Content recommendation method and device based on feature engineering, electronic device, and medium

Abstract

The content recommendation method and device based on feature engineering, the electronic equipment and the medium provided in the application relate to the technical field of artificial intelligence. The method comprises: performing feature encoding processing on current click position information according to feature engineering configuration information to obtain current position features; reading target object features from a preset object encoding file according to current object information; reading target content features from a preset content encoding file according to current click content information; performing feature extraction according to feature extraction information to obtain target feature encoding data; and performing click prediction on the target feature encoding data through a preset prediction model to obtain a target click score. When the prediction model performs click prediction on the target feature encoding data, the position features, the object features and the content features are combined, so that the target click score obtained by prediction is more accurate, and the accuracy of content recommendation according to the target click score is higher.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a content recommendation method and apparatus, electronic device, and medium based on feature engineering. Background Technology

[0002] Feature engineering is a process of transforming raw data into features that better represent business logic. Currently, content recommendation based on feature engineering generally involves: acquiring raw data, encoding features into the raw data, using an offline-developed prediction model to predict clicks on the encoded data, obtaining prediction results, and then recommending content based on these results. In online content recommendation, online data has high real-time performance, but offline prediction models are trained on offline data, which has lower real-time performance. This inconsistency in real-time performance between online and offline data affects the accuracy of recommended content when offline prediction models are deployed online. Therefore, providing a feature engineering-based content recommendation method that can improve the accuracy of recommended content has become a pressing technical problem. Summary of the Invention

[0003] The main objective of this application is to propose a content recommendation method, apparatus, electronic device, and medium based on feature engineering, which can improve the accuracy of recommended content.

[0004] To achieve the above objectives, a first aspect of this application proposes a content recommendation method based on feature engineering, the method comprising:

[0005] Obtain the current click data, which includes the current object information of the clicked object, the current click content information, and the current click position information;

[0006] Read feature engineering configuration information and feature extraction information from the preset configuration file;

[0007] The current click location information is subjected to feature encoding processing based on the feature engineering configuration information to obtain the current location feature;

[0008] The target object features are read from a preset object encoding file based on the current object information;

[0009] Based on the currently clicked content information, the target content features are read from a preset content encoding file;

[0010] Based on the feature extraction information, feature extraction is performed on the current location feature, the target object feature, and the target content feature to obtain target feature encoded data;

[0011] The target feature encoding data is decoded to obtain the target object, the target display content, and the target display position; wherein, the target object includes the clicked object, or a non-clicked object;

[0012] Click prediction is performed on the target feature encoding data using a preset prediction model to obtain a target click score; the target click score is used to characterize the probability that the target display content located at the target display position is clicked by the target object;

[0013] The target content is recommended to the target object based on the target click score and the target display position.

[0014] In some embodiments, the current click data further includes: the current click time; after performing click prediction on the target feature encoding data using a preset prediction model to obtain the target click score, the method further includes:

[0015] The current object information is subjected to feature encoding processing based on the feature engineering configuration information to obtain the current object features;

[0016] Based on the feature engineering configuration information, the currently clicked content information is subjected to feature encoding processing to obtain the current content features;

[0017] Based on the feature extraction information, feature extraction is performed on the current object features, the current content features, and the current position features to obtain current feature encoding data;

[0018] If the target feature encoding data is the same as the current feature encoding data, then obtain the display time of the target feature encoding data;

[0019] If the target feature encoding data is the same as the current feature encoding data, then obtain the display time of the target feature encoding data;

[0020] The time difference is calculated by comparing the current click time with the display time.

[0021] The target click score is adjusted based on the time difference.

[0022] In some embodiments, before performing click prediction on the target feature encoding data using a preset prediction model to obtain the target click score, the method further includes:

[0023] Pre-training the prediction model specifically includes:

[0024] Obtain historical display data and its original tags; the historical display data includes historical object information, historical display content information, and historical display location information of the historical object.

[0025] Based on the feature engineering configuration information, the historical object information is subjected to feature encoding processing to obtain historical object features;

[0026] Based on the feature engineering configuration information, the historical display content information is subjected to feature encoding processing to obtain historical content features;

[0027] Based on the feature engineering configuration information, the historical display location information is subjected to feature encoding processing to obtain historical location features;

[0028] Based on the feature extraction information, feature extraction is performed on the historical object features, historical content features, and historical location features to obtain training feature encoding data;

[0029] Based on the original labels, determine the original training score of the training feature encoding data;

[0030] The target training score is obtained by predicting the score of the training feature-encoded data using the prediction model.

[0031] The loss value is obtained by calculating the loss based on the original training score and the target training score;

[0032] The parameters of the prediction model are adjusted based on the loss value to obtain the trained prediction model.

[0033] In some embodiments, obtaining historical display data and obtaining the original tags of the historical display data includes:

[0034] Obtain the object behavior log of the historical object;

[0035] The object behavior log is read to obtain the historical object information, the historical display content information, and the historical display location information;

[0036] The historical object information, the historical display content information, and the historical display location information are merged to obtain the historical display data;

[0037] The object behavior log is read by time to obtain the content annotation time of the historical display data. The content annotation time includes: content exposure time or content click time.

