CN111159382B - Method and device for constructing and using knowledge model of conversational system - Google Patents

Abstract

The invention discloses a method and device for constructing and using a knowledge model of a conversation system, wherein a method for constructing a knowledge model of a conversation system includes: abstracting at least one subject associated with the knowledge point from the knowledge point in the conversation system; The at least one topic constructs a system topic tree; based on the system topic tree, an association relationship between knowledge points is determined; a conversational system knowledge model is constructed at least based on the system topic tree and the association relationship between each knowledge point. The method of this embodiment abstracts the knowledge points into at least one topic, then uses the topic to construct a topic tree, and then determines the association relationship between the knowledge points according to the topic tree, and then a conversational system knowledge model can be constructed for subsequent heuristics. During the session, questions or answers are recommended to the user, so that questions or content that are more in line with the user's interests can be recommended to the user.

Description

Method and device for constructing and using knowledge model of conversational system

technical field

The invention belongs to the technical field of conversational systems, and in particular relates to a method and device for constructing and using a knowledge model of conversational systems.

Background technique

In related technologies, in general conversational system application scenarios, such as smart speakers, smart TVs, including current in-vehicle equipment, etc., when the user says a question, the intelligent customer service automatically asks and answers questions and dialogues, the basic process of such a system. It receives a question from the user or a sentence said by the user, and then does some processing in the system to give the user a reply.

Heuristic conversations keep conversations going by capturing some of the connections behind the questions.

In the process of realizing the present application, the inventor found that the existing products or technologies are mainly heuristic conversations based on themes or scenarios, without considering the interests of users themselves.

SUMMARY OF THE INVENTION

Embodiments of the present invention provide a method and device for constructing and using a knowledge model of a conversational system, which are used to solve at least one of the above technical problems.

In a first aspect, an embodiment of the present invention provides a method for constructing a knowledge model of a conversational system, including: abstracting at least one topic associated with the knowledge point from a knowledge point in a conversational system; constructing a system topic based on the at least one topic tree; determining an association relationship between knowledge points based on the system topic tree; and constructing a conversational system knowledge model based at least on the system theme tree and the association relationship between the knowledge points.

In a second aspect, an embodiment of the present invention provides a method for using a knowledge model of a conversation system, including: constructing a user interest model based on basic information of a user and historical conversation data of the user, wherein the user interest model includes at least one knowledge points; perform semantic map-based matching on the knowledge points in the user interest model and the knowledge points in the conversation system knowledge model constructed according to the method of the first aspect; generate at least one semantic map according to the matching result of the semantic map matching knowledge points, wherein the knowledge points are associated with at least one question; and recommending the questions associated with the at least one matching knowledge point to the user.

In a third aspect, an embodiment of the present invention provides an apparatus for constructing a knowledge model of a conversational system, including: a knowledge point abstraction module configured to abstract at least one topic associated with the knowledge point from the knowledge point in the conversational system; a topic tree a building module, configured to build a system theme tree based on the at least one theme; an association relationship determination module, configured to determine the association relationship between knowledge points based on the system theme tree; and a knowledge model building module, configured to at least based on all A conversational system knowledge model is constructed based on the relationship between the system topic tree and the knowledge points.

In a fourth aspect, an embodiment of the present invention provides a device for using a knowledge model of a conversation system, including: a user interest model building module configured to build a user interest model based on basic information of a user and historical conversation data of the user, wherein the The user interest model includes at least one knowledge point; the semantic graph matching module is configured to perform the matching between the knowledge point in the user interest model and the knowledge point in the conversation system knowledge model constructed according to the method described in the first aspect. Semantic graph-based matching; a matching knowledge point generation module, configured to generate at least one matched knowledge point according to the matching result of the semantic graph, wherein the knowledge point is associated with at least one question; At least one matching knowledge point associated question is recommended to the user.

A fifth aspect provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, The instructions are executed by the at least one processor, so that the at least one processor can execute the steps of the method for constructing a knowledge model of a conversational system according to any embodiment of the present invention.

In a sixth aspect, an embodiment of the present invention further provides a computer program product, the computer program product includes a computer program stored on a non-volatile computer-readable storage medium, the computer program includes program instructions, and when the program is When the instructions are executed by a computer, the computer is made to execute the steps of the method for constructing a knowledge model of a conversational system according to any embodiment of the present invention.

The method of this embodiment abstracts the knowledge points into at least one topic, then uses the topic to construct a topic tree, and then determines the association relationship between the knowledge points according to the topic tree, and then a conversational system knowledge model can be constructed for subsequent heuristics. During the session, questions or answers are recommended to the user, so that questions or content that are more in line with the user's interests can be recommended to the user.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without any creative effort.

