CN115982453A - Content recommendation method and device based on feature engineering, electronic equipment and medium - Google Patents

Abstract

The application provides a content recommendation method and device based on feature engineering, electronic equipment and a medium, and relates to the technical field of artificial intelligence. The method comprises the following steps: performing feature coding processing on the current click position information according to the feature engineering configuration information to obtain current position features; reading target object characteristics from a preset object code file according to current object information; reading target content characteristics from a preset content coding file according to the current clicked content information; performing feature extraction according to the feature extraction information to obtain target feature coded data; click prediction is carried out on the target feature coded data through a preset prediction model, and a target click score is obtained; when the prediction model performs click prediction on the target feature coded data, the position features, the object features and the content features are combined, so that the target click score obtained through prediction is more accurate, and the accuracy of content recommendation according to the target click score is higher.

Description

Method and device for content recommendation based on feature engineering, electronic equipment, medium

technical field

The present application relates to the technical field of artificial intelligence, in particular to a feature engineering-based content recommendation method and device, electronic equipment, and media.

Background technique

Feature engineering is a process of transforming raw data into features that better represent business logic. At this stage, content recommendation based on feature engineering generally includes: obtaining original data, then performing feature encoding on the original data, and then performing click prediction on the encoded data through an offline-developed prediction model to obtain the prediction result, and recommending content based on the prediction result . In online content recommendation, the real-time performance of online data is high, but the prediction model developed offline is trained from offline data, and the real-time performance of offline data is low. Due to the real-time inconsistency between online data and offline data, when the prediction model developed offline is applied to online deployment, the accuracy of recommended content is affected. Therefore, how to provide a content recommendation method based on feature engineering, which can improve the accuracy of recommended content, has become an urgent technical problem to be solved.

Contents of the invention

The main purpose of the embodiments of the present application is to provide a content recommendation method and device, electronic equipment, and media based on feature engineering, which can improve the accuracy of recommended content.

In order to achieve the above purpose, the first aspect of the embodiment of the present application proposes a content recommendation method based on feature engineering, the method includes:

Acquiring current click data, the current click data includes current object information of the clicked object, current click content information, and current click position information;

Read feature engineering configuration information and feature extraction information from a preset configuration file;

performing feature encoding processing on the current click position information according to the feature engineering configuration information to obtain the current position feature;

Reading target object characteristics from a preset object encoding file according to the current object information;

Reading target content characteristics from a preset content encoding file according to the currently clicked content information;

performing feature extraction on the current location feature, the target object feature, and the target content feature according to the feature extraction information to obtain target feature encoding data;

Decoding the target feature encoding data to obtain a target object, target display content, and target display position; wherein, the target object includes the click object, or is not the click object;

Click prediction is performed on the target feature coding data through a preset prediction model to obtain a target click score; the target click score is used to represent: the target display content at the target display position is clicked by the target object probability;

Recommending the target display content to the target object according to the target click score and the target display position.

In some embodiments, the current click data further includes: the current click time, and after performing click prediction on the target feature encoding data through a preset prediction model to obtain a target click score, the method further includes:

performing feature encoding processing on the current object information according to the feature engineering configuration information to obtain the current object features;

performing feature encoding processing on the currently clicked content information according to the feature engineering configuration information to obtain current content features;

performing feature extraction on the current object feature, the current content feature, and the current location feature according to the feature extraction information to obtain current feature encoding data;

If the target characteristic encoding data is the same as the current characteristic encoding data, then obtain the display time of the target characteristic encoding data;

If the target characteristic encoding data is the same as the current characteristic encoding data, then obtain the display time of the target characteristic encoding data;

calculating the difference between the current click time and the display time to obtain a time difference;

The target click score is adjusted according to the time difference.

In some embodiments, before performing click prediction on the target feature encoding data through a preset prediction model to obtain a target click score, the method further includes:

Pre-training the prediction model specifically includes:

Obtain historical display data, and obtain the original label of the historical display data; the historical display data includes historical object information, historical display content information, and historical display position information of historical objects;

performing feature encoding processing on the historical object information according to the feature engineering configuration information to obtain historical object features;

performing feature encoding processing on the historical display content information according to the feature engineering configuration information to obtain historical content features;

performing feature encoding processing on the historical display location information according to the feature engineering configuration information to obtain historical location features;

performing feature extraction on the historical object features, historical content features and the historical location features according to the feature extraction information to obtain training feature encoding data;

determining a raw training score for the training feature-encoded data based on the raw label;

performing score prediction on the training feature encoding data through the prediction model to obtain a target training score;

performing loss calculation according to the original training score and the target training score to obtain a loss value;

The parameters of the prediction model are adjusted according to the loss value to obtain the trained prediction model.

In some embodiments, the obtaining the historical display data and obtaining the original label of the historical display data includes:

Obtaining an object behavior log of the historical object;

Reading information from the object behavior log to obtain the historical object information, the historical display content information, and the historical display position information;

combining the historical object information, the historical display content information and the historical display position information to obtain the historical display data;

Read the time of the object behavior log to obtain the content marking time of the historical display data, the content marking time includes: content exposure time or content click time;

The original tag is determined according to the tagging time of the content.

In some embodiments, before reading the object behavior log, the method further includes:

Update the object behavior log, specifically including:

acquiring object data and an object identification code, the object data being used to provide the historical object information;

Obtaining content data and content identification codes and using the content data to provide the historical content information;

Recording the object identification code and the content identification code in the object behavior log;

The object data is associated with the object behavior log through the object identification code, and the content data is associated with the object behavior log through the content identification code.

