Content determination method, apparatus, computer-readable storage medium and computer equipment

Abstract

This application discloses a content determination method, apparatus, computer-readable storage medium, and computer device. The method involves: acquiring at least one behavioral feature corresponding to the historical behavioral data of a target object; acquiring at least two attribute features corresponding to each of at least two candidate contents, where the attribute features are the features corresponding to the attribute data of the candidate contents; calculating the target object's degree of interest in each candidate content based on the behavioral features and attribute features; and determining the target candidate content based on the degree of interest. The content determination method provided by this application can improve the accuracy of content determination, thereby improving the accuracy of content data recommended by the recommendation system to users.

Description

Technical Field

[0001] This application relates to the field of content recommendation technology, specifically to a content determination method, apparatus, computer-readable storage medium, and computer device. Background Technology

[0002] With the continuous development of internet technology, people's lives have become inseparable from the internet. In the internet age, with the rapid expansion of content, people face increasing pressure in choosing content, which reduces the efficiency of content use and leads to information overload.

[0003] A recommendation system is a personalized recommendation system that recommends information and products that users are interested in based on their content needs and interests. A good recommendation system not only provides users with personalized content and services but also builds a close relationship with them.

[0004] However, in some cases, such as in internet advertising recommendation scenarios, the content recommended by the recommendation system to users is not accurate enough. Summary of the Invention

[0005] This application provides a content determination method, apparatus, computer-readable storage medium, and computer device. The content determination method can effectively improve the accuracy of content determination in a recommendation system, thereby improving the accuracy of content recommended to users.

[0006] The first aspect of this application provides a method for determining content, the method comprising:

[0007] Obtain at least one behavioral feature corresponding to the historical behavioral data of the target object;

[0008] Obtain at least two dimensions of attribute features corresponding to each of at least two candidate contents, wherein the attribute features are the features corresponding to the attribute data of the candidate contents;

[0009] The degree of interest of the target object in each candidate content is calculated based on the behavioral characteristics and attribute characteristics.

[0010] Target candidate content is determined based on the level of interest.

[0011] A second aspect of this application provides a content determining apparatus, the apparatus comprising:

[0012] The first acquisition unit is used to acquire at least one behavioral feature corresponding to the historical behavioral data of the target object;

[0013] The second acquisition unit is used to acquire at least two dimensions of attribute features corresponding to each candidate content in at least two candidate contents, wherein the attribute features are features corresponding to the attribute data of the candidate content.

[0014] A calculation unit is used to calculate the target object's degree of interest in each candidate content based on the behavioral features and the attribute features;

[0015] A determining unit is used to determine target candidate content based on the degree of interest.

[0016] In some embodiments, the computing unit includes:

[0017] A calculation subunit is used to calculate a first weight coefficient for each behavioral feature based on the behavioral feature and the attribute feature;

[0018] The processing subunit is used to perform weighted processing on the at least one behavioral feature according to the first weight coefficient of each behavioral feature to obtain the interest feature corresponding to each candidate content.

[0019] A subunit is defined for determining the degree of interest of the target object in each candidate content based on the interest features.

[0020] In some embodiments, the computing subunit includes:

[0021] The first acquisition module is used to acquire the second weight coefficient corresponding to each attribute feature;

[0022] The calculation module is used to calculate a first weight coefficient for each behavioral feature based on the behavioral feature, the attribute feature, and the second weight coefficient.

[0023] In some embodiments, the acquisition module includes:

[0024] The extraction submodule is used to extract the target attribute data associated with each attribute data contained in each historical behavior data in the historical behavior data.

[0025] The determination submodule is used to determine the second weight coefficient corresponding to each attribute data according to the target attribute data, wherein each attribute data is each of the at least two attribute data corresponding to each candidate content.

[0026] In some embodiments, the computing module includes:

[0027] The acquisition submodule is used to acquire the behavior feature vector corresponding to each behavior feature and the attribute feature vector corresponding to each attribute feature.

[0028] The processing submodule is used to perform element-level multiplication on any target behavior feature and each attribute feature, and to perform weighted processing on the processing result according to the second weight coefficient, and to perform weighted processing on the target behavior feature according to the first weight coefficient corresponding to the weighted processing result.

[0029] The calculation submodule is used to traverse each behavior feature based on the calculation method of the first weight coefficient corresponding to the target behavior feature, and calculate the first weight coefficient of each behavior feature.

[0030] In some embodiments, the determining subunit includes:

[0031] The second acquisition module is used to acquire the trained interest level prediction model;

[0032] The prediction module is used to input the interest features into the interest level prediction model to obtain the target object's interest level for each candidate content.