1 is a flowchart of a method for constructing a knowledge model of a conversational system according to an embodiment of the present invention;

2 is a flowchart of another method for constructing a knowledge model of a conversational system provided by an embodiment of the present invention;

3 is a flowchart of a method for using a conversational system knowledge model provided by an embodiment of the present invention;

4 is a flowchart of another method for using a knowledge model of a conversation system provided by an embodiment of the present invention;

5 is a flowchart of another method for using a knowledge model of a conversational system provided by an embodiment of the present invention;

FIG. 6 is a specific flowchart of a system provided by an embodiment of the present invention;

7 is a specific example diagram of a model structure of user interest points based on topics and knowledge points provided by an embodiment of the present invention;

FIG. 8 is a semantic representation of a user question and a semantic representation of a dialogue system knowledge system provided by an embodiment of the present invention;

9 is a block diagram of an apparatus for constructing a knowledge model of a conversation system according to an embodiment of the present invention;

10 is a block diagram of a device for using a knowledge model of a conversation system provided by an embodiment of the present invention;

FIG. 11 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed ways

In order to make the purposes, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments These are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Please refer to FIG. 1 , which shows a flowchart of an embodiment of a method for constructing a knowledge model of a conversational system of the present application. The method for constructing a knowledge model of a conversational system in this embodiment can be applied to a terminal with an intelligent voice dialogue function, such as an intelligent Children's story machines, smart conversation toys, devices that include smart story playback, etc.

As shown in Figure 1, in step 101, at least one topic associated with the knowledge point is abstracted from the knowledge point in the conversation system;

In step 102, a system theme tree is constructed based on at least one theme;

In step 103, the association relationship between the knowledge points is determined based on the system theme tree;

In step 104, a conversational system knowledge model is constructed at least based on the system topic tree and the association relationship between each knowledge point.

In this embodiment, for step 101, the device for constructing the knowledge model of the conversational system abstracts at least one topic associated with the knowledge point from the knowledge point in the conversational system, wherein the knowledge point can be, for example, Li Bai, which can be abstracted as a poet , or can be abstracted into a game character to construct a theme, such as a poet or a game character.

After that, for step 102, a systematic topic tree is constructed based on at least one topic, for example, major poets such as Du Fu are derived from the poet Li Bai to construct the topic tree. Or the theme tree constructed by major game characters such as Di Renjie is derived from the game character Li Bai.

Then, for step 103, the relationship between the knowledge points is determined based on the system theme tree, for example, the relationship between Li Bai and Du Fu and other major poets is determined based on the theme tree of poets, or Li Bai and Di Renjie are determined based on the theme tree of game characters. The relationship between the major game characters.

Finally, for step 104, a conversational system knowledge model is constructed based on at least the relationship between the system topic tree and each knowledge point. For example, a conversational system knowledge model is constructed based on the relationship between the poet topic tree and each knowledge point, wherein the knowledge Models are abstractions that formalize and structure knowledge.

The method of this embodiment abstracts the knowledge points into at least one topic, then uses the topic to construct a topic tree, and then determines the association relationship between the knowledge points according to the topic tree, and then a conversational system knowledge model can be constructed for subsequent heuristics. During the session, questions or answers are recommended to the user, so that questions or content that are more in line with the user's interests can be recommended to the user.

Referring further to FIG. 2 , it shows a flowchart of another method for constructing a knowledge model of a conversation system provided by an embodiment of the present application. This flowchart is mainly a flowchart for the steps further defined in step 103 in flowchart 1 . The knowledge point includes a question and answer pair and a knowledge graph, the question and answer pair includes a first entity, and the knowledge graph includes a second entity and an association relationship between the second entities.

As shown in Figure 2, in step 201, when the first entity in the theme tree is the same as the second entity, the first entity and the second entity are associated and fused;

In step 202, the association relationship between the knowledge points is determined based on the subject tree after association fusion.

In this embodiment, for step 201, when the first entity in the topic tree is the same as the second entity, the device for constructing the conversational system knowledge model associates and fuses the first entity with the second entity, for example, combining the first entity in the topic tree with the second entity. The poet Bai Juyi is the first entity, and the film and television character Bai Juyi is the second entity, and then they are associated and merged to form a merged theme tree;

Finally, for step 202, the association relationship between knowledge points is determined based on the theme tree after association fusion, for example, the association relationship between knowledge points is determined based on the theme tree after the fusion of poets and film and television characters to build a conversational system knowledge model.

The solution of this embodiment determines the association relationship between knowledge points by performing relationship fusion on entities with the same theme tree species, so that the same entity can be fused to make the relationship between the knowledge points more closely, so that there is no relationship between two knowledge points. contact with knowledge points.

Please refer to FIG. 3 , which shows a flowchart of an embodiment of a method for using a knowledge model of a conversation system provided by the present application.