In some embodiments, the object data includes object basic information and object behavior information, the content data includes content basic information and content statistical information, and the object data is associated with the object behavior through the object identification code log, and associate the content data with the object behavior log through the content identification code, including:

Establishing a mapping relationship between the object basic information and the object behavior information to obtain an object wide table;

Establishing a mapping relationship between the content basic information and the content statistical information to obtain a wide content table;

The object wide table is associated with the object behavior log through the object identification code, and the content wide table is associated with the object behavior log through the content identification code.

In some embodiments, the reading feature engineering configuration information and feature extraction information from a preset configuration file includes:

Get the current feature configuration version number;

Reading the feature engineering configuration information and the feature extraction information from the configuration file according to the current feature configuration version number.

In order to achieve the above purpose, the second aspect of the embodiment of the present application proposes a content recommendation device based on feature engineering, the device includes:

The click data acquisition module is used to acquire the current click data, the current click data includes the current object information of the click object, the current click content information, and the current click position information;

The configuration information acquisition module is used to read feature engineering configuration information and feature extraction information from a preset configuration file;

A position feature determination module, configured to perform feature encoding processing on the current click position information according to the feature engineering configuration information to obtain the current position feature;

An object feature determination module, configured to read target object features from a preset object encoding file according to the current object information;

A content feature determination module, configured to read target content features from a preset content encoding file according to the currently clicked content information;

A feature extraction module, configured to perform feature extraction on the current location feature, the target object feature, and the target content feature according to the feature extraction information, to obtain target feature encoding data;

A feature decoding module, configured to decode the target feature encoding data to obtain a target object, target display content, and target display position; wherein, the target object includes the click object, or is not the click object;

A model prediction module, configured to perform click prediction on the target feature encoding data through a preset prediction model to obtain a target click score; the target click score is used to represent: the target display content at the target display position is The probability of the target object being clicked;

A content recommendation module, configured to recommend the target display content to the target object according to the target click score and the target display position.

In order to achieve the above purpose, the third aspect of the embodiments of the present application proposes an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned computer program when executing the computer program. The method described in the first aspect.

In order to achieve the above purpose, the fourth aspect of the embodiments of the present application proposes a storage medium, the storage medium is a computer-readable storage medium, the storage medium stores a computer program, and the computer program is implemented when the computer program is executed by a processor. The method described in the first aspect above.

The feature engineering-based content recommendation method and device, electronic equipment, and media proposed in this application perform feature encoding and feature extraction by acquiring multi-dimensional information such as object information, content information, and location information to obtain target feature encoding data. The data includes not only location features, but also object features and content features. In this way, when the prediction model performs click prediction on the target feature encoding data, the dimension of the features is increased, and the location features, object features and content features are combined to make the prediction The obtained target click score is more accurate, so that the accuracy of content recommendation based on the target click score is higher.

Description of drawings

FIG. 1 is a flowchart of a feature engineering-based content recommendation method provided in an embodiment of the present application;

FIG. 2 is a flowchart of a content recommendation method based on feature engineering provided by another embodiment of the present application;

Fig. 3 is the flowchart of step S201 in Fig. 2;

FIG. 4 is a flowchart of a content recommendation method based on feature engineering provided by another embodiment of the present application;

Fig. 5 is the flowchart of step S404 in Fig. 4;

FIG. 6 is a flowchart of a content recommendation method based on feature engineering provided by another embodiment of the present application;

FIG. 7 is a block diagram of a module structure of a content recommendation device based on feature engineering provided by an embodiment of the present application;

FIG. 8 is a schematic diagram of a hardware structure of an electronic device provided by an embodiment of the present application.

Detailed ways

In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, not to limit the present application.

It should be noted that although the functional modules are divided in the schematic diagram of the device, and the logical sequence is shown in the flowchart, in some cases, it can be executed in a different order than the module division in the device or the flowchart in the flowchart. steps shown or described. The terms "first", "second" and the like in the specification and claims and the above drawings are used to distinguish similar objects, and not necessarily used to describe a specific sequence or sequence.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application, and are not intended to limit the present application.

First, analyze some nouns involved in this application:

Artificial Intelligence (AI): It is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence; artificial intelligence is a branch of computer science. Intelligence attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a manner similar to human intelligence. Research in this field includes robotics, language recognition, image recognition, natural language processing, and expert systems. Artificial intelligence can simulate the information process of human consciousness and thinking. Artificial intelligence is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

Feature engineering is a process of transforming raw data into features that better represent business logic, so that applying these features to predictive models can improve model prediction accuracy for unseen data and improve machine learning performance. For example, in the fields of search, advertising, recommendation, etc., the data determines the upper limit of the effect, and the algorithm can only determine the degree of approaching the upper limit, while feature engineering is the bridge between data and algorithms. Feature engineering has a big impact on business outcomes. The main reason is that feature engineering can significantly improve model performance, greatly simplify model complexity, and reduce model maintenance costs. High-quality features can also show good results on simple linear models.

At this stage, content recommendation based on feature engineering generally includes: obtaining online data, then performing feature encoding on the online data, and then performing click prediction on the encoded data through an offline-developed prediction model to obtain the prediction result, and based on the prediction result. Content recommendation. In online content recommendation, the real-time performance of online data is high, but the prediction model developed offline is trained from offline data, and the real-time performance of offline data is low. Due to the real-time inconsistency between online data and offline data, when the prediction model developed offline is applied to online deployment, the accuracy of recommended content is affected. Therefore, how to provide a content recommendation method based on feature engineering, which can improve the accuracy of recommended content, has become an urgent technical problem to be solved.