[0038] The original tag is determined based on the time indicated by the content.

[0039] In some embodiments, the method further includes, before reading the object behavior log:

[0040] Updating the object behavior log specifically includes:

[0041] Obtain object data and object identifier code, wherein the object data is used to provide the historical object information;

[0042] The content data and content identifier code are obtained, and the content data is used to provide the historical content information;

[0043] Record the object identifier and the content identifier to the object behavior log;

[0044] The object data is associated with the object behavior log through the object identifier code, and the content data is associated with the object behavior log through the content identifier code.

[0045] In some embodiments, the object data includes basic object information and object behavior information, and the content data includes basic content information and content statistics. The step of associating the object data with the object behavior log using the object identifier and the content data with the object behavior log using the content identifier includes:

[0046] A mapping relationship is established between the object's basic information and the object's behavioral information to obtain an object wide table;

[0047] A mapping relationship is established between the basic content information and the content statistics information to obtain a wide content table;

[0048] The object wide table is associated with the object behavior log through the object identifier code, and the content wide table is associated with the object behavior log through the content identifier code.

[0049] In some embodiments, reading feature engineering configuration information and feature extraction information from a preset configuration file includes:

[0050] Get the current feature configuration version number;

[0051] Based on the current feature configuration version number, the feature engineering configuration information and the feature extraction information are read from the configuration file.

[0052] To achieve the above objectives, a second aspect of this application proposes a content recommendation apparatus based on feature engineering, the apparatus comprising:

[0053] The click data acquisition module is used to acquire the current click data, which includes the current object information of the clicked object, the current click content information, and the current click position information.

[0054] The configuration information acquisition module is used to read feature engineering configuration information and feature extraction information from a preset configuration file;

[0055] The location feature determination module is used to perform feature encoding processing on the current click location information according to the feature engineering configuration information to obtain the current location feature;

[0056] An object feature determination module is used to read target object features from a preset object encoding file based on the current object information;

[0057] The content feature determination module is used to read target content features from a preset content encoding file based on the currently clicked content information;

[0058] The feature extraction module is used to extract features from the current location features, the target object features, and the target content features based on the feature extraction information to obtain target feature encoding data;

[0059] The feature decoding module is used to decode the target feature encoding data to obtain the target object, the target display content, and the target display position; wherein, the target object includes the clicked object or a non-clicked object;

[0060] The model prediction module is used to predict clicks on the target feature encoding data using a preset prediction model to obtain a target click score; the target click score is used to characterize the probability that the target display content located at the target display position is clicked by the target object;

[0061] The content recommendation module is used to recommend the content displayed for the target to the target object based on the target click score and the target display position.

[0062] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0063] To achieve the above objectives, a fourth aspect of the present application provides a storage medium, which is a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0064] The content recommendation method, apparatus, electronic device, and medium proposed in this application based on feature engineering 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. This target feature encoding data includes location features, object features, and content features. In this way, when the prediction model performs click prediction on the target feature encoding data, the feature dimension is increased, combining location features, object features, and content features, making the predicted target click score more accurate, and thus the accuracy of content recommendation based on the target click score is higher. Attached Figure Description

[0065] Figure 1 This is a flowchart of the content recommendation method based on feature engineering provided in the embodiments of this application;

[0066] Figure 2 This is a flowchart of a content recommendation method based on feature engineering provided in another embodiment of this application;

[0067] Figure 3 yes Figure 2 The flowchart of step S201 in the text;

[0068] Figure 4 This is a flowchart of a content recommendation method based on feature engineering provided in another embodiment of this application;

[0069] Figure 5 yes Figure 4 The flowchart of step S404 in the document;

[0070] Figure 6 This is a flowchart of a content recommendation method based on feature engineering provided in another embodiment of this application;

[0071] Figure 7 This is a block diagram of the module structure of the content recommendation device based on feature engineering provided in the embodiments of this application;

[0072] Figure 8 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0074] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0075] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0076] First, let's analyze some of the terms used in this application:

[0077] Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0078] Feature engineering is the process of transforming raw data into features that better represent business logic. Applying these features to predictive models improves the accuracy of predictions on unseen data and enhances machine learning performance. For example, in fields like search, advertising, and recommendation, data determines the upper limit of performance, while algorithms can only determine the degree to which they approximate that limit. Feature engineering acts as a bridge between data and algorithms. Feature engineering has a significant impact on business performance. This is primarily because it can significantly improve model performance, greatly simplify model complexity, reduce model maintenance costs, and high-quality features can demonstrate good results even on simple linear models.

