels corresponding to the set of sample input subsets are included, and the sample output subsets corresponding to the set of sample input subsets are used as the outputs of the recognition submodels corresponding to the set of sample input subsets to train the recognition submodels corresponding to the set of sample input subsets, includes:

randomly sampling the set of sample input subsets for each of the tree models to obtain a sample input subset, the sample input subset comprising a smaller number of the sample inputs than the set of sample input subsets;

randomly sampling each sample input included in the sampling input subset to obtain a sampling sample input corresponding to the sample input, wherein the sampling sample input belongs to the sample input;

taking a sampling sample input corresponding to each sample input included in the sampling input subset as an input of the tree model, and taking a sample output corresponding to each sample input included in the sampling input subset as an output of the tree model, so as to train the tree model;

the output of the tree integration model is determined from the output of each of the tree models.

8. The method of claim 3, wherein the game knowledge graph is established by:

acquiring object information of a plurality of objects, game information of a plurality of games and content information of a plurality of contents;

establishing an object node corresponding to each object, a game node corresponding to each game and a content node corresponding to each content;

and establishing edges among the nodes according to a preset association rule.

9. The method according to claim 8, wherein the establishing edges between the plurality of nodes according to a preset association rule comprises:

according to the object information of each object, determining a game associated object meeting the association rule with the object, wherein the game associated object comprises: at least one of an object, a game and a content, and establishing an edge between an object node corresponding to the object and a node corresponding to the game-related object;

according to the content information of each content, determining a content associated object meeting the association rule with the content, wherein the content associated object comprises: games and/or contents, and establishing edges between content nodes corresponding to the contents and nodes corresponding to the content-associated objects;

and determining the associated games meeting the association rules with the games according to the game information of each game, and establishing edges between the game nodes corresponding to the games and the game nodes corresponding to the associated games.

10. The method of claim 9, wherein determining a target game vector for characterizing a target game and an object-to-be-identified vector for characterizing an object to be identified based on a pre-established game knowledge graph comprises:

determining game vectors for characterizing each of the games based on the game knowledge graph;

determining an object vector for characterizing each of the objects based on the game knowledge graph and each of the game vectors;

and taking an object vector for representing the object to be identified as the object vector to be identified, and taking a game vector for representing a target game as the target game vector.

11. The method of claim 10, wherein determining game vectors for characterizing each of the games based on the game knowledge graph comprises:

determining an object game map according to edges between object nodes and game nodes in the game knowledge map, determining an object content map according to edges between object nodes and content nodes in the game knowledge map, and determining a game map according to edges between game nodes and game nodes in the game knowledge map;

determining each game vector according to a preset graph representation algorithm, the object game map, the object content map and the game map;

the object game graph is used for representing the association of any two game nodes in an object dimension, the object content graph is used for representing the association of any two game nodes in a content dimension, and the game graph is used for representing the association of any two game nodes in a game dimension.

12. The method of claim 10, wherein determining an object vector for characterizing each of the objects based on the game knowledge graph and each of the game vectors comprises:

determining an object game map according to edges between object nodes and game nodes in the game knowledge map, and determining an object content map according to edges between object nodes and content nodes in the game knowledge map;

determining each of the object vectors from each of the game vectors, the object game map and the object content map;

wherein the object game graph is used for representing the association of each object node and each game node in a game dimension, and the object content graph is used for representing the association of each object node and each game node in a content dimension.

13. An apparatus for knowledge-graph based object recognition, the apparatus comprising:

14. A computer-readable medium, on which a computer program is stored, characterized in that the program, when being executed by processing means, carries out the steps of the method of any one of claims 1-12.

15. An electronic device, comprising:

a storage device having a computer program stored thereon;

processing means for executing the computer program in the storage means to carry out the steps of the method according to any one of claims 1 to 12.