[0033] In some embodiments, the content determination apparatus provided in this application further includes:

[0034] The third acquisition unit is used to acquire target attribute data of at least one dimension of the target candidate content;

[0035] The recommendation unit is used to recommend the target candidate content to the target user based on the target attribute data.

[0036] In some embodiments, the first acquisition unit includes:

[0037] The first acquisition subunit is used to acquire at least one historical behavior data of the target object within a preset time period;

[0038] The first mapping subunit is used to map the historical behavior data to a vector space to obtain at least one behavioral feature of the target object.

[0039] In some embodiments, the second acquisition unit includes:

[0040] The second acquisition subunit is used to acquire resource attribute data, channel attribute data and associated content attribute data for each of at least two candidate contents;

[0041] The second mapping subunit is used to map the resource attribute data, channel attribute data and associated content attribute data to the vector space respectively to obtain the resource attribute features, channel attribute features and associated content attribute features corresponding to each candidate content.

[0042] The computing unit is also used for:

[0043] The degree of interest of the target object in each candidate content is calculated based on the behavioral characteristics, resource attribute characteristics, channel attribute characteristics, and associated content attribute characteristics.

[0044] A third aspect of this application also provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to perform the steps of the content determination method provided in the first aspect of this application.

[0045] The fourth aspect of this application provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the content determination method provided in the first aspect of this application.

[0046] The fifth aspect of this application provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps in the content determination method provided in the first aspect.

[0047] The content determination method provided in this application involves: acquiring at least one behavioral feature corresponding to the historical behavioral data of a target object; acquiring at least two attribute features corresponding to each of at least two candidate contents, wherein the attribute features are the features corresponding to the attribute data of the candidate contents; calculating the target object's degree of interest in each candidate content based on the behavioral features and attribute features; and determining the target candidate content based on the degree of interest.

[0048] Therefore, the content determination method provided in this application can bidirectionally calculate the user's level of interest in each candidate content based on the user's historical behavioral characteristics and the multi-dimensional attribute characteristics of the candidate content, and further determine the target content to recommend to the user based on the level of interest. Because the multi-dimensional attribute characteristics of the candidate content are considered when calculating the user's level of interest, the calculated level of interest is more accurate, thereby improving the accuracy of determining the target content and consequently improving the accuracy of the content recommended to the user. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of a scenario defined in this application;

[0051] Figure 2 This is a flowchart illustrating the content determination method provided in this application;

[0052] Figure 3 This is another flowchart illustrating the content determination method provided in this application;

[0053] Figure 4 This is a schematic diagram of the behavioral feature weight coefficient evaluation model provided in this application;

[0054] Figure 5 This is a schematic diagram illustrating the effect of pruning attribute feature vectors in this application;

[0055] Figure 6 This is a heatmap of the attribute data of the target advertising content in this application across various dimensions;

[0056] Figure 7 This is a schematic diagram of the structure of the content determining device provided in this application;

[0057] Figure 8 This is a schematic diagram of the structure of the computer device provided in this application. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0059] This invention provides a content determination method, apparatus, computer-readable storage medium, and terminal. The content determination method can be used in the content determination apparatus. The content determination apparatus can be integrated into a computer device, which can be a terminal or a server. The terminal can be a mobile phone, tablet computer, laptop computer, smart TV, wearable smart device, personal computer (PC), or in-vehicle terminal, etc. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or 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, Content Delivery Network (CDN), and big data and artificial intelligence platforms. The server can also be a node in a blockchain.

[0060] Please see Figure 1This is a schematic diagram of a scenario for the content determination method provided in this application. As shown in the figure, computer device A acquires at least one behavioral feature corresponding to the historical behavioral data of the target object; acquires at least two attribute features corresponding to each of at least two candidate contents, wherein the attribute features are the features corresponding to the attribute data of the candidate contents; calculates the target object's degree of interest in each candidate content based on the behavioral features and attribute features; and determines the target candidate content based on the degree of interest.

[0061] It should be noted that, Figure 1 The illustrated content determination scenario diagram is merely an example. The content determination scenario described in the embodiments of this application is intended to more clearly illustrate the technical solution of this application and does not constitute a limitation on the technical solution provided by this application. Those skilled in the art will understand that as content determination scenarios evolve and new business scenarios emerge, the technical solution provided by this application is equally applicable to similar technical problems.

[0062] The implementation scenarios described above will be explained in detail below.