Based on this, the main purpose of the embodiment of the present application is to propose a content recommendation method based on feature engineering and a content recommendation device, electronic device, and storage medium based on feature engineering, and perform feature Extraction to obtain target feature coding data, the target feature coding data includes not only high real-time location features, but also relatively low real-time object features and content features, so that the prediction model is predicting the target feature coding data In this case, it can not only combine the high real-time online features, but also take into account the relatively low real-time offline features, so that the prediction result (target click score) is more accurate, so that the accuracy of content recommendation based on the target click score is higher. In addition, in the content recommendation in the embodiment of the present application, not only the target click score but also the target display position are considered, which helps to improve the click rate of the target object on the target display content.

The content recommendation method based on feature engineering provided in the embodiment of the present application is applied to the server, and may also be software running on the terminal or the server. The server side can be configured as an independent physical server, or as a server cluster or distributed system composed of multiple physical servers, and can also be configured to provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, Cloud servers for basic cloud computing services such as cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; software can be applications that implement content recommendation methods based on feature engineering, but are not limited to above form.

The application can be used in numerous general purpose or special purpose computer system environments or configurations. For example: server computers, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, distributed computing environments including any of the above systems or devices, and the like. This application may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including storage devices.

The embodiment of the present application provides a content recommendation method based on feature engineering and a content recommendation device, electronic device, and storage medium based on feature engineering. Specifically, the following embodiments will be used to describe the content recommendation method based on feature engineering in the embodiment of the present application. .

It should be noted that, in each specific implementation of the present application, when it comes to related processing based on user information, user behavior data, user history data, user location information and other data related to user identity or characteristics, all will first Obtain the user's permission or consent, and the collection, use and processing of these data will comply with the relevant laws, regulations and standards of the relevant countries and regions. In addition, when the embodiment of this application needs to obtain the user's sensitive personal information, the user's separate permission or separate consent will be obtained through a pop-up window or jump to a confirmation page, etc. After the user's separate permission or separate consent is clearly obtained, the Obtain necessary user-related data for the normal operation of this embodiment of the application.

Fig. 1 is an optional flow chart of the content recommendation method based on feature engineering provided by the embodiment of the present application, which may include but not limited to steps S101 to S109.

Step S101, obtaining current click data, which includes current object information of the clicked object, current click content information, and current click position information;

Step S102, reading feature engineering configuration information and feature extraction information from a preset configuration file;

Step S103, performing feature encoding processing on the current click position information according to the feature engineering configuration information to obtain the current position feature;

Step S104, read the target object characteristics from the preset object encoding file according to the current object information;

Step S105, read the target content characteristics from the preset content encoding file according to the currently clicked content information;

Step S106, perform feature extraction on the current location feature, target object feature and target content feature according to the feature extraction information, to obtain target feature encoding data;

Step S107, decoding the target feature coded data to obtain the target object, target display content and target display position; wherein, the target object includes a click object or a non-click object;

In step S108, click prediction is performed on the target feature coding data through a preset prediction model to obtain a target click score; the target click score is used to represent: the probability that the target display content at the target display position is clicked by the target object;

Step S109, recommending the target display content to the target object according to the target click score and the target display position.

From steps S101 to S109 shown in the embodiment of the present application, by acquiring multi-dimensional information such as object information, content information, and location information, feature encoding and feature extraction are performed to obtain target feature encoding data, the target feature encoding data includes location features, and Including object characteristics and content characteristics. Because the location feature is obtained by clicking on the object, the real-time performance is high, while the object feature and content feature are obtained from the encoded file, and the real-time performance is relatively low. In this way, when the prediction model performs click prediction on the target feature encoding data, it can not only combine the features with high real-time performance, but also take into account the features with relatively low real-time performance, so that the prediction result (target click score) is more accurate. The accuracy rate of content recommendation is higher when the target click score is used. In addition, in the content recommendation in the embodiment of the present application, not only the target click score but also the target display position are considered, which helps to improve the click rate of the target object on the target display content.

In step S101 of some embodiments, current click data may be obtained from the client. Taking video recommendation as an example, record the clicked videos in the click log. Therefore, in an example, the click log of the client is obtained, the current object information, current click content information and current click position information of the click object are read from the click log, and the current object information, current click content information and current click position information Merge to get the current click data.

In another example, obtain the current browsing page of the client; obtain the object identification code according to the login account of the current browsing page; read the current object information from the preset object database according to the object identification code; receive a click signal; respond to the click signal, Obtain current click content information and current click position information from the browsing page; combine current object information, current click content information and current click position information to obtain current click data.

It should be noted that the embodiments of the present application refer to click objects, history objects and target objects. The above three objects all refer to users, but may be different users. For example, client 1 corresponds to an object A, and client 2 corresponds to an object B. If the object A clicks in real time on the browsing page of the client 1, then the object A is the clicked object, and the current click data of the object A can be acquired in this embodiment of the present application. Since the embodiment of the present application will use the current object information to read the target object feature from the preset object encoding file, and the target object feature may be the object feature of object A or the object feature of object B, therefore, according to the target object The target object obtained by the feature may be the clicked object, or other objects other than the clicked object. The historical object appears in the pre-training model stage of this application, regardless of whether object A or object B clicks on the browsing page, it can be a historical object.

Both current object information and historical object information may refer to object data. Object data generally includes object basic information and object behavior information. The basic information of the object includes gender, age, current city, etc. The object behavior information includes the news list clicked, the news list favorited, the news list liked, etc.

Both current click content information and historical display content information refer to content data, and content data includes content basic information and content statistical information; content basic information includes primary/secondary categories, publishers, keywords, etc. Content statistics include favorites, likes, comments, etc.

The current click position information refers to the display position of the content on the browsing page when the click object clicks on the content. Taking video recommendation as an example, the client displays multiple videos on the browsing page at the same time, and each video has a corresponding display position on the browsing page. The browsing page will change the displayed video with the operation of the object, that is, the display position of the video will change with the operation of the object during the display process. When the click signal of the object is received, the current click position is obtained according to the display position of the clicked video in response to the click signal.