[0079] Currently, content recommendation based on feature engineering generally involves: acquiring online data, encoding features into the online data, using an offline-developed prediction model to predict clicks on the encoded data, obtaining the prediction results, and then recommending content based on these results. In online content recommendation, online data has high real-time performance, but offline prediction models are trained on offline data, which has lower real-time performance. This inconsistency in real-time performance between online and offline data affects the accuracy of recommended content when offline prediction models are deployed online. Therefore, providing a feature engineering-based content recommendation method that can improve the accuracy of recommended content has become an urgent technical problem to be solved.

[0080] Based on this, the main objective of this application is to propose a content recommendation method, device, electronic device, and storage medium based on feature engineering. By acquiring multi-dimensional information such as object information, content information, and location information, feature extraction is performed to obtain target feature encoding data. This target feature encoding data includes both highly real-time location features and relatively less real-time object and content features. In this way, when the prediction model predicts based on the target feature encoding data, it can combine highly real-time online features with relatively less real-time offline features, making the prediction result (target click score) more accurate, thereby increasing the accuracy of content recommendation based on the target click score. Furthermore, in the content recommendation of this application, not only the target click score but also the target display location is considered, which helps to improve the click-through rate of the target object on the displayed content.

[0081] The feature-engineered content recommendation method provided in this application can be applied to the server side, or it can be software running on a terminal or server side. The server side can be configured as an independent physical server, or as a server cluster or distributed system composed of multiple physical servers, or as a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the feature-engineered content recommendation method, etc., but is not limited to the above forms.

[0082] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include server computers, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0083] This application provides a content recommendation method, device, electronic device, and storage medium based on feature engineering, which are specifically described through the following embodiments. First, the content recommendation method based on feature engineering in this application is described.

[0084] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards of the relevant countries and regions. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirects to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data for the proper functioning of the embodiments of this application obtained.

[0085] Figure 1 This is an optional flowchart of the content recommendation method based on feature engineering provided in the embodiments of this application, which may include, but is not limited to, steps S101 to S109.

[0086] Step S101: Obtain the current click data, which includes the current object information of the clicked object, the current click content information, and the current click position information;

[0087] Step S102: Read feature engineering configuration information and feature extraction information from the preset configuration file;

[0088] Step S103: Perform feature encoding processing on the current click location information according to the feature engineering configuration information to obtain the current location feature;

[0089] Step S104: Read the target object features from the preset object encoding file based on the current object information;

[0090] Step S105: Read the target content features from the preset content encoding file based on the currently clicked content information;

[0091] Step S106: Based on the feature extraction information, feature extraction is performed on the current location features, target object features, and target content features to obtain target feature encoding data;

[0092] Step S107: Decode the target feature encoding data to obtain the target object, target display content, and target display position; wherein, the target object includes clicked objects or non-clicked objects;

[0093] Step S108: Click prediction is performed on the target feature encoding data using a preset prediction model to obtain the target click score; the target click score is used to characterize the probability that the target display content at the target display position is clicked by the target object;

[0094] Step S109: Recommend the target content to the target audience based on the target click score and the target display position.

[0095] Steps S101 to S109 of this embodiment involve acquiring multi-dimensional information such as object information, content information, and location information, performing feature encoding and feature extraction to obtain target feature encoded data. This target feature encoded data includes location features, as well as object features and content features. Since location features are obtained by clicking on the object, they have high real-time performance, while object features and content features are obtained from the encoded file, resulting in relatively lower real-time performance. In this way, when the prediction model performs click prediction on the target feature encoded data, it can combine features with high real-time performance while also considering features with relatively lower real-time performance, making the prediction result (target click score) more accurate. This leads to a higher accuracy rate for content recommendation based on the target click score. Furthermore, in the content recommendation of this embodiment, not only the target click score but also the target display position is considered, which helps to improve the click-through rate of the target object on the target displayed content.

[0096] In step S101 of some embodiments, current click data can be obtained from the client. Taking video recommendation as an example, the clicked video information is recorded in the click log. Therefore, in one example, the client's click log is obtained, and the current object information, current click content information, and current click location information of the clicked object are read from the click log. The current object information, current click content information, and current click location information are then merged to obtain the current click data.

[0097] In another example, the client's current browsing page is obtained; the object identifier is obtained based on the login account of the current browsing page; the current object information is read from the preset object database based on the object identifier; a click signal is received; in response to the click signal, the current click content information and the current click location information are obtained from the browsing page; the current object information, the current click content information, and the current click location information are merged to obtain the current click data.

[0098] It should be noted that the embodiments of this application mention clicked object, historical object, and target object. All three objects refer to users, but they may be different users. For example, client 1 corresponds to object A, and client 2 corresponds to object B. If object A is clicked in real time on the browsing page of client 1, then object A is the clicked object, and the embodiments of this application can obtain the current click data of object A. Since the embodiments of this application read the target object features from a preset object encoding file through the current object information, and the target object features may be the object features of object A or object B, the target object obtained based on the target object features may be the clicked object or other objects that are not clicked. Historical objects, on the other hand, appear in the pre-training model stage of this application; regardless of whether object A or object B clicks on the browsing page, they can be historical objects.