[0063] In related technologies, recommender systems, when recommending content to users, can acquire users' historical behavior data and candidate content data, calculate the user's level of interest in each candidate content data, and recommend content to the user based on this level of interest. However, currently, the calculation of user interest in candidate content generally only considers the content data itself or a single-dimensional attribute of the content data, resulting in poor accuracy in calculating user interest in candidate content, and consequently, low accuracy in the content recommended by the recommender system. To address this, this application provides a content determination method that can calculate the user's level of interest in each candidate content based on multi-dimensional attribute features, thereby making the calculated user interest in candidate content more accurate and improving the accuracy of the content recommended by the recommender system.

[0064] This application will describe the embodiments from the perspective of a content determination device, which can be integrated into a computer device. The computer device can be a terminal or a server. The terminal can be a mobile phone, tablet computer, laptop computer, smart TV, wearable smart device, personal computer (PC), or in-vehicle terminal, etc. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or 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, content delivery network (CDN) acceleration services, and big data and artificial intelligence platforms. Figure 2 The diagram shown is a flowchart illustrating the content determination method provided in this application. The method includes:

[0065] Step 101: Obtain at least one behavioral feature corresponding to the historical behavioral data of the target object.

[0066] The target object can be a specific user, a user's identity document (ID), or a virtual artificial intelligence object.

[0067] Historical behavioral data of the target object can be data corresponding to the target object's historical operations on content data. This content data can be content provided by various applications, such as news and information applications, short video applications, e-commerce platforms, and instant messaging applications. Specifically, content data can include text information, video information, short video content, product information, and online advertising content. The target object's historical operations on this content data can include clicks, reading, watching, liking, saving, registration, and purchasing behaviors.

[0068] To calculate the target object's interest level for each candidate content based on its historical behavior, and then recommend matching candidate content to the target object according to the interest level, the target object's historical behavior needs to be transformed into feature data that the recommendation system can process, resulting in behavioral features corresponding to each historical behavior data point. The target object can have one or multiple historical behavior data points, and each historical behavior data point corresponds to a behavioral feature.

[0069] In some embodiments, obtaining at least one behavioral feature corresponding to the historical behavioral data of the target object includes:

[0070] 1. Obtain at least one historical behavior data of the target object within a preset time period;

[0071] 2. Map historical behavior data to a vector space to obtain at least one behavioral feature of the target object.

[0072] In this embodiment, at least one historical behavior data of the target object within a preset time period can be obtained. Then, each historical behavior data point is mapped to a vector space to obtain at least one behavioral feature of the target object. That is, embedding can be used to transform historical behavior data into computationally readable vector data. Embedding is a term in deep learning, specifically referring to mapping high-dimensional raw data (such as images, sentences, etc.) to low-dimensional data, making the high-dimensional raw data usable for computation. For example, word embedding maps sentences composed of words to a representation vector. Therefore, when a user's historical behavior data is expressed in text form, the text can be mapped to a vector space to obtain the behavioral features corresponding to the historical behavior data.

[0073] Step 102: Obtain at least two dimensions of attribute features corresponding to each of the at least two candidate contents.

[0074] When determining recommended content data for a target audience, it is generally necessary to identify the content data that best matches the target audience or the content data that the target audience is most interested in from a content library. This content library can include multiple candidate content items, such as at least two. Specifically, when the candidate content is advertising content, at least two dimensions of attribute features for each candidate ad in the advertising library can be obtained.

[0075] In related technologies, such as the Deep Interest Network for Click-Through Rate Prediction (DIN) and Deep Interest Evolution Network for Click-Through Rate Prediction, when calculating the relevance between a target ad and a user's historical behavior, only a single dimension of the target ad's attribute features are often considered, such as the advertiser's identity document (ID) or ad ID, ignoring the impact of other features on user behavior. This is especially true for features at the channel and related content dimensions.

[0076] In this embodiment, the at least two-dimensional attribute features corresponding to each candidate content can be the features corresponding to the at least two-dimensional attribute data of each candidate content. Specifically, the at least two-dimensional attribute data can include resource attribute data, channel attribute data, and associated content attribute data of the candidate content. Of course, the attribute data of the candidate content can also include attribute data of other dimensions, such as time-dimensional attribute data, etc. Here, resource attribute data, channel attribute data, and associated content attribute data are only examples.

[0077] Specifically, taking the candidate content as the advertising content as an example, the resource attribute data of the advertising content can include advertiser ID data, ad ID data, and ad content data; the channel attribute data of the advertising content can be the data of the advertising channel, such as advertising on a WeChat Official Account or advertising on WeChat Moments; the related content attribute data of the advertising content can be the context data of the ad, such as user activity data and ad data before and after the ad when advertising on WeChat Moments, or the post data before and after the ad when advertising on a WeChat Official Account. For the same user, the probability of an ad being clicked or converted varies depending on the advertising channel. For example, some users frequently use WeChat Moments but rarely view WeChat Official Account content; therefore, advertising on WeChat Moments will have a higher probability of clicks or conversions compared to advertising on WeChat Official Accounts. Furthermore, even when using the same advertising channel, the placement of the ad can influence the probability of a user clicking on it. Specifically, an ad for basketball shoes posted on a WeChat official account featuring basketball game highlights is more likely to be clicked or converted than an ad posted on a food promotion account. In other words, the characteristics of the advertising channel and the related content of the ad placement also affect user behavior.