In step S102 of some embodiments, feature engineering configuration information and feature extraction information are set in a configuration file in advance. The feature engineering configuration information is mainly to encode the features of the data. For example, the feature encoding process is performed on the current click position information according to the feature engineering configuration information to obtain the current position feature.

In an example, step S102 may include: acquiring a current feature configuration version number; reading feature engineering configuration information and feature extraction information from a configuration file according to the current feature configuration version number.

Specifically, feature configuration file: feature engineering configuration information and feature extraction information are configurable, and the feature configuration file is the same in the offline development stage and the online real-time recommendation stage. It can solve the problem that the project code needs to be updated frequently, and the release cycle is long, and effectively improve the efficiency of feature iteration. In addition, different feature iterative experiments and feature versions can be configured to make feature versions controllable.

It should be noted that the configuration file in the embodiment of the present application is the same in the offline development stage and the online real-time recommendation stage. Due to the difference between offline data and online data, the feature engineering configuration information obtained from the configuration file may not be able to process online data. For example, in the browsing page of the client, if the browsing page M displays three videos at a time, the display positions of the videos can be divided into three display positions: upper, middle and lower. If the click object clicks on the video whose position is "up", then the current position information is "up". If the browsing page N displays five videos at a time, the display positions of the videos can be divided into five display positions: upper, upper middle, middle, lower middle, and lower. If the click object clicks on the video whose position is "upper middle", then the current position information is "upper middle". It can be found that different location information may be obtained from different browsing pages, and the location information of browsing page M and browsing page N cannot be treated equally.

Therefore, in an example, before step S103, the mapping configuration information is obtained, and the format of the current location information is adjusted through the mapping configuration information to obtain the adjusted current location information. Then perform feature encoding on the adjusted current position information according to the feature engineering configuration information to obtain the current position feature. Specifically, after format adjustment, the "upper middle" of the browsing page will correspond to either "upper" or "middle".

In another example, the result of format adjustment is detected, and if the result is that the format cannot be adjusted, the historical location information of the last click is obtained, and then feature encoding is performed on the historical location information according to the feature engineering configuration information to obtain the current location feature. In addition, if the result is that the format cannot be adjusted, a reminder message is sent to the server. The reminder message is used to feed back the fact that the feature engineering configuration information cannot be encoded normally, so that developers can update or optimize the configuration file.

In step S103 of some embodiments, feature encoding is performed on the current location information according to the feature engineering configuration information to obtain the current location feature. The feature engineering configuration information includes: information to be processed and matching information of feature processing methods. For example, feature engineering configuration information includes: object information, object feature processing method; content information, content feature processing method; location information, location feature processing method, etc. Therefore, in an example, the location feature processing method is read from the feature engineering configuration information according to the current location information, and the current location information is subjected to feature processing through the location feature processing method to obtain the current location feature.

In step S104 of some embodiments, the characteristics of the target object are read from a preset object encoding file according to the current object information. Specifically, the object encoding file includes multiple candidate object characteristics such as gender characteristics, age characteristics, educational background characteristics, and information click sequence characteristics. The current object information includes gender information, age information, educational background information, and the like. A plurality of candidate object features are screened according to the current object information to obtain target object features; wherein the target object features include at least one candidate object feature. For example, if the current object information includes gender information and age information, read the gender characteristics from the object encoding file according to the gender information, and read the age characteristics from the object encoding file according to the age information, and splice the gender characteristics and age characteristics, Get the characteristics of the target object.

In step S105 of some embodiments, the target content feature is read from a preset content encoding file according to the currently clicked content information. Specifically, the content coding file includes multiple candidate content features such as primary/secondary category features, keyword features, click volume features, collection volume features, like volume features, and comment volume features. The current content information includes primary/secondary category information, keyword information, click volume information, collection volume information, like volume information, comment volume information, etc. A plurality of candidate content features are screened according to the current content information to obtain a target content feature; wherein, the target content feature includes at least one candidate content feature. For example, if the current content information includes keyword information and comment amount information, then read the keyword features from the content encoding file according to the keyword information, and read the comment amount features from the content encoding file according to the comment amount information, and convert the keyword features Splicing with the comment quantity feature to get the target content feature.

In step S106 of some embodiments, feature extraction is performed on the current location feature, the target object feature and the target content feature according to the feature extraction information to obtain target feature encoding data. Specifically, the feature extraction information is mainly to perform feature extraction on multiple features, and this feature extraction is also called feature merging or feature splicing. For example, according to the feature extraction information, the current location feature, the target object feature and the target content feature are feature spliced to obtain the target feature encoding data.

In an example, before performing feature splicing, feature dimension processing may be performed on the current location feature, target object feature, and target content feature, so that the current location feature, target object feature, and target content feature all have the same feature dimension.

In another example, step S106 specifically includes: obtaining the feature category of the target content feature to obtain the original feature category; if the original feature category is a continuous category, performing discretization processing on the target content feature to obtain the target discrete content feature; The location features, target object features and target discrete content features are combined to obtain target feature encoding data.

For example, the number of likes feature in the target content feature is a continuous feature. In the model prediction, it is necessary to discretize the number of likes feature to get better results. Therefore, the number of likes feature needs to be discretized, such as log(likes Quantitative features*10+1).