As shown in Figure 3, in step 301, a user interest model is constructed based on the user's basic information and the user's historical session data, wherein the user interest model includes at least one knowledge point;

In step 302, matching based on semantic graph is performed on the knowledge points in the user interest model and the knowledge points in the conversation system knowledge model constructed by the method of the above embodiment;

In step 303, at least one matched knowledge point is generated according to the matching result of the semantic graph, wherein the knowledge point is associated with at least one question;

In step 304, the questions associated with the at least one matched knowledge point are recommended to the user.

In this embodiment, for step 301. The device for constructing the knowledge model of the conversation system collects basic information such as the user's age, gender, etc. and the user's historical conversation data based on the user's basic information and the user's historical conversation data to construct a user interest model, for example , based on the data such as "Li Bai" and "Du Fu" in the user's historical conversation to build the user's interest model of Tang poetry, or based on the data of "Su Shi", "Li Qingzhao" in the user's historical conversation to build the user's Song poetry interest model.

Afterwards, for step 302, the knowledge points in the user interest model and the knowledge points in the conversation system knowledge model constructed according to the method described in this application are matched based on the semantic graph, and then, for step 303, according to the matching of the semantic graph As a result, at least one matching knowledge point is generated, and finally, for step 304, the questions associated with the at least one matching knowledge point are recommended to the user.

For example, according to the method for building a knowledge model in the above-mentioned embodiment, a knowledge model is created for "Tang poetry" in the user interest model, and then the knowledge points in "Tang poetry" are matched based on the semantic graph, and then based on the semantic graph matching The result is generated such as "Zhang Jiuling". At least one matching knowledge point such as "Du Mu", and finally recommends questions related to at least one matching knowledge point such as "Zhang Jiuling" and "Du Mu" to the user.

The method of this embodiment constructs a user interest model based on the user's existing data, and then recommends to the user questions that match the knowledge points in the user interest model, so that the user interest model constructed based on the user data can be recommended to the user more suitable for For questions of user interest, the recommended questions have a higher degree of matching with the user, and the user experience is better.

Referring further to FIG. 4 , it shows a flowchart of another embodiment of a method for using a knowledge model of a conversational system provided by the present application. This flowchart is mainly a flowchart for the steps further defined in step 301 in flowchart 3, "constructing a user interest model based on the user's basic information and the user's historical session data". Among them, the user's historical session data includes the user's active questions and answers, and the system's recommended knowledge and answers accepted by the user.

As shown in Figure 4, in step 401, hierarchical clustering is performed on the user's basic information and the user's historical session data to generate a user topic tree;

In step 402, the user's active questions and answers and the user-accepted system recommendation knowledge and answers are constructed into knowledge triples including questions, answers, and associations between questions and answers, wherein the questions and answers form a question-and-answer pair;

In step 403, the knowledge triplet is associated with the user topic tree;

In step 404, using the entity linking technology, the entities in the question and answer pairs in the triples are mapped to the knowledge triples of the conversational system knowledge model, and the mapped knowledge triples are added to the user topic tree. In knowledge points, a user interest model is formed;

In step 405, the semantic similarity between each question-answer pair is calculated, and the semantic similarity is added to the user interest model.

In this embodiment, for step 401, the device for constructing the conversational system knowledge model will perform hierarchical clustering on basic information such as the user's gender, age, and the user's historical conversational data to generate a user topic tree, wherein the hierarchical clustering attempts to The dataset is divided at different levels to form a tree-like clustering structure. The data set can be divided by a "bottom-up" aggregation strategy or a "top-down" splitting strategy. For example, "Tang Poetry" is a cluster, and the next layer of clustering includes "Li Bai", "Du Fu", "Zhang Jiuling" and other clusters, and for example, the "Tang Poetry" cluster contains "Tang Poetry". ”, “Song Ci” and other clusters, and so on, and will not be repeated here.

Then, for step 402, the device for constructing the knowledge model of the conversational system will construct the questions and answers actively asked by the user and the system recommended knowledge and answers accepted by the user into a knowledge ternary including questions, answers, and the relationship between questions and answers For example, first, the user actively asks the device "What is the full text of Li Bai's Jing Ye Si?", and then the device gives the answer and asks the user whether to accept other Li Bai's poems recommended by the system. If the user accepts the system's recommended knowledge, ask the question. , the answer and the system recommendation knowledge form a knowledge triple, in which the question and the answer form a question-and-answer pair;

Afterwards, for step 403, the device for constructing the knowledge model of the conversational system will associate the knowledge triplet with the user topic tree;

Then, for step 404, the device for constructing the conversational system knowledge model uses the entity linking technology to map the entities in the question-and-answer pair in the triplet to the knowledge triplet of the conversational system knowledge model, and maps the mapped knowledge triples to the knowledge triplet of the conversational system knowledge model. Groups are added to the knowledge points under the user topic tree to form a user interest model; among them, entity linking technology is an important method to solve the problem of named entity ambiguity. In order to achieve the elimination of entity ambiguity.