[0099] Both current object information and historical object information refer to object data. Object data generally includes basic object information and object behavior information. Basic object information includes gender, age, current city, etc. Object behavior information includes the list of clicked news items, the list of saved news items, the list of liked news items, etc.

[0100] Both currently clicked content information and historically displayed content information refer to content data, which includes basic content information and content statistics. Basic content information includes first / secondary categories, publisher, keywords, etc. Content statistics include the number of favorites, likes, comments, etc.

[0101] The current click location information refers to the position of the content on the browsing page when the clicked object clicks on it. Taking video recommendations as an example, the client displays multiple videos simultaneously on the browsing page, each with a corresponding display position. The browsing page changes the displayed videos based on the object's actions; that is, the video's display position changes accordingly. When a click signal is received, the system responds by determining the current click position based on the current display position of the clicked video.

[0102] In step S102 of some embodiments, feature engineering configuration information and feature extraction information are set in advance in the configuration file. The feature engineering configuration information mainly performs feature encoding on the data. For example, feature encoding processing is performed on the current click location information according to the feature engineering configuration information to obtain the current location feature.

[0103] In one example, step S102 may include: obtaining the current feature configuration version number; and reading feature engineering configuration information and feature extraction information from the configuration file based on the current feature configuration version number.

[0104] Specifically, the feature configuration file contains configurable feature engineering and feature extraction information, and the same configuration file is used in both offline development and online real-time recommendation phases. This solves the problem of frequent project code updates and long release cycles, effectively improving feature iteration efficiency. Furthermore, different feature iteration experiments can use configurable feature versions, ensuring feature version control.

[0105] It should be noted that the configuration file in this application embodiment is the same in both the offline development stage and the online real-time recommendation stage. Since offline data and online data differ, the feature engineering configuration information obtained from the configuration file may not be able to process online data. For example, in a client's browsing page, if browsing page M displays three videos at a time, the video display positions can be divided into top, middle, and bottom. If the clicked object clicks on the video at the "top" position, the current position information will be "top". If browsing page N displays five videos at a time, the video display positions can be divided into top, upper middle, middle, lower middle, and bottom. If the clicked object clicks on the video at the "upper middle" position, the current position information will be "upper middle". It can be seen that different browsing pages may yield different position information, and the position information of browsing page M and browsing page N cannot be processed equally.

[0106] Therefore, in one example, before step S103, mapping configuration information is obtained, and the format of the current location information is adjusted using the mapping configuration information to obtain the adjusted current location information. Then, feature encoding is performed on the adjusted current location information according to the feature engineering configuration information to obtain the current location feature. Specifically, after the format adjustment, "upper middle" on the browsing page can correspond to either "upper" or "middle".

[0107] In another example, the result of format adjustment is checked. If the result indicates that the format cannot be adjusted, the historical location information of the last click is retrieved. Then, feature encoding is performed on the historical location information according to the feature engineering configuration information to obtain the current location feature. Additionally, if the result indicates that the format cannot be adjusted, an alert message is sent to the server. This alert message provides feedback on the inability to properly encode the feature engineering configuration information, allowing developers to update or optimize the configuration file.

[0108] In step S103 of some embodiments, the current location information is feature-encoded according to the feature engineering configuration information to obtain the current location features. The feature engineering configuration information includes matching information between the information to be processed and the feature processing method. For example, the 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 one 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 feature-processed using the location feature processing method to obtain the current location features.

[0109] In step S104 of some embodiments, target object features are read from a preset object encoding file based on the current object information. Specifically, the object encoding file includes multiple candidate object features such as gender features, age features, education level features, and information click sequence features. The current object information includes gender information, age information, and education level information, etc. Multiple candidate object features are filtered based on 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, then the gender feature is read from the object encoding file based on the gender information, and the age feature is read from the object encoding file based on the age information. The gender feature and the age feature are then concatenated to obtain the target object features.

[0110] In step S105 of some embodiments, target content features are read from a preset content encoding file based on the currently clicked content information. Specifically, the content encoding file includes multiple candidate content features such as first / secondary category features, keyword features, click volume features, collection volume features, like volume features, and comment volume features. The current content information includes first / secondary category information, keyword information, click volume information, collection volume information, like volume information, and comment volume information. Multiple candidate content features are filtered based on the current content information to obtain target content features; wherein, the target content features include at least one candidate content feature. For example, if the current content information includes keyword information and comment volume information, then keyword features are read from the content encoding file based on the keyword information, and comment volume features are read from the content encoding file based on the comment volume information. The keyword features and comment volume features are then concatenated to obtain the target content features.

[0111] In step S106 of some embodiments, feature extraction is performed on the current location feature, target object feature, and target content feature based on feature extraction information to obtain target feature encoding data. Specifically, the feature extraction information mainly involves extracting features from multiple features; this feature extraction is also known as feature merging or feature concatenation. For example, feature concatenation is performed on the current location feature, target object feature, and target content feature based on feature extraction information to obtain target feature encoding data.

[0112] In one example, before feature concatenation, the current location feature, target object feature, and target content feature can be processed to ensure that they all have the same feature dimension.