[0078] In considering the impact of advertising features on user behavior, this application's embodiments, in addition to considering the features of the advertisement itself, such as advertiser ID and advertisement ID, also consider the features corresponding to the advertisement's delivery channel, the features corresponding to the advertisement's associated content, and other unlisted advertisement-related features. Feature extraction of advertisement features from multiple dimensions describes the relationship between user behavior and advertisement features from multiple perspectives, making the predicted user interest in the advertisement more accurate.

[0079] Step 103: Calculate the target object's level of interest in each candidate content based on the behavioral features and attribute features.

[0080] Here, the behavioral feature refers to at least one behavioral feature of the aforementioned target object, and the attribute feature refers to at least two dimensions of attribute features corresponding to each of the aforementioned at least two candidate contents. In related technologies, the target object's interest in each candidate content is often calculated based on only the target object's behavioral feature and one dimension of the candidate content's attribute feature. For example, when calculating the target object's interest in each advertisement content, only the target object's behavioral feature and the advertisement's resource attribute feature are often considered, that is, only the corresponding attribute features such as advertiser data, advertisement ID data, and advertisement content data are considered to calculate the target object's interest in each candidate advertisement content, without considering other dimensions of attribute features such as channel attribute data and associated content attribute data.

[0081] In this embodiment of the application, by comprehensively considering the influence of the multi-dimensional attribute features of the candidate content on the degree of interest, the accuracy of the calculated degree of interest of the target object on each candidate content can be effectively improved, thereby improving the accuracy of the determination of the target candidate content.

[0082] In some embodiments, the degree of interest of the target object in each candidate content is calculated based on behavioral characteristics and attribute characteristics in at least two dimensions, including:

[0083] 1. Calculate the first weight coefficient for each behavioral feature based on behavioral features and attribute features in at least two dimensions;

[0084] 2. Weight at least one behavioral feature based on the first weight coefficient of each behavioral feature to obtain the interest feature corresponding to each candidate content;

[0085] 3. Determine the target audience's level of interest in each candidate content based on interest characteristics.

[0086] Here, the first weight coefficient for each behavioral feature is the weight coefficient corresponding to each behavioral feature and each target candidate content. For example, if there are three behavioral features A, B, and C, and three candidate contents a, b, and c, then for candidate content a, behavioral features A, B, and C each have a weight coefficient corresponding to candidate content a; for candidate content b and candidate content c, behavioral features A, B, and C also each have a weight coefficient corresponding to the candidate content. To distinguish it from other subsequent weight coefficients, this can be referred to as the first weight coefficient.

[0087] The first weight coefficient of each behavioral feature represents its contribution to the level of interest, where level of interest refers to the target audience's interest in the target content. Generally, the higher the first weight coefficient of a behavioral feature, the greater its contribution to the level of interest. For example, if behavioral feature A corresponds to clicking on candidate content 'a', behavioral feature B corresponds to liking candidate content 'a', and behavioral feature C corresponds to paying for candidate content 'a', then the weight coefficient of behavioral feature A is less than the weight coefficient of behavioral feature B, and the weight coefficient of behavioral feature B is less than the weight coefficient of behavioral feature C.

[0088] In related technologies, traditional click-through rate (CTR) prediction models often use weighted pooling or average pooling of users' historical behavioral features. However, in reality, different historical user behaviors have varying impacts on predicting click-through rates or conversion rates for different ads. Simply using weighted pooling or average pooling of users' historical behavioral features fails to reflect the correlation between different user historical behaviors and users' interest in ads, resulting in inaccurate predictions of user interest. Therefore, in this embodiment, a first weight coefficient for each behavioral feature can be calculated based on behavioral features and at least two dimensions of attribute features. This determines the weight coefficient corresponding to each behavioral feature, differentiating and accurately reflecting the impact of different behavioral features on users' click-through rates for target content, thereby improving the accuracy of predicted user interest in ads.

[0089] In this embodiment, a first weight coefficient for each behavioral feature corresponding to each candidate content can be calculated using behavioral features and multiple attribute features corresponding to each candidate content. Then, the first weight coefficients for each behavioral feature are used to perform a weighted sum of the multiple behavioral features corresponding to each candidate content to obtain the interest feature corresponding to each candidate content. This interest feature represents the degree of interest of the target object in the candidate content and can be a feature vector.