Finally, for step 405, the semantic similarity between each question-answer pair is calculated, and the semantic similarity is added to the user interest model.

The method of this embodiment forms a user topic tree by performing hierarchical clustering on some existing information of the user, then constructs a knowledge triplet, associates the knowledge triplet with the user topic tree, constructs a user interest model, and finally constructs a user interest model. Calculate the semantic similarity of the question and answer pairs in the user interest model and add it to the user interest model, so that a user interest model containing semantic similarity can be constructed, which is convenient for recommending questions with higher voice similarity to users in subsequent heuristic conversations or Answer.

Referring further to FIG. 5 , it shows a flowchart of another embodiment of a method for using a knowledge model of a conversation system provided by the present application. This flow chart is mainly for the flow of steps further defined in the steps before step 401 of "Hierarchical clustering of basic user information and user's historical session data to generate a user topic tree" in Flowchart 4. It is mainly applicable to users with very little historical session data, for example, the user's historical session data may be less than a preset threshold. The preset threshold may be user-defined, which is not limited in this application.

As shown in Figure 5, in step 501, search for users with the same basic information according to the user's basic information;

In step 502, the user's question-and-answer pair set and the knowledge triplet set are constructed based on the user's question-answer pair set and the knowledge triplet set with the same basic information;

In step 503, the semantic similarity between the user's active question and the questions in the constructed question-and-answer pair set is calculated, and the question-and-answer pair corresponding to the top N questions with the highest semantic similarity is selected;

In step 504, calculate the semantic similarity between the answer triplet corresponding to the system recommended answer accepted by the user and the question in the constructed knowledge triplet, and select the top M knowledge triples with the highest semantic similarity;

In step 505, the question-answer pairs corresponding to the top N questions and the top M knowledge triples are used as the user's initial historical conversation data.

In this embodiment, for step 501, if the user's historical session data is particularly small, firstly, the device for constructing the knowledge model of the session system will search for users with the same basic information according to the user's basic information; then, for step 502, based on the same basic information Then, for step 503, the semantic similarity between the user's active question and the questions in the constructed question and answer pair set is calculated, Select the question and answer pairs corresponding to the top N questions with the highest semantic similarity; then, for step 504, calculate the semantic similarity between the answer triplet corresponding to the system recommended answer accepted by the user and the question in the constructed knowledge triplet, The top M knowledge triples with the highest semantic similarity are selected; finally, for step 505, the question and answer pairs corresponding to the top N questions and the top M knowledge triples are used as the user's initial historical session data.

The solution provided by the embodiments of the present application comprehensively considers the knowledge points that users pay attention to, and the association between knowledge points, so as to achieve more fine-grained modeling. The unified modeling of knowledge points can use the graph matching algorithm to find the knowledge points that match the user's interest points. Combined with time sensitivity, location sensitivity and event sensitivity, users can more accurately describe user interests and calculate quickly and accurately. Users' interest points, build the user's initial interest model and recommend more accurate conversational content for users.

In some optional embodiments, the semantic graph matching includes node semantic similarity matching and path semantic similarity matching.

The following describes some problems encountered by the inventor in the process of implementing the present invention and a specific embodiment of the finalized solution, so that those skilled in the art can better understand the solution of the present application.

A simple process of heuristic conversation is as follows: First, the dialogue interaction is actively guided according to the user's question. The user asks a question, and the system will list some related questions based on this question or ask the user if he wants to know. The user's question is connected to the knowledge point in various forms. Of course, after this dialogue, we call it a knowledge point. In the way of a knowledge point, connecting a knowledge point may be for a specific problem, and this problem may have various types. Ask the law, we all think it is a knowledge point.

The invention proposes a personalized heuristic dialogue technology based on knowledge correlation and user interest. First of all, this patent proposes a joint modeling method based on topics and knowledge points. Based on user historical session record data, hierarchical modeling of user interests is carried out, including topic layer and knowledge point layer, and topic layer supports hierarchical structure. Second, this patent models and organizes knowledge points, and the modeling method includes topic abstraction of knowledge points, topic tree construction, and association of knowledge points; third, predictive modeling of new users' interest points to construct initial interest Model. Fourth, match the user's interest graph and knowledge graph, and combine time-sensitive, location-sensitive and event-sensitive to find the knowledge points that users are potentially interested in, and recommend more accurate conversation content for users.

The invention realizes a personalized heuristic conversation method by uniformly modeling the topic association of user interest points and knowledge and the association of knowledge points, improves the efficiency of conversation, and efficiently achieves the communication goal of the user.