In step S107 of some embodiments, the target characteristic encoding data is decoded to obtain the target object, the target display content and the target display position; wherein, the target object includes a click object or a non-click object. Specifically, feature separation is performed on the target feature encoding data to obtain object features to be decoded, content features to be decoded, and location features to be decoded; object features to be decoded, content features to be decoded, and location features to be decoded are respectively decoded to obtain the target object , targeted display content, and targeted placement. It should be noted that the characteristics of the object to be decoded are not completely the same as the characteristics of the target object, the characteristics of the content to be decoded are not completely the same as the characteristics of the target content, and the characteristics of the location to be decoded are not completely the same as the characteristics of the current location. Since there is no direct connection between the current location features, target content features, and target object features, if feature extraction is not performed through feature extraction information, but feature decoding and model prediction of candidate steps are directly performed, it may affect the prediction model for click prediction the accuracy rate.

In an example, the features of the object to be decoded are decoded to obtain the target object, which specifically includes:

Perform decoding processing on the characteristics of the object to be decoded to obtain target object information; the target object information includes at least one of the following: gender information, age information, and education information;

Filter out the target object from the preset object set according to the target object information.

Specifically, if the target object information is gender information, target objects with consistent gender information are screened out from the object set. If the target object information is age information, an age interval is determined according to the age information, and target objects meeting the age interval are screened out from the object set.

In another example, the features of the content to be decoded are decoded to obtain the target display content, which specifically includes:

Perform decoding processing on the characteristics of the content to be decoded to obtain target content information; the target content information includes at least one of the following: keyword information, number of likes information;

The target display content is filtered out from the preset content set according to the target content information.

Specifically, if the target content information includes keyword information, the target display content including the keyword information is screened out from the content set. If the target content information includes the number of likes, the likes threshold is determined according to the likes information, and the target display content greater than the likes threshold is screened out from the content set.

It should be noted that the prediction model in this embodiment of the present application may be a logistic regression model (LogisticRegression, LR), or an online learning model (Follow the Regularized Leader, FTRL).

Please refer to FIG. 2. In some other embodiments, before step S108, the content recommendation method based on feature engineering further includes: pre-training the prediction model, specifically including but not limited to steps S201 to S207:

Step S201, obtaining historical display data, and obtaining the original label of the historical display data; the historical display data includes historical object information, historical display content information, and historical display location information of historical objects;

Step S202, perform feature encoding processing on historical object information according to feature engineering configuration information to obtain historical object features; perform feature encoding processing on historical display content information according to feature engineering configuration information to obtain historical content features; The location information is subjected to feature encoding processing to obtain historical location features;

Step S203, perform feature extraction on historical object features, historical content features and historical location features according to the feature extraction information to obtain training feature encoding data;

Step S204, based on the original label, determine the original training score of the training feature encoding data;

Step S205, predicting the score of the training feature coding data through the prediction model to obtain the target training score;

Step S206, performing loss calculation according to the original training score and the target training score to obtain a loss value;

Step S207, adjusting the parameters of the prediction model according to the loss value to obtain a trained prediction model.

Steps S201 to S207 illustrated in the embodiment of the present application, feature encoding is performed on historical display data through feature engineering configuration information to obtain offline features, which include historical object features, historical content features, and historical location features. Input the offline features into the prediction model for click prediction, and obtain the target training score. The target training score is used to represent the predicted probability that the historical content at the historical position is clicked by the historical object. The original tag is used to indicate whether the historical content has been clicked by the historical object. If yes, the original label is a positive label, if not, the original label is a negative label. If the original label is positive, the original training score is higher, and if the original label is negative, the original training score is lower. The loss value is determined according to the original training score and the target training score, and the parameters of the prediction model are adjusted according to the loss value to obtain the trained prediction model. In an example, the difference calculation is performed on the original training score and the target training score to obtain a loss value.

In an example, the original tag marks the time of the content, and the time of the content marking includes the exposure time of the content or the click time of the content. If the content tagging time is the content exposure time, use the preset exposure score as the original training score; if the content tagging time is the content click time, then calculate the difference between the content exposure time and the content click time to get the time coefficient, according to the time coefficient The product of the exposure score and the exposure score is calculated to obtain the original training score. The greater the difference between the content exposure time and the content click time, the smaller the time coefficient. However, the time coefficient is at least greater than the preset coefficient threshold, and the preset coefficient threshold is at least 1.

Please refer to FIG. 3 , in some other embodiments, step S201 may include but not limited to include steps S301 to S305:

Step S301, acquiring object behavior logs of historical objects;

Step S302, read information from the object behavior log to obtain historical object information, historical display content information, and historical display location information;

Step S303, combining historical object information, historical display content information and historical display location information to obtain historical display data;

Step S304, read the time of the object behavior log to obtain the content marking time of the historical display data, the content marking time includes: content exposure time or content click time;

Step S305, determining the original tag according to the tagged time of the content.

From step S301 to step S305 shown in the embodiment of the present application, the object behavior log stores historical object information, historical display content information, historical display position information and content annotation time. By reading object behavior logs, historical display data and original labels can be quickly obtained.

It should be noted that, in addition to object behavior logs, historical object information can be stored in the hbase database, historical display content information can be stored in the redis database, historical location information can be stored in the redis database, and content annotation time can be stored in the redis database. In another embodiment, step S201 may further include: obtaining historical object information, historical display content information, historical display location information, and content annotation time from the hbase database and redis data through the data interface.

Please refer to FIG. 4 , in some other embodiments, before step S301, the content recommendation method based on feature engineering further includes: updating the object behavior log, specifically including but not limited to steps S401 to S404:

Step S401, acquiring object data and object identification code, the object data is used to provide historical object information;

Step S402, acquiring content data and content identification code, the content data is used to provide historical content information;

Step S403, recording the object identification code and the content identification code in the object behavior log;

Step S404, associate the object data with the object behavior log through the object identification code, and associate the content data with the object behavior log through the content identification code.