[0090] Furthermore, the degree of interest of the target object in each candidate content can be determined based on this interest feature, and this degree of interest can be an interest value.

[0091] In some embodiments, calculating a first weight coefficient for each behavioral feature based on behavioral features and attribute features in at least two dimensions includes:

[0092] 1.1 Obtain the second weight coefficient corresponding to each attribute feature in at least two dimensions;

[0093] 1.2 Calculate the first weight coefficient for each behavioral feature based on the behavioral features, at least two dimensions of attribute features, and the second weight coefficient.

[0094] The at least two dimensions of attribute features are the attribute features corresponding to at least two dimensions of each candidate content. In this embodiment, since the contribution of each attribute feature to the degree of interest may vary among the multiple dimensions of attribute features corresponding to each candidate content, in order to further improve the accuracy of interest calculation and thus improve the accuracy of target candidate content determination, the weight coefficient of each attribute feature corresponding to each content can be obtained first. Here, to distinguish it from the weight coefficient of the aforementioned behavioral features, it can be called the second weight coefficient. The second weight coefficient here represents the degree of influence of each dimension of attribute features on the user's behavior, or the degree of importance. After obtaining the second weight coefficient of each attribute feature corresponding to each candidate content, the first weight coefficient of each behavioral feature among the multiple behavioral features corresponding to each candidate content is calculated based on the multiple dimensions of attribute features corresponding to each candidate content, the second weight coefficient of each attribute feature, and the multiple behavioral features of the target object.

[0095] In some embodiments, obtaining the second weight coefficient corresponding to each attribute feature in at least two dimensions of attribute features includes:

[0096] 1.1.1 Extract the target attribute data associated with each attribute data contained in each historical behavior data from the historical behavior data;

[0097] 1.1.2. Determine the second weight coefficient corresponding to each attribute data according to the target attribute data, wherein each attribute data is each of the at least two attribute data corresponding to each candidate content.

[0098] In this embodiment, target attribute data associated with each attribute data can be extracted from the historical behavior data of the target object. The target attribute data is data related to the behavioral scenario corresponding to the historical behavior data. For example, when the candidate content is an advertisement for a sports product, the difference in click-through rate (CTR) obtained by placing the advertisement on different channels is much greater than the difference in CTR obtained by placing the advertisement in different contextual content positions. That is, the influence of channel attribute features on the CTR increase of this advertisement is greater than that of associated content attribute features. Therefore, for this advertisement, the weight coefficient corresponding to the channel attribute features is generally higher than the weight coefficient corresponding to the associated content attribute features. In other words, different attribute features have different impacts on the target object's interest level, and this difference can be characterized by a second weight coefficient for each attribute feature. In this embodiment, a weight coefficient can also be set for each dimension of attribute features to achieve pruning of attribute features of different dimensions, and even remove attribute features with less influence. This makes the contribution of attribute features corresponding to attribute data of different dimensions to the degree of interest more consistent with the actual situation, and makes the calculated result of the target's interest level for each candidate content more accurate, thus making the determined target candidate content more accurate. In this process, weight coefficients are set for the attribute features of each dimension. An initial weight coefficient can be set first, and then the initial weight coefficient can be iteratively updated during the deep learning process to learn a more accurate weight coefficient for each attribute feature.

[0099] In some embodiments, more historical behavior data of objects towards the target candidate content can be obtained, and then this historical behavior data can be labeled to obtain training data. The labeled data can be the weight coefficients corresponding to each attribute feature. Further, a pre-defined neural network model can be trained based on the training data to obtain the trained neural network model.

[0100] Then, once the historical behavior data corresponding to each candidate content is determined, the weight coefficient corresponding to each attribute feature can be determined through the neural network model described above.

[0101] In some embodiments, calculating a first weight coefficient for each behavioral feature based on behavioral features, at least two dimensions of attribute features, and a second weight coefficient includes:

[0102] 1.2.1 Obtain the behavioral feature vector corresponding to each behavioral feature and the attribute feature vector corresponding to each attribute feature;

[0103] 1.2.2 Perform element-level multiplication on any target behavior feature and each attribute feature, and weight the processing result according to the second weight coefficient, and determine the first weight coefficient corresponding to the target behavior feature according to the weighted processing result.

[0104] 1.2.3. The calculation method based on the first weight coefficient corresponding to the target behavior feature is to traverse each behavior feature and calculate the first weight coefficient of each behavior feature.