[0113] 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, then discretizing the target content feature to obtain the target discrete content feature; merging the current position feature, the target object feature and the target discrete content feature to obtain the target feature encoding data.

[0114] For example, the number of likes in the target content features is a continuous feature. The model needs to discretize the number of likes feature to get better results. Therefore, the number of likes feature needs to be discretized, such as log(number of likes feature * 10 + 1).

[0115] In step S107 of some embodiments, the target feature encoding data is decoded to obtain the target object, target display content, and target display position; wherein, the target object includes a clickable object or a non-clickable object. Specifically, the target feature encoding data is feature-separated to obtain the object feature to be decoded, the content feature to be decoded, and the position feature to be decoded; the object feature to be decoded, the content feature to be decoded, and the position feature to be decoded are decoded respectively to obtain the target object, target display content, and target display position. It should be noted that the object feature to be decoded is not exactly the same as the target object feature, the content feature to be decoded is not exactly the same as the target content feature, and the position feature to be decoded is not exactly the same as the current position feature. Since there is no direct relationship between the current position feature, the target content feature, and the target object feature, if feature extraction is not performed through feature extraction information, but feature decoding and candidate step model prediction are performed directly, it may affect the accuracy of the prediction model in click prediction.

[0116] In one example, the features of the object to be decoded are decoded to obtain the target object, specifically including:

[0117] The features of the object to be decoded are decoded to obtain the target object information; the target object information includes at least one of the following: gender information, age information, and education information;

[0118] Target objects are selected from a preset set of objects based on the target object information.

[0119] Specifically, if the target object information is gender, then target objects with the same gender information are selected from the object set. If the target object information is age, then an age range is determined based on the age information, and target objects that match the age range are selected from the object set.

[0120] In another example, the features of the content to be decoded are decoded to obtain the target display content, specifically including:

[0121] The features of the content to be decoded are decoded to obtain target content information; the target content information includes at least one of the following: keyword information, likes information;

[0122] The target content is selected from a preset content set based on the target content information.

[0123] Specifically, if the target content information includes keyword information, then target display content that covers the keyword information is selected from the content set. If the target content information includes likes information, then a likes threshold is determined based on the likes information, and target display content with likes exceeding the likes threshold is selected from the content set.

[0124] It should be noted that the prediction model in the embodiments of this application can be a logistic regression model (LR) or an online learning model (Follow the Regularized Leader, FTRL).

[0125] Please see Figure 2 In some other embodiments, prior to step S108, the content recommendation method based on feature engineering further includes: a pre-trained prediction model, specifically including but not limited to steps S201 to S207:

[0126] Step S201: Obtain historical display data and its original tags; historical display data includes historical object information, historical display content information, and historical display location information of historical objects.

[0127] 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; perform feature encoding processing on historical display location information according to feature engineering configuration information to obtain historical location features;

[0128] Step S203: Based on the feature extraction information, feature extraction is performed on historical object features, historical content features, and historical location features to obtain training feature encoding data;

[0129] Step S204: Determine the original training score of the training feature encoding data based on the original labels;

[0130] Step S205: The target training score is obtained by predicting the score of the training feature encoding data through the prediction model.

[0131] Step S206: Calculate the loss based on the original training score and the target training score to obtain the loss value;

[0132] Step S207: Adjust the parameters of the prediction model based on the loss value to obtain the trained prediction model.

[0133] Steps S201 to S207, as illustrated in this embodiment, involve feature encoding of historical display data using feature engineering configuration information to obtain offline features. These offline features include historical object features, historical content features, and historical location features. The offline features are then input into a prediction model for click prediction to obtain a target training score. The target training score represents the predicted probability that historical content at a historical location is clicked by a historical object. The original label indicates whether the historical content was clicked by a historical object. If yes, the original label is positive; otherwise, it is negative. A positive original label results in a higher original training score, and a negative original label results in a lower original training score. A loss value is determined based on the original training score and the target training score. The parameters of the prediction model are then adjusted based on the loss value to obtain the trained prediction model. In one example, the difference between the original training score and the target training score is calculated to obtain the loss value.

[0134] In one example, the original label is the content annotation time, which includes either the content exposure time or the content click time. If the content annotation time is the content exposure time, the preset exposure score is used as the original training score; if the content annotation time is the content click time, the difference between the content exposure time and the content click time is calculated to obtain a time coefficient. The original training score is then calculated by multiplying the time coefficient and the exposure score. The larger the difference between the content exposure time and the content click time, the smaller the time coefficient. However, the time coefficient must be at least greater than a preset coefficient threshold, which has a minimum value of 1.

[0135] Please see Figure 3 In other embodiments, step S201 may include, but is not limited to, steps S301 to S305:

[0136] Step S301: Obtain the object behavior log of the historical object;

[0137] Step S302: Read information from the object behavior log to obtain historical object information, historical display content information, and historical display location information;

[0138] Step S303: Merge the historical object information, historical display content information, and historical display location information to obtain historical display data;

[0139] Step S304: Read the object behavior log by time to obtain the content annotation time of the historical display data. The content annotation time includes: content exposure time or content click time.