From step S401 to step S404 shown in the embodiment of the present application, since the data volume of the object data and content data is too large, if the object data and content data are directly stored in the object behavior log, the data volume of the object behavior log will be too large. Affects the reading efficiency of object behavior logs. Therefore, in the embodiment of the present application, an object identification code is set for object data, and a content identification code is set for content data, and data association is performed through the identification code, which reduces the storage capacity of object behavior logs and improves reading efficiency.

Please refer to FIG. 5. In some other embodiments, the object data includes object basic information and object behavior information, and the content data includes content basic information and content statistical information. Step S404 specifically includes but is not limited to steps S501 to S503:

Step S501, establishing a mapping relationship between the object basic information and the object behavior information, and obtaining the object wide table;

Step S502, establishing a mapping relationship between content basic information and content statistical information to obtain a wide content table;

Step S503, associate the object wide table with the object behavior log through the object identification code, and associate the content wide table with the object behavior log through the content identification code.

From step S501 to step S503 shown in the embodiment of this application, user basic information includes gender, age, current city, etc., and user behavior information includes clicked information list, favorited or liked information list, etc. The object wide table refers to the association of user basic information and behavior information to form a full user portrait table. Basic content information includes primary/secondary categories, publishers, keywords, etc. Content statistical information includes favorites, likes, comments, etc. The content wide table refers to the association of content basic information and content statistical information to form a full content portrait table. By establishing wide tables for object data and content data, it is convenient to manage and read data.

In step S108 of some embodiments, click prediction is performed on the target feature coding data through a preset prediction model to obtain a target click score; the target click score is used to represent: the probability that the target display content at the target display position is clicked by the target object .

Please refer to FIG. 6. In some other embodiments, the current click data also includes: the current click time. After step S108, the content recommendation method based on feature engineering may also include, but is not limited to, step S601 to step S605:

Step S601, perform feature encoding processing on the current object information according to the feature engineering configuration information to obtain current object features; perform feature encoding processing on the currently clicked content information according to the feature engineering configuration information to obtain current content features;

Step S602, perform feature extraction on the current object feature, current content feature and current location feature according to the feature extraction information to obtain current feature encoding data;

Step S603, if the target characteristic coding data is the same as the current characteristic coding data, obtain the display time of the target characteristic coding data;

Step S604, calculating the difference between the current click time and the display time to obtain the time difference;

Step S605, adjusting the target click score according to the time difference.

From step S601 to step S605 shown in the embodiment of the present application, if the target feature encoding data is the same as the current feature encoding data, the prediction model will output a higher target click score, then the current click data of the clicked object is obtained, and To recommend the same content to the clicked object, in order to prevent the same target object from receiving the same recommended content all the time, this embodiment of the present application judges whether the target feature code data is the same as the current feature code data. If they are different, there is no need to adjust the target click score, and the content recommendation is performed directly according to the target click score. If they are the same, the target click score needs to be decreased so that other content with higher target click score can be recommended to the target object. More specifically, in the embodiment of the present application, the target click score is adjusted through the time difference between the display time and the current click time. Both the display time and the current click time can be obtained from the system timer of the client. The current click time refers to the corresponding system time when the click object is clicked. The display time refers to the corresponding system time when the target display content is recommended to the target object for display. The adjustment coefficient can be calculated according to the time difference, and the adjusted target click score can be obtained according to the product of the adjustment coefficient and the target click score. The larger the time difference, the larger the adjustment coefficient, and the range of the adjustment coefficient is (0,1). It should be noted that if the time difference is larger, it means that the display time is farther away from the current click time. At this time, the larger the adjustment coefficient, the smaller the decrease in the target click score, even if the same display content is repeatedly recommended to the same There is no major problem with a target audience.

In step S109 of some embodiments, the target display content is recommended to the target object according to the target click score and the target display position. In an example, step S109 specifically includes: if the target click score is greater than the preset click score threshold, obtaining the browsing page of the target object; if the target click score is less than or equal to the click score threshold, reducing the target display position and obtaining the target The browse page of the object; displaying the target display content on the target display position of the browse page.

It should be noted that if a certain target object is a new user, the knowledge of the user is too little, and the preferences of the target object may not be accurately captured when recommending content, resulting in a decrease in the accuracy of content recommendation, and thus the accuracy of content recommendation. The click-through rate drops, which is not conducive to maintaining the new user. Therefore, the embodiment of the present application proposes to read the object encoding file through the current object information of the clicked object, so as to obtain the target object features associated with the current object information. The target object feature can represent the clicked object or the non-clicked object . In the case that the feature of the target object represents a non-click object, the non-click object may be another user who has similar object information to the click object, and in this case, the accuracy of content recommendation is relatively high. If the user is a new user, the content recommendation for the new user can be implemented, which helps to maintain the new user.

Please refer to FIG. 7. The embodiment of the present application also provides a content recommendation device based on feature engineering, which can implement the above-mentioned content recommendation method based on feature engineering. FIG. 7 is a block diagram of the module structure of the content recommendation device based on feature engineering provided by the embodiment of the present application. , the device includes: a click data acquisition module 701, a configuration information acquisition module 702, a location feature determination module 703, an object feature determination module 704, a content feature determination module 705, a feature extraction module 706, a feature decoding module 707, a model prediction module 708 and Content recommendation module 709 . Wherein, the click data acquiring module 701 is used to acquire the current click data, and the current click data includes the current object information of the clicked object, the current click content information, and the current click position information; the configuration information acquisition module 702 is used to read Feature engineering configuration information and feature extraction information; the position feature determination module 703 is used to perform feature encoding processing on the current click position information according to the feature engineering configuration information to obtain the current position feature; the object feature determination module 704 is used to select from presets according to the current object information The object encoding file reads the target object feature; the content feature determination module 705 is used to read the target content feature from the preset content encoding file according to the content information of the current click; the feature extraction module 706 is used to extract the current position feature, Perform feature extraction on the target object features and target content features to obtain target feature encoding data; the feature decoding module 707 is used to decode the target feature encoding data to obtain the target object, target display content and target display position; wherein, the target object includes click object, or a non-click object; the model prediction module 708 is used to perform click prediction on the target feature encoding data through a preset prediction model to obtain a target click score; the target click score is used to represent: the target display content at the target display position is targeted The probability of an object being clicked; the content recommendation module 709 is configured to recommend the target display content to the target object according to the target click score and the target display position.