[0105] In the embodiments of this application, behavioral features and attribute features can be specifically represented in the form of vectors, that is, to obtain the behavioral feature vector corresponding to each behavioral feature, and the attribute feature vector corresponding to each attribute feature in the multiple dimensions of attribute features corresponding to each candidate content.

[0106] For any target behavior feature vector, the element-wise product between it and multiple attribute feature vectors of each candidate content can be calculated. Specifically, for example, given two-dimensional vectors c = (c1, d1) and d = (c2, d2), the element-wise product between vectors c and d can be expressed as:<c,d> = (c1*c2, d1*d2). After calculating the element-wise multiplication between the target behavior feature vector and the feature vector of each attribute of the candidate content, the vectors corresponding to these element-wise multiplications are weighted and pooled according to the second weight coefficient corresponding to each attribute feature of any candidate content, resulting in the interaction vector between the target behavior feature vector and the candidate content. Normalizing (softmax) this interaction vector yields the weight coefficients corresponding to the target behavior feature vector and the candidate content. Iterating through each behavior feature yields the first weight coefficient corresponding to each behavior feature and the candidate content. Here, the candidate content is any one of at least two candidate contents. Furthermore, iterating through each candidate content yields the first weight coefficient for each behavior feature corresponding to each candidate content.

[0107] In some embodiments, determining the target object's level of interest in each candidate content based on interest characteristics includes:

[0108] 3.1 Obtain the trained interest prediction model;

[0109] 3.2 Input the interest features into the interest level prediction model to obtain the target object's interest level for each candidate content.

[0110] In this embodiment, after determining the target object's interest features for each candidate content, these interest features can be input into a preset interest prediction model for processing, resulting in the interest level output by the interest prediction model. This interest level can be an interest value, and the interest prediction model can be a deep click-through rate (CTR) model.

[0111] In some embodiments, obtaining attribute features of at least two dimensions corresponding to each of the at least two candidate contents includes:

[0112] A. Obtain resource attribute data, channel attribute data, and associated content attribute data for each of at least two candidate contents;

[0113] B. Map the resource attribute data, channel attribute data, and associated content attribute data to the vector space respectively to obtain the resource attribute features, channel attribute features, and associated content attribute features corresponding to each candidate content.

[0114] The degree of interest of the target object in each candidate content is calculated based on behavioral and attribute characteristics, including:

[0115] C. Calculate the target object's interest level in each candidate content based on behavioral characteristics, resource attribute characteristics, channel attribute characteristics, and associated content attribute characteristics.

[0116] In this embodiment, the at least two-dimensional attribute features corresponding to each candidate content can specifically include resource attribute features, channel attribute features, and associated content attribute features. Specifically, resource attribute data, channel attribute data, and associated content attribute data corresponding to each candidate content can be obtained first. Then, these attribute data can be mapped to a vector space to obtain the feature vector corresponding to each attribute data, i.e., the feature vector corresponding to at least two dimensions for each candidate content. Thus, the step of calculating the target object's interest level in each candidate content based on behavioral features and attribute features can be transformed into calculating the interest level corresponding to each candidate content based on behavioral features and the resource attribute features, channel attribute features, and associated content attributes corresponding to each candidate content.

[0117] It's important to understand that resource attribute data, channel attribute data, and related content attribute data represent only a few relatively important dimensions of the candidate content's multi-dimensional attribute data, and do not represent all dimensions of the candidate content's attribute data. Furthermore, the candidate content's attribute data may also include content object data, data on the content object's location, content management system data, and so on.

[0118] Step 104: Determine target candidate content based on the stated level of interest.

[0119] Specifically, target candidate content can be determined based on the target audience's level of interest in each candidate content. This level of interest can be defined as an interest value, and generally, the candidate content with the highest interest value is selected as the target candidate content. In other words, at least two candidate contents can be ranked according to their interest values, and the candidate content with the highest interest value is selected as the target candidate content.

[0120] In some embodiments, the content determination method provided in this application may further include:

[0121] 1. Obtain target attribute data for at least two dimensions of the target candidate content;

[0122] 2. Recommend the target candidate content to the target user based on the target attribute data.

[0123] Since each candidate content has at least two dimensions of attribute data, the target candidate content that the target object is most interested in also has at least two dimensions of target attribute data.

[0124] At least two dimensions of target attribute data indicate the resource attributes, channel attributes, and associated content attributes of the target candidate content. Based on the resource attributes, the resource data corresponding to the target candidate content can be obtained, such as advertiser information, ad ID, and ad content; based on the channel attributes, the placement channel of the target candidate content can be obtained, such as advertising on WeChat Moments or WeChat Official Accounts; based on the associated content attributes, the content associated with the target candidate content can be determined, such as the context information of the ad, thereby determining the ad placement location.