The technical innovation of the present invention:

1. Joint modeling based on topics, knowledge points and user interest points.

Most of the traditional user modeling methods use topic modeling. In addition to considering topic information, this patent also comprehensively considers the knowledge points that users pay attention to and the association between knowledge points to achieve more fine-grained modeling, which can be more accurate. User interests are described.

2. Prediction of new users' points of interest.

The traditional user modeling method, which uses the user's history records, often has a cold start problem, which is not suitable for scenarios with only a small number of user history records. According to the association of knowledge points and the similarity between users, the present patent can quickly and accurately calculate the user's interest points, and build a user's initial interest model.

3. User interest point and knowledge point recommendation based on graph matching.

Unified modeling of user interest points and knowledge points can use graph matching algorithm to find knowledge points that match user interest points, and combine time sensitivity, location sensitivity and event sensitivity to recommend more accurate conversation content for users.

FIG. 6 shows a flow chart of a system provided by an embodiment of the present application.

As shown in Figure 6, the system mainly includes three parts. The left part models the interests of users who lack historical session data. Such users only have basic user information or registration information and a small amount of session records. For such users, the system predicts a user interest model based on a small amount of conversation records and a conversational system knowledge model, and gives an initialized user interest model.

The middle part is the users with rich historical session data. Such users have rich session history. Here, the basic information of the user, the user's active questions and answers, the knowledge and answers recommended by the system, and the three types of data pairs that have been accepted by the user are used here. Users conduct interest modeling.

The right part is the modeling process of the knowledge system in the conversation system, including the construction of the topic tree, including the construction of the knowledge map of entities and relationships, the association of knowledge points and topic trees; the sorting of knowledge in question and answer, including the correspondence between questions and answers, May be one-to-one, one-to-many relationships, semantic similarity between questions, associations between entities involved in question-answer pairs and entities in the knowledge graph, etc.

Then, match the user interest model and the system knowledge model based on the semantic graph, calculate the most matching knowledge point, and recommend related questions to the user.

FIG. 7 shows the structure of a user interest point model based on topics and knowledge points.

The data sources of user interest modeling include basic user information, user active questions and answers, knowledge and answers recommended by the system, and three types of data accepted by users.

The user interest model is represented in the form of a semantic graph. Figure 7 shows the model structure, including five elements: topic, knowledge point, question-answer pair, entity, and attribute. The topic layer supports hierarchical structure, that is, a topic can have multiple subtopics, such as topic A includes two sub-topics, topic B and topic C; under the topic are knowledge points, for example, topic D includes knowledge point 1 and knowledge point 2; knowledge points can belong to multiple topics; knowledge points include question-and-answer pairs and knowledge in the knowledge graph, with It is represented in the form of triples, including entities, relationships and attributes, and attribute values. For example, knowledge point 1 includes question-and-answer pair 1 and entities E1, E2 and their relationships.

The first step: According to the user's active questions and answers, the knowledge and answers recommended by the system, do hierarchical clustering to obtain a topic tree with a hierarchical structure. At the same time, the user's active questions and the knowledge recommended by the system are constructed into question-answer pairs or knowledge triples, which are associated with the topic tree.

Step 2: Use entity linking technology to map the entity mentions in the question-and-answer pair to the system knowledge model, and add the mapped knowledge triples to the knowledge points under the corresponding topic tree. Among them, the entity mentioned means the manifestation of the entity in the text. For example, the entity Peking University may be in the form of Peking University in the text.

Step 3: Calculate the semantic similarity between the question-answer pairs and add the semantic similarity to the user interest model.

User Interest Prediction and Interest Model Initialization Based on Semantic Similarity Propagation

For users with less user historical session data, a user interest propagation model based on semantic similarity is proposed. The modified model is used to predict user interest, and the interest model is initialized.

For the sake of illustration, it is assumed that user A has only one active question Q1 that is solved by the question-and-answer pair, and a question Q2 that is recommended by the system and accepted by A is solved by the knowledge triplet.

The first step is to find similar users based on user registration/basic information, such as user set C with similar age, gender, and geographic location, and the question-answer pairs of users in C form the set CQA, and the knowledge triples form the set CKB.

In the second step, the semantic similarity of questions in Q1 and CQA is calculated, and the top N questions are selected as candidates. The specific algorithm can be simple edit distance, set distance, or express the problem as a vector and calculate the cosine distance of the vector.

The third step is to calculate the semantic similarity between the answer triple <e1,r1,e2> corresponding to Q2 and the questions in CKB. Assuming that the triplet in CKB is <e1*,r2,e2*>, the calculation method can be the minimum distance from e1 and e2 to e1* and e2* in the knowledge graph.

In the fourth step, the first N question-and-answer pairs with the highest similarity are taken from the results of the second step, and the first M knowledge triples with the shortest distance are taken from the results of the third step, and N and M are used as the user's initial interest points. The user interest point model is constructed using the method for constructing the model in the foregoing embodiment.