[0140] Step S305: Determine the original tag based on the content annotation time.

[0141] In steps S301 to S305 of this embodiment, the object behavior log stores historical object information, historical display content information, historical display location information, and content annotation time. By reading the object behavior log, historical display data and original tags can be quickly obtained.

[0142] It should be noted that, in addition to object behavior logs, historical object information can be stored in an HBase database, historical display content information can be stored in a Redis database, historical location information can be stored in a Redis database, and content annotation time can be stored in a 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 database through a data interface.

[0143] Please see Figure 4 In some other embodiments, prior to 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:

[0144] Step S401: Obtain object data and object identifier code. The object data is used to provide historical object information.

[0145] Step S402: Obtain content data and content identifier code. The content data is used to provide historical content information.

[0146] Step S403: Record the object identifier code and content identifier code to the object behavior log;

[0147] Step S404: Associate object data with object behavior log using object identifier code, and associate content data with object behavior log using content identifier code.

[0148] In steps S401 to S404 of this embodiment, due to the large volume of object data and content data, directly storing the object data and content data in the object behavior log would result in an excessively large object behavior log volume, affecting the efficiency of reading the object behavior log. Therefore, this embodiment sets object identifier codes for object data and content identifier codes for content data. By associating data through identifier codes, the storage volume of the object behavior log is reduced, and the reading efficiency is improved.

[0149] Please see Figure 5 In other embodiments, object data includes object basic information and object behavior information, and content data includes content basic information and content statistics. Step S404 specifically includes, but is not limited to, steps S501 to S503:

[0150] Step S501: Establish a mapping relationship between the object's basic information and the object's behavioral information to obtain the object wide table;

[0151] Step S502: Establish a mapping relationship between basic content information and content statistics information to obtain a wide content table;

[0152] Step S503: Associate the object wide table with the object behavior log using the object identifier code, and associate the content wide table with the object behavior log using the content identifier code.

[0153] Steps S501 to S503, as illustrated in this embodiment, involve user basic information including gender, age, current city, etc., and user behavior information including clicked news lists, favorited or liked news lists, etc. The object wide table refers to linking user basic information and behavior information together to form a complete user profile table. Content basic information includes first / secondary categories, publishers, keywords, etc., and content statistics include the number of favorites, likes, comments, etc. The content wide table refers to linking content basic information and content statistics together to form a complete content profile table. By establishing wide tables for object data and content data, data management and retrieval are facilitated.

[0154] In step S108 of some embodiments, click prediction is performed on the target feature encoding data using a preset prediction model to obtain a target click score; the target click score is used to characterize the probability that the target display content at the target display position is clicked by the target object.

[0155] Please see Figure 6 In other embodiments, the current click data further includes the current click time. After step S108, the content recommendation method based on feature engineering may also include, but is not limited to, steps S601 to S605:

[0156] Step S601: Perform feature encoding processing on the current object information according to the feature engineering configuration information to obtain the current object features; perform feature encoding processing on the currently clicked content information according to the feature engineering configuration information to obtain the current content features;

[0157] Step S602: Based on the feature extraction information, extract features from the current object features, current content features, and current location features to obtain the current feature encoding data;

[0158] Step S603: If the target feature encoding data is the same as the current feature encoding data, then obtain the display time of the target feature encoding data;

[0159] Step S604: Calculate the difference between the current click time and the display time to obtain the time difference;

[0160] Step S605: Adjust the target click score based on the time difference.

[0161] In steps S601 to S605 of this embodiment, 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. This means the current click data of the clicked object is obtained, and the same content is recommended to that clicked object again. To avoid the same target object receiving the same recommended content repeatedly, this embodiment determines whether the target feature encoding data and the current feature encoding data are the same. If they are different, there is no need to adjust the target click score; content recommendation is directly based on the target click score. If they are the same, the target click score needs to be lowered so that content with higher target click scores can be recommended to the target object. More specifically, this embodiment adjusts the target click score based on 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 client's system timer. The current click time refers to the system time obtained when the clicked object clicks. The display time refers to the system time obtained if the target display content is recommended to the target object for display. An adjustment coefficient can be calculated based on the time difference, and the adjusted target click score is obtained by multiplying the adjustment coefficient by the target click score. The larger the time difference, the larger the adjustment coefficient, which ranges from 0 to 1. It's important to note that a larger time difference indicates a greater gap between the display time and the current click time. In this case, a larger adjustment coefficient results in a smaller decrease in the target click score, meaning that even if the same content is repeatedly recommended to the same target audience, there will be little problem.

[0162] In step S109 of some embodiments, target display content is recommended to the target object based on the target click score and the target display position. In one example, step S109 specifically includes: if the target click score is greater than a preset click score threshold, then the browsing page of the target object is obtained; if the target click score is less than or equal to the click score threshold, then the target display position is lowered and the browsing page of the target object is obtained; and the target display content is displayed at the target display position on the browsing page.