[0125] In this way, after obtaining the resource data, channel data, and related content data of the target candidate content, accurate recommendations can be made for the target candidate content based on the above data.

[0126] As described above, the content determination method provided in this application embodiment involves: obtaining at least one behavioral feature corresponding to the historical behavioral data of the target object; obtaining at least two attribute features corresponding to each of the at least two candidate contents, wherein the attribute features are the features corresponding to the attribute data of the candidate contents; calculating the target object's degree of interest in each candidate content based on the behavioral features and attribute features; and determining the target candidate content based on the degree of interest.

[0127] Therefore, the content determination method provided in this application can bidirectionally calculate the user's level of interest in each candidate content based on the user's historical behavioral characteristics and the multi-dimensional attribute characteristics of the candidate content, and further determine the target content to recommend to the user based on the level of interest. Because the multi-dimensional attribute characteristics of the candidate content are considered when calculating the user's level of interest, the calculated level of interest is more accurate, thereby improving the accuracy of determining the target content and consequently improving the accuracy of the content recommended to the user.

[0128] Furthermore, this application adds appropriate weight coefficients to each attribute feature corresponding to each candidate content, which can identify and remove irrelevant attribute features, making the calculated weight coefficients of each behavioral feature more accurate. This improves the accuracy of determining target candidate content, and consequently improves the accuracy of content recommendation by the recommendation system.

[0129] This application also provides a content determination method, which can be used in a computer device, which can be a terminal or a server. In this embodiment, an advertisement as the content will be described in detail. Figure 3 The diagram shown is another flowchart illustrating the content determination method provided in this application. The method specifically includes:

[0130] Step 201: The computer device acquires the historical behavioral characteristics of the target user.

[0131] The historical behavioral features of the target user refer to the features corresponding to the target user's historical behavioral data. A user's historical behavior can include actions such as clicking, browsing, liking, saving, donating coins, placing orders, and making payments related to content data. In this embodiment, the content data can specifically be advertising data. The features corresponding to the historical behavioral data can be feature vectors corresponding to the historical behavioral data. Specifically, word embedding processing can be performed on the descriptive text of the historical behavioral data to obtain H behavioral feature vectors v corresponding to each historical behavior of the user towards the advertisement. i Where H≥1, H≥i≥1, v i Let be the i-th behavioral feature vector of the user.

[0132] Step 202: The computer device obtains the multi-dimensional attribute features of each candidate advertisement in the advertisement library.

[0133] The advertising library can be a preset advertising library, specifically all advertising resources stored in a preset computer device, which can be a server.

[0134] The multi-dimensional attribute features of each candidate ad can be represented as feature vectors corresponding to the multi-dimensional attribute data of each candidate ad. Specifically, the multi-dimensional attribute data can include resource attribute data, channel attribute data, and associated content attribute data. Similarly, word embedding methods can be used to transform the multi-dimensional attribute data of each candidate ad into a multi-dimensional attribute feature vector for each candidate ad. Specifically, for any target candidate ad, its corresponding M attribute feature vectors e can be determined. j Where M≥2, M≥j≥1, e j Let be the feature vector of the j-th attribute of the target candidate advertisement.

[0135] Step 203: The computer device obtains the first weight coefficient corresponding to each attribute feature of each candidate advertisement.

[0136] For each attribute feature of each candidate advertisement, its corresponding weight coefficient r can be obtained. F(j) This can be referred to as the first weighting coefficient. Here, F(j) represents the attribute data corresponding to the j-th attribute feature of the target candidate advertisement, r F(j) The first weight coefficient is the j-th attribute feature of the target candidate advertisement.

[0137] Among them, the first weight coefficient corresponding to each attribute feature of the target candidate advertisement can be preset with a random value, and then pruned during the model training process to automatically identify and remove features of irrelevant dimensions.

[0138] Step 204: The computer device calculates the second weight coefficient corresponding to each historical behavior feature and each candidate advertisement based on the historical behavior features, the attribute features of each candidate advertisement, and the corresponding first weight coefficient.

[0139] Among them, the behavioral feature vector v corresponding to the user's historical behavioral characteristics is determined. i The attribute feature vector e corresponding to the multiple dimensions of the target candidate advertisement's attribute features. j and the first weight coefficient r corresponding to each attribute feature vector F(j) Then, the second weight coefficient of the target candidate advertisement corresponding to each behavioral feature can be calculated using the following formula (1):

[0140] Formula (1)

[0141] in, The second weight coefficient is the target candidate ad corresponding to the user's i-th behavioral feature vector. For vector v i and e j Perform element-level multiplication. () represents the softmax normalization process applied to the vector within the parentheses.