It should be noted that the specific implementation of the device for recommending content based on feature engineering is basically the same as the specific embodiment of the method for recommending content based on feature engineering, and will not be repeated here.

The embodiment of the present application also provides an electronic device, the electronic device includes: a memory, a processor, a program stored on the memory and operable on the processor, and a data bus for realizing connection and communication between the processor and the memory, and the program When executed by the processor, the above webpage performance evaluation method is realized. The electronic device may be any intelligent terminal including a tablet computer, a vehicle-mounted computer, and the like.

Please refer to FIG. 8. FIG. 8 illustrates a hardware structure of an electronic device in another embodiment. The electronic device includes:

The processor 801 can be implemented by a general-purpose CPU (Central Processing Unit, central processing unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute Relevant programs to realize the technical solutions provided by the embodiments of the present application;

The memory 802 may be implemented in the form of a read-only memory (Read Only Memory, ROM), a static storage device, a dynamic storage device, or a random access memory (Random Access Memory, RAM). The memory 802 can store operating systems and other application programs. When implementing the technical solutions provided by the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 802 and called by the processor 801 to execute the implementation of this application. An example content recommendation method based on feature engineering;

The input/output interface 803 is used to realize information input and output;

The communication interface 804 is used to realize the communication interaction between the device and other devices, and the communication can be realized through a wired method (such as USB, network cable, etc.), or can be realized through a wireless method (such as a mobile network, WIFI, Bluetooth, etc.);

A bus 805, which transmits information between various components of the device (such as a processor 801, a memory 802, an input/output interface 803, and a communication interface 804);

The processor 801 , the memory 802 , the input/output interface 803 and the communication interface 804 are connected to each other within the device through the bus 805 .

The embodiment of the present application also provides a storage medium, the storage medium is a computer-readable storage medium for computer-readable storage, the storage medium stores one or more programs, and one or more programs can be executed by one or more processors , to implement the above-mentioned content recommendation method based on feature engineering.

As a non-transitory computer-readable storage medium, memory can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory optionally includes memory located remotely from the processor, and these remote memories may be connected to the processor via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The feature engineering-based content recommendation method, feature engineering-based content recommendation device, electronic equipment, and storage medium provided in the embodiments of the present application perform feature encoding and feature extraction by acquiring multi-dimensional information such as object information, content information, and location information to obtain Target feature coding data, the target feature coding data includes not only high real-time location features, but also relatively low real-time object features and content features, so that when the prediction model performs click prediction on the target feature coding data, It can not only combine high real-time features, but also take into account relatively low real-time features, so that the prediction result (target click score) is more accurate, so that the accuracy of content recommendation based on the target click score is higher. In addition, in the content recommendation in the embodiment of the present application, not only the target click score but also the target display position are considered, which helps to improve the click rate of the target object on the target display content.

The embodiments described in the embodiments of the present application are to illustrate the technical solutions of the embodiments of the present application more clearly, and do not constitute a limitation to the technical solutions provided by the embodiments of the present application. Those skilled in the art know that with the evolution of technology and new For the emergence of application scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

Those skilled in the art can understand that the technical solutions shown in Figures 1-6 do not constitute a limitation to the embodiments of the present application, and may include more or fewer steps than those shown in the illustrations, or combine certain steps, or be different A step of.

The device embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, the functional modules/units in the system, and the device can be implemented as software, firmware, hardware, and an appropriate combination thereof.

The terms "first", "second", "third", "fourth", etc. (if any) in the description of the present application and the above drawings are used to distinguish similar objects and not necessarily to describe specific sequence or sequence. It is to be understood that the data so used are interchangeable under appropriate circumstances such that the embodiments of the application described herein can be practiced in sequences other than those illustrated or described herein. Furthermore, the terms "comprising" and "having", as well as any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product or device comprising a sequence of steps or elements is not necessarily limited to the expressly listed instead, may include other steps or elements not explicitly listed or inherent to the process, method, product or apparatus.

It should be understood that in this application, "at least one (item)" means one or more, and "multiple" means two or more. "And/or" is used to describe the association relationship of associated objects, indicating that there can be three types of relationships, for example, "A and/or B" can mean: only A exists, only B exists, and A and B exist at the same time , where A and B can be singular or plural. The character "/" generally indicates that the contextual objects are an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one item (piece) of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c ", where a, b, c can be single or multiple.

In the several embodiments provided in this application, it should be understood that the disclosed devices and methods may be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or May be integrated into another system, or some features may be ignored, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.

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 may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including multiple instructions to make an electronic device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), magnetic disk or optical disc, etc., which can store programs. medium.

The preferred embodiments of the embodiments of the present application have been described above with reference to the accompanying drawings, which does not limit the scope of rights of the embodiments of the present application. Any modifications, equivalent replacements and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall fall within the scope of rights of the embodiments of the present application.