[0163] It should be noted that if a target user is new, insufficient information about that user may prevent accurate content recommendations from capturing their preferences, leading to decreased accuracy and consequently lower click-through rates, which is detrimental to retaining the new user. Therefore, this application proposes reading the object encoding file using the current object information of the clicked object to obtain target object features associated with that information. These target object features can represent either the clicked object or a non-clicked object. When the target object features represent a non-clicked object, this non-clicked object could be another user with similar object information to the clicked object, resulting in higher content recommendation accuracy. If the user is new, content recommendations can be tailored to them, helping to retain new users.

[0164] Please see Figure 7 This application also provides a content recommendation apparatus based on feature engineering, which can implement the above-described content recommendation method based on feature engineering. Figure 7The block diagram of the feature engineering-based content recommendation device provided in this application embodiment is shown. 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 a content recommendation module 709. Specifically, the click data acquisition module 701 acquires current click data, which includes current object information, current click content information, and current click location information; the configuration information acquisition module 702 reads feature engineering configuration information and feature extraction information from a preset configuration file; the location feature determination module 703 performs feature encoding processing on the current click location information according to the feature engineering configuration information to obtain current location features; the object feature determination module 704 reads target object features from a preset object encoding file according to the current object information; the content feature determination module 705 reads target content features from a preset content encoding file according to the current click content information; and the feature extraction module 706... 6 is used to extract features from the current location features, target object features, and target content features based on feature extraction information 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 clicked objects or non-clicked objects; the model prediction module 708 is used to perform click prediction on the target feature encoding data through a preset prediction model to obtain the target click score; the target click score is used to characterize the probability that the target display content at the target display position is clicked by the target object; the content recommendation module 709 is used to recommend the target display content to the target object based on the target click score and the target display position.

[0165] It should be noted that the specific implementation of this feature engineering-based content recommendation device is basically the same as the specific implementation of the feature engineering-based content recommendation method described above, and will not be repeated here.

[0166] This application also provides an electronic device, which includes: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for communication between the processor and the memory. When the program is executed by the processor, it implements the aforementioned webpage performance evaluation method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0167] Please see Figure 8 , Figure 8 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0168] The processor 801 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0169] The memory 802 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 802 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 802 and is called and executed by the processor 801 using the feature engineering-based content recommendation method of the embodiments of this application.

[0170] The 803 input / output interface is used to implement information input and output.

[0171] The communication interface 804 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0172] Bus 805 transmits information between various components of the device (e.g., processor 801, memory 802, input / output interface 803, and communication interface 804);

[0173] The processor 801, memory 802, input / output interface 803, and communication interface 804 are connected to each other within the device via bus 805.

[0174] This application embodiment also provides a storage medium, which is a computer-readable storage medium for computer-readable storage. The storage medium stores one or more programs, which can be executed by one or more processors to implement the above-described content recommendation method based on feature engineering.

[0175] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0176] The feature-engineered content recommendation method, device, electronic device, and storage medium provided in this application embodiment acquire multi-dimensional information such as object information, content information, and location information, perform feature encoding and feature extraction, and obtain target feature-encoded data. This target feature-encoded data includes both highly real-time location features and relatively less real-time object and content features. In this way, when the prediction model performs click prediction on the target feature-encoded data, it can combine features with high real-time performance while also considering features with relatively low real-time performance, making the prediction result (target click score) more accurate. Therefore, the accuracy of content recommendation based on the target click score is higher. Furthermore, the content recommendation in this application embodiment considers not only the target click score but also the target display location, which helps to improve the click-through rate of the target object on the target displayed content.

[0177] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0178] It will be understood by those skilled in the art that Figure 1-6 The technical solutions shown do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0179] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0180] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0181] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0182] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) 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) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0183] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0184] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0185] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0186] If the integrated unit is implemented as a software functional 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 this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0187] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A content recommendation method based on feature engineering, characterized in that, The method includes: Obtain current click data, which includes the current object information of the clicked object, the current click content information, the current click location information, and the current click time; Read feature engineering configuration information and feature extraction information from the preset configuration file; The current click location information is subjected to feature encoding processing based on the feature engineering configuration information to obtain the current location feature; The target object features are read from a preset object encoding file based on the current object information; Based on the currently clicked content information, the target content features are read from a preset content encoding file; Based on the feature extraction information, feature extraction is performed on the current location feature, the target object feature, and the target content feature to obtain target feature encoded data; The target feature encoding data is decoded to obtain the target object, the target display content, and the target display position; wherein, the target object includes the clicked object, or a non-clicked object; Click prediction is performed on the target feature encoding data using a preset prediction model to obtain a target click score; the target click score is used to characterize the probability that the target display content located at the target display position is clicked by the target object; The current object information is subjected to feature encoding processing based on the feature engineering configuration information to obtain the current object features; Based on the feature engineering configuration information, the currently clicked content information is subjected to feature encoding processing to obtain the current content features; Based on the feature extraction information, feature extraction is performed on the current object features, the current content features, and the current position features to obtain current feature encoding data; If the target feature encoding data is the same as the current feature encoding data, then obtain the display time of the target feature encoding data; The time difference is calculated by comparing the current click time with the display time. The target click score is adjusted based on the time difference. The target content is recommended to the target object based on the target click score and the target display position.