Matching of system knowledge semantic graph and user interest semantic graph: Semantic graph matching considers node semantic similarity and path semantic similarity at the same time.

Figure 8 shows the semantic representation of the user question (upper part), and the semantic representation of the dialogue system body of knowledge (lower part).

Figure 8 presents the semantic representation of the user question and the semantic representation of the dialogue system knowledge system. Suppose the user accesses the system, asks a question Q, and hits question 1, that is, the questions in Q and question 1 are semantically equivalent.

The first step is to construct the path T1 from Q&A 1 to the topic. As shown in the upper part of Figure 8, this path includes knowledge point A of Q&A 1, sub-topic 2, first-level topic 1, and also includes Q&A 1 in All questions and answers under the same knowledge point are recorded as QA1 and all knowledge triples KB1.

The second step is to select the topic path T2 from the system knowledge system, as shown in the lower part of Figure 8, such as primary topic 1 and subtopic 3, knowledge point B, question and answer set QA2, knowledge point triplet KB2

The third step is to calculate the similarity of the topic paths obtained in the first step and the second step, while considering the weight of the topic, the similarity of KB1 and KB2 is averaged, and the similarity of QA1 and QA2 is also averaged. The similarity calculation method can be the edit distance or the cosine distance of the word vector after word embedding.

Sim(T1,T2)=w1*sim(topic 1, topic 3)+w2*sim(subtopic 2,subtopic 4)+average(sim(QA1,QA2))+average(sim(KB1,KB2)) , where sim means similarity, that is, similarity.

The fourth step is to increase the weight of the corresponding QA or knowledge triplet according to time, location, and event sensitivity. For example, if the current time is the Mid-Autumn Festival, then the weight of the QA or knowledge triplet meeting about the Mid-Autumn Festival will be increased. Location and event handling methods are similar.

Update of the semantic graph of the knowledge system of conversational systems and the semantic graph of user interests: when a question is recommended to a user. When the user selects a question recommended by the system, the question will be added to the user interest model. If the user does not select the recommended question, the recommendation index of the question with higher semantic similarity to the question in the user interest model is reduced by a certain value.

This patent recommends more interesting knowledge points for users by integrating the relevance of knowledge and user interest points, thereby completing a more efficient personalized heuristic conversation. Personalized heuristic conversations have the following advantages: First, it improves the efficiency of communication and accelerates convergence to topics that users are interested in. Second, improve user satisfaction.

Please refer to FIG. 9 , which shows a block diagram of an apparatus for constructing a knowledge model of a conversation system provided by an embodiment of the present application.

As shown in FIG. 9 , an apparatus 900 for constructing a knowledge model of a conversation system includes: a knowledge point abstraction module 910 , a topic tree construction module 920 , an association relationship determination module 930 and a knowledge model construction module 940 .

The knowledge point abstraction module 910 is configured to abstract at least one topic associated with the knowledge point from the knowledge point in the conversation system; the topic tree construction module 920 is configured to construct a system topic tree based on the at least one topic; the association A relationship determination module 930, configured to determine an association relationship between knowledge points based on the system topic tree; and a knowledge model building module 940, configured to at least based on the system theme tree and the association relationship between the various knowledge points Build a conversational system knowledge model.

Please refer to FIG. 10 , which shows a block diagram of an apparatus for constructing a knowledge model of a conversation system provided by an embodiment of the present application.

As shown in FIG. 10 , an apparatus 1000 for using a knowledge model of a conversation system includes: a user interest model building module 1010 , a semantic graph matching module 1020 , a matching knowledge point generating module 1030 and a recommendation module 1040 .

The user interest model building module 1010 is configured to build a user interest model based on the basic information of the user and the historical session data of the user, wherein the user interest model includes at least one knowledge point; the semantic graph matching module 1020 is configured to In order to perform semantic graph-based matching between the knowledge points in the user interest model and the knowledge points in the conversation system knowledge model constructed according to the method described in this application; the matching knowledge point generation module 1030 is configured to match the knowledge points according to the semantic graph. The matching result of , generates at least one matched knowledge point, wherein the knowledge point is associated with at least one question; and a recommendation module 1040 is configured to recommend the question associated with the at least one matched knowledge point to the user.

It should be understood that the modules recited in FIGS. 9 and 10 correspond to various steps in the methods described with reference to FIGS. 1 , 2 , 3 , 4 and 5 . Therefore, the operations and features described above with respect to the method and the corresponding technical effects are also applicable to the modules in FIG. 5 , which will not be repeated here.

It is worth noting that the modules in the embodiments of the present disclosure are not used to limit the solution of the present disclosure. For example, the attribute analysis module may be described as a module that analyzes the basic attributes of the user based on the acquired voiceprint information of the user. In addition, the relevant functional modules may also be implemented by a hardware processor, for example, the attribute analysis module may also be implemented by a processor, which will not be repeated here.