Claims (10)

1. A content recommendation method based on feature engineering, characterized in that the method comprises: Acquiring current click data, the current click data includes current object information of the clicked object, current click content information, and current click position information; Read feature engineering configuration information and feature extraction information from a preset configuration file; performing feature encoding processing on the current click position information according to the feature engineering configuration information to obtain the current position feature; Reading target object characteristics from a preset object encoding file according to the current object information; Reading target content characteristics from a preset content encoding file according to the currently clicked content information; performing feature extraction on the current location feature, the target object feature, and the target content feature according to the feature extraction information to obtain target feature encoding data; Decoding the target feature encoding data to obtain a target object, target display content, and target display position; wherein, the target object includes the click object, or is not the click object; Click prediction is performed on the target feature coding data through a preset prediction model to obtain a target click score; the target click score is used to represent: the target display content at the target display position is clicked by the target object probability; Recommending the target display content to the target object according to the target click score and the target display position. 2. The method according to claim 1, wherein the current click data further includes: the current click time, when the preset prediction model is used to perform click prediction on the target feature coding data, the target click is obtained After scoring, the method also includes: performing feature encoding processing on the current object information according to the feature engineering configuration information to obtain the current object features; performing feature encoding processing on the currently clicked content information according to the feature engineering configuration information to obtain current content features; performing feature extraction on the current object feature, the current content feature, and the current location feature according to the feature extraction information to obtain current feature encoding data; If the target characteristic encoding data is the same as the current characteristic encoding data, then obtain the display time of the target characteristic encoding data; calculating the difference between the current click time and the display time to obtain a time difference; The target click score is adjusted according to the time difference. 3. The method according to any one of claims 1 to 2, characterized in that, before the click prediction is performed on the target feature encoding data through a preset prediction model to obtain the target click score, the method further include: Pre-training the prediction model specifically includes: Obtain historical display data, and obtain the original label of the historical display data; the historical display data includes historical object information, historical display content information, and historical display position information of historical objects; performing feature encoding processing on the historical object information according to the feature engineering configuration information to obtain historical object features; performing feature encoding processing on the historical display content information according to the feature engineering configuration information to obtain historical content features; performing feature encoding processing on the historical display location information according to the feature engineering configuration information to obtain historical location features; performing feature extraction on the historical object features, historical content features and the historical location features according to the feature extraction information to obtain training feature encoding data; determining a raw training score for the training feature-encoded data based on the raw label; performing score prediction on the training feature encoding data through the prediction model to obtain a target training score; performing loss calculation according to the original training score and the target training score to obtain a loss value; The parameters of the prediction model are adjusted according to the loss value to obtain the trained prediction model. 4. The method according to claim 3, wherein said obtaining historical display data and obtaining the original label of said historical display data comprises: Obtaining an object behavior log of the historical object; Reading information from the object behavior log to obtain the historical object information, the historical display content information, and the historical display position information; combining the historical object information, the historical display content information and the historical display position information to obtain the historical display data; Read the time of the object behavior log to obtain the content marking time of the historical display data, the content marking time includes: content exposure time or content click time; The original tag is determined according to the tagging time of the content. 5. The method according to claim 4, wherein, before the object behavior log is read, the method further comprises: Update the object behavior log, specifically including: acquiring object data and an object identification code, the object data being used to provide the historical object information; Obtaining content data and content identification codes and using the content data to provide the historical content information; Recording the object identification code and the content identification code in the object behavior log; The object data is associated with the object behavior log through the object identification code, and the content data is associated with the object behavior log through the content identification code. 6. The method according to claim 5, wherein the object data includes object basic information and object behavior information, the content data includes content basic information and content statistical information, and the object identification code will The object data is associated with the object behavior log, and the content data is associated with the object behavior log through the content identification code, including: Establishing a mapping relationship between the object basic information and the object behavior information to obtain an object wide table; Establishing a mapping relationship between the content basic information and the content statistical information to obtain a wide content table; The object wide table is associated with the object behavior log through the object identification code, and the content wide table is associated with the object behavior log through the content identification code. 7. The method according to any one of claims 1 to 2, wherein said reading feature engineering configuration information and feature extraction information from a preset configuration file includes: Get the current feature configuration version number; Reading the feature engineering configuration information and the feature extraction information from the configuration file according to the current feature configuration version number. 8. A content recommendation device based on feature engineering, characterized in that the device comprises: The click data acquisition module is used to acquire the current click data, the current click data includes the current object information of the click object, the current click content information, and the current click position information; The configuration information acquisition module is used to read feature engineering configuration information and feature extraction information from a preset configuration file; A position feature determination module, configured to perform feature encoding processing on the current click position information according to the feature engineering configuration information to obtain the current position feature; An object feature determination module, configured to read target object features from a preset object encoding file according to the current object information; A content feature determination module, configured to read target content features from a preset content encoding file according to the currently clicked content information; A feature extraction module, configured to perform feature extraction on the current location feature, the target object feature, and the target content feature according to the feature extraction information, to obtain target feature encoding data; A feature decoding module, configured to decode the target feature encoding data to obtain a target object, target display content, and target display position; wherein, the target object includes the click object, or is not the click object; A model prediction module, configured to perform click prediction on the target feature encoding data through a preset prediction model to obtain a target click score; the target click score is used to represent: the target display content at the target display position is The probability of the target object being clicked; A content recommendation module, configured to recommend the target display content to the target object according to the target click score and the target display position. 9. An electronic device, characterized in that, the electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements any one of claims 1 to 7 when executing the computer program. A content recommendation method based on feature engineering. 10. A computer-readable storage medium, the storage medium stores a computer program, characterized in that, when the computer program is executed by a processor, the content based on feature engineering according to any one of claims 1 to 7 is realized recommended method.

Images