2. The method according to claim 1, characterized in that, Before performing click prediction on the target feature encoding data using a preset prediction model to obtain the target click score, the method further includes: Pre-training the prediction model specifically includes: Obtain historical display data and its original tags; the historical display data includes historical object information, historical display content information, and historical display location information of the historical object. Based on the feature engineering configuration information, the historical object information is subjected to feature encoding processing to obtain historical object features; Based on the feature engineering configuration information, the historical display content information is subjected to feature encoding processing to obtain historical content features; Based on the feature engineering configuration information, the historical display location information is subjected to feature encoding processing to obtain historical location features; Based on the feature extraction information, feature extraction is performed on the historical object features, historical content features, and historical location features to obtain training feature encoding data; Based on the original labels, determine the original training score of the training feature encoding data; The target training score is obtained by predicting the score of the training feature-encoded data using the prediction model. The loss value is obtained by calculating the loss based on the original training score and the target training score; The parameters of the prediction model are adjusted based on the loss value to obtain the trained prediction model.

3. The method according to claim 2, characterized in that, The step of obtaining historical display data and obtaining the original tags of the historical display data includes: Obtain the object behavior log of the historical object; The object behavior log is read to obtain the historical object information, the historical display content information, and the historical display location information; The historical object information, the historical display content information, and the historical display location information are merged to obtain the historical display data; The object behavior log is read by time to obtain the content annotation time of the historical display data. The content annotation time includes: content exposure time or content click time. The original tag is determined based on the time indicated by the content.

4. The method according to claim 3, characterized in that, Before reading the object behavior log, the method further includes: Updating the object behavior log specifically includes: Obtain object data and object identifier code, wherein the object data is used to provide the historical object information; The content data and content identifier code are obtained, and the content data is used to provide the historical content information; Record the object identifier and the content identifier to the object behavior log; The object data is associated with the object behavior log through the object identifier code, and the content data is associated with the object behavior log through the content identifier code.

5. The method according to claim 4, characterized in that, The object data includes basic object information and object behavior information, and the content data includes basic content information and content statistics. The step of associating the object data with the object behavior log using the object identifier and the content data with the object behavior log using the content identifier includes: A mapping relationship is established between the object's basic information and the object's behavioral information to obtain an object wide table; A mapping relationship is established between the basic content information and the content statistics information to obtain a wide content table; The object wide table is associated with the object behavior log through the object identifier code, and the content wide table is associated with the object behavior log through the content identifier code.

6. The method according to claim 1, characterized in that, The step of reading feature engineering configuration information and feature extraction information from a preset configuration file includes: Get the current feature configuration version number; Based on the current feature configuration version number, the feature engineering configuration information and the feature extraction information are read from the configuration file.

7. A content recommendation device based on feature engineering, characterized in that, The device includes: The click data acquisition module is used to acquire the current click data, which includes the current object information of the clicked object, the current click content information, the current click location information, and the current click time. The configuration information acquisition module is used to read feature engineering configuration information and feature extraction information from a preset configuration file; The location feature determination module is used to perform feature encoding processing on the current click location information according to the feature engineering configuration information to obtain the current location feature; An object feature determination module is used to read target object features from a preset object encoding file based on the current object information; The content feature determination module is used to read target content features from a preset content encoding file based on the currently clicked content information; The feature extraction module is used to extract features from the current location features, the target object features, and the target content features based on the feature extraction information to obtain target feature encoding data. The feature decoding module is used to decode the target feature encoding data to obtain the target object, the target display content, and the target display position; wherein, the target object includes the clicked object or a non-clicked object; The model prediction module is used to predict clicks on the target feature encoding data using a preset prediction model to obtain a target click score; the target click score is used to characterize the probability that the target display content located at the target display position is clicked by the target object; The content recommendation module is used to recommend the content displayed for the target to the target object based on the target click score and the target display position; After performing click prediction on the target feature-encoded data using a preset prediction model to obtain the target click score, the device is further used to: The current object information is subjected to feature encoding processing based on the feature engineering configuration information to obtain the current object features; Based on the feature engineering configuration information, the currently clicked content information is subjected to feature encoding processing to obtain the current content features; Based on the feature extraction information, feature extraction is performed on the current object features, the current content features, and the current position features to obtain current feature encoding data; If the target feature encoding data is the same as the current feature encoding data, then obtain the display time of the target feature encoding data; The time difference is calculated by comparing the current click time with the display time. The target click score is adjusted based on the time difference.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the feature-engineered content recommendation method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the content recommendation method based on feature engineering as described in any one of claims 1 to 6.

Images