In other embodiments, embodiments of the present invention further provide a non-volatile computer storage medium, where the computer storage medium stores computer-executable instructions, and the computer-executable instructions can execute the session system in any of the foregoing method embodiments The construction method of knowledge model;

As an embodiment, the non-volatile computer storage medium of the present invention stores computer-executable instructions, and the computer-executable instructions are set to:

At least one topic associated with the knowledge point is abstracted from the knowledge point in the conversation system;

constructing a system theme tree based on the at least one theme;

Determine the association relationship between the knowledge points based on the system theme tree;

A conversational system knowledge model is constructed at least based on the system topic tree and the association relationship between the knowledge points.

The non-volatile computer-readable storage medium can include a stored program area and a stored data area, wherein the stored program area can store an operating system and an application program required by at least one function; data created by the use of the device, etc. In addition, the non-volatile computer-readable storage medium may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the non-transitory computer-readable storage medium may optionally include memory located remotely from the processor, the remote memory being connectable through a network to the means for constructing the knowledge model of the conversational system. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

An embodiment of the present invention further provides a computer program product, the computer program product includes a computer program stored on a non-volatile computer-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by a computer, the computer is made to execute the above A method for constructing knowledge model of any conversational system.

FIG. 11 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. As shown in FIG. 11 , the device includes: one or more processors 1110 and a memory 1120 . In FIG. 11 , one processor 1110 is used as an example. The apparatus of the method for constructing a knowledge model of a conversational system may further include: an input device 1130 and an output device 1140 . The processor 1110, the memory 1120, the input device 1130, and the output device 1140 may be connected through a bus or in other ways, and the connection through a bus is taken as an example in FIG. 11 . The memory 1120 is the aforementioned non-volatile computer-readable storage medium. The processor 1110 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 1120 , that is, implementing the method for constructing the knowledge model of the conversation system in the above method embodiments. The input device 1130 may receive input numerical or character information, and generate key signal input related to user settings and function control of the above-mentioned devices. The output device 1140 may include a display device such as a display screen.

The above product can execute the method provided by the embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the method. For technical details not described in detail in this embodiment, reference may be made to the method provided by the embodiment of the present invention.

As an embodiment, the above electronic device is applied to an apparatus for constructing a knowledge model of a conversational system for a client, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores a instructions to be executed by at least one processor, the instructions being executed by at least one processor to enable at least one processor to:

Based on the acquired voiceprint information of the user, analyze the basic attributes of the user;

recommending a first set of stories to the user for selection based on the user's underlying attributes;

Judging whether the user selects any story in the first set of stories and recording the user's selection, wherein the any story has at least one story attribute and each story attribute corresponds to a weight value;

Update the weight value of each story attribute of the user based on the user's selection;

Based on the basic attribute of the user and the updated weight value of each story attribute of the user, a second set of stories is recommended to the user for selection.

The electronic devices in the embodiments of the present application exist in various forms, including but not limited to:

(1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions, and its main goal is to provide voice and data communication. Such terminals include: smart phones (eg iPhone), multimedia phones, feature phones, and low-end phones.

(2) Ultra-mobile personal computer equipment: This type of equipment belongs to the category of personal computers, has computing and processing functions, and generally has the characteristics of mobile Internet access. Such terminals include: PDAs, MIDs, and UMPC devices, such as iPads.

(3) Portable entertainment equipment: This type of equipment can display and play multimedia content. Such devices include: audio and video players (eg iPod), handheld game consoles, e-books, as well as smart toys and portable car navigation devices.

(4) Server: A device that provides computing services. The composition of the server includes a processor, a hard disk, a memory, a system bus, etc. The server is similar to a general computer architecture, but due to the need to provide highly reliable services, the processing power, stability , reliability, security, scalability, manageability and other aspects of high requirements.

(5) Other electronic devices with data interaction function.

The device embodiments described above are only illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place , or distributed to multiple network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

From the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and certainly can also be implemented by hardware. Based on this understanding, the above-mentioned technical solutions can be embodied in the form of software products in essence or the parts that make contributions to the prior art, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic Disks, optical discs, etc., include instructions for causing a computer device (which may be a personal computer, server, or network device, etc.) to perform the methods of various embodiments or portions of embodiments.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or some technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (5)

1. A method of using a conversational system knowledge model, comprising:

constructing a user interest model based on basic information of a user and historical conversation data of the user, wherein the user interest model comprises at least one knowledge point;

matching the knowledge points in the user interest model and the knowledge points in the session system knowledge model based on a semantic graph;

generating at least one matched knowledge point according to a matching result of the semantic graph, wherein the knowledge point is associated with at least one question;

recommending a question associated with the at least one matched knowledge point to the user,

wherein the construction of the session system knowledge model comprises the following steps:

abstracting, by a knowledge point in a conversational system, at least one topic associated with the knowledge point;

building a system topic tree based on the at least one topic;

determining an incidence relation between knowledge points based on the system theme tree;

constructing a conversation system knowledge model at least based on the incidence relation between the system subject tree and the knowledge points,

the method for constructing the user interest model based on the basic information of the user and the historical conversation data of the user comprises the following steps of:

carrying out hierarchical clustering on the basic information of the user and the historical conversation data of the user to generate a user theme tree;

constructing the active questions and answers of the users and the system recommendation knowledge and answers received by the users into knowledge triples containing questions, answers and incidence relations among the questions and the answers, wherein the questions and the answers form question-answer pairs;

associating the knowledge triples onto the user topic tree;

mapping the entities in the question-answer pairs in the triples to the knowledge triples of the session system knowledge model by using an entity link technology, and adding the mapped knowledge triples to the knowledge points under the user topic tree to form a user interest model;

calculating semantic similarity between the question-answer pairs, adding the semantic similarity into the user interest model,

if the historical conversation data of the user is less than a preset threshold, before the hierarchical clustering is performed on the basic information of the user and the historical conversation data of the user to generate the user topic tree, the method further comprises the following steps:

searching users with the same basic information according to the basic information of the users;

constructing a question-answer pair set and a knowledge triple set of the user based on the question-answer pair set and the knowledge triple set of the user with the same basic information;

calculating the semantic similarity between the active questions of the user and the questions in the constructed question-answer pair set, and selecting question-answer pairs corresponding to the first N questions with the highest semantic similarity;

calculating semantic similarity between answer triples corresponding to the system recommended answers received by the user and the questions in the constructed knowledge triples, and selecting the first M knowledge triples with the highest semantic similarity;

and taking the question-answer pairs corresponding to the first N questions and the first M knowledge triples as initial historical conversation data of the user.

2. The method of claim 1, wherein the semantic graph matching comprises node semantic similarity matching and path semantic similarity matching.

3. An apparatus for using a knowledge model of a conversational system, comprising:

the user interest model building module is configured to build a user interest model based on basic information of a user and historical conversation data of the user, wherein the user interest model comprises at least one knowledge point;

the semantic graph matching module is configured to perform semantic graph-based matching on the knowledge points in the user interest model and the knowledge points in the session system knowledge model;

the matching knowledge point generating module is configured to generate at least one matching knowledge point according to a matching result of the semantic graph, wherein the knowledge point is associated with at least one question;

a recommendation module configured to recommend questions associated with the at least one matched knowledge point to the user,

wherein the construction of the session system knowledge model comprises the following steps:

abstracting at least one topic associated with a knowledge point in a conversation system by the knowledge point;

building a system topic tree based on the at least one topic;

determining an incidence relation among knowledge points based on the system theme tree;

constructing a conversation system knowledge model at least based on the incidence relation between the system theme tree and the knowledge points,

the method for constructing the user interest model based on the basic information of the user and the historical session data of the user comprises the following steps:

carrying out hierarchical clustering on the basic information of the user and the historical conversation data of the user to generate a user theme tree;

constructing the active questions and answers of the users and the system recommendation knowledge and answers received by the users into knowledge triples containing questions, answers and incidence relations among the questions and the answers, wherein the questions and the answers form question-answer pairs;

associating the knowledge triples onto the user topic tree;

mapping the entities in the question-answer pairs in the triples to the knowledge triples of the session system knowledge model by using an entity linking technology, and adding the mapped knowledge triples to the knowledge points under the user topic tree to form a user interest model;

calculating the semantic similarity between the question-answer pairs, adding the semantic similarity into the user interest model,

if the historical conversation data of the user is less than a preset threshold value, before hierarchical clustering is performed on the basic information of the user and the historical conversation data of the user to generate a user theme tree:

searching users with the same basic information according to the basic information of the users;

constructing a question-answer pair set and a knowledge triple set of the user based on the question-answer pair set and the knowledge triple set of the user with the same basic information;

calculating the semantic similarity between the active questions of the user and the questions in the constructed question-answer pair set, and selecting question-answer pairs corresponding to the first N questions with the highest semantic similarity;

calculating semantic similarity between answer triples corresponding to the system recommended answers received by the user and the questions in the constructed knowledge triples, and selecting the first M knowledge triples with the highest semantic similarity;

and taking the question-answer pairs corresponding to the first N questions and the first M knowledge triples as initial historical conversation data of the user.

4. An electronic device, comprising: at least one processor, and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the steps of the method of any of claims 1-2.

5. A storage medium having stored thereon a computer program, characterized in that the program, when being executed by a processor, is adapted to carry out the steps of the method of any one of claims 1 to 2.

Images