[0142] Then, each candidate ad can be iterated through, and the second weight coefficient of each candidate ad corresponding to each user's features can be calculated according to the above formula.

[0143] Step 205: The computer device performs weighted processing on the user's historical behavior features according to the second weighting coefficient to obtain the user's interest features for each candidate advertisement.

[0144] For the aforementioned target candidate advertisements, since the second weight coefficient corresponding to each user's behavioral characteristics has been determined, the behavioral characteristics of each user can be weighted and pooled using the aforementioned second weight coefficient to obtain the user's interest characteristics in the target candidate advertisements. The specific calculation formula is as follows:

[0145] Formula (1)

[0146] in, For user u, v represents the interest characteristics of the target candidate ads. i Let i be the user's i-th behavioral feature vector. It can be calculated using the formula (1) above.

[0147] Similarly, each candidate ad can be iterated through to obtain the user's interest characteristics for each candidate ad. .

[0148] Step 206: The computer device inputs the user's interest features for each candidate advertisement into a preset interest prediction model to obtain the user's interest value for each candidate advertisement.

[0149] The preset interest prediction model can be a deep click-through rate (CTR) prediction model. After obtaining the feature vectors corresponding to the user's interest features for each candidate content, these vectors can be input into the deep CTR model for prediction to obtain the user's interest value for each candidate advertisement.

[0150] Step 207: The computer device determines the target candidate advertisement based on the user's interest value for each candidate advertisement.

[0151] This process involves determining the interest value for each candidate ad and sorting the candidates based on their interest values ​​from highest to lowest. The candidate ad with the highest interest value is then identified as the target candidate ad.

[0152] Step 208: The computer device recommends target candidate advertisements to the user.

[0153] This involves acquiring attribute data from multiple dimensions of the target candidate advertisement. Specifically, this includes resource attribute data, channel attribute data, and related content attribute data. Resource attribute data can include advertiser information, advertisement ID information, and advertisement content information; channel attribute data can include information about the channels through which the target candidate advertisement is placed, such as web pages, official accounts, or WeChat Moments; related content attribute data can include contextual relationship data within the target channels, which can be used to determine the precise placement of the target candidate advertisement. Once this multi-dimensional attribute information is obtained, the advertisement can be precisely targeted based on this information.

[0154] Specifically, the above solution can be used for personalized advertising recommendations. Online advertising is the most direct and transparent way for most internet companies to monetize traffic. Taking an instant messaging application's information sharing platform as an example, when a user opens the information sharing platform and refreshes the list, the advertising recommendation system receives the refresh request, obtains the user's historical behavior data and multi-dimensional feature data of multiple candidate advertisements, then calculates the user's level of interest in each candidate advertisement, determines the target candidate advertisement based on the level of interest, and then adaptively delivers the target candidate advertisement. It is understood that in the specific embodiments of this application, data related to user historical behavior data is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions.

[0155] Among them, such as Figure 4 The diagram shows a model for calculating the weight coefficients corresponding to each behavioral feature in the content determination method provided in this application. The function of this model corresponds to the aforementioned formula (1), which can calculate the behavioral feature weight coefficients corresponding to each user's behavioral features. Specifically, multiple-dimensional attribute feature vectors can be weighted and pooled, that is, the resource attribute feature vector, channel attribute feature vector, and associated content attribute feature vector are weighted and pooled based on the attribute weight coefficients corresponding to each resource attribute feature vector to obtain the weighted and pooled target attribute feature. Then, the target attribute feature and multiple behavioral feature vectors are used to perform vector dot product calculation to obtain the behavioral feature weight coefficients of each behavioral feature for one target candidate advertisement. Here, the behavioral feature weight coefficients are behavioral feature weight coefficients for a target candidate advertisement. Based on the behavioral feature weight coefficients, the behavioral feature vectors are weighted and pooled again to obtain the user's interest features for the target candidate advertisement.

[0156] The following data comparison will more clearly demonstrate the performance improvement of the model provided in this application compared to the models in related technologies. Table 1 below shows a performance comparison table of the target model provided in this application and several models in related technologies.

[0157]

[0158] Table 1: Model Performance Comparison Table

[0159] Table 1 shows the performance parameters of various models running on public datasets. The weighted pooling model is a traditional pooling method that directly weights and pools at least one user behavioral feature to obtain the user's interest features for each candidate advertisement. The Deep Interest Network for Click-Through Rate Prediction (DIN) is a user interest modeling model that considers the impact of different target advertisements on user behavior modeling and assigns different weights to different historical user behaviors. Specifically, when calculating the weight of each user's historical behavior, the behavioral feature vector of that historical behavior and the feature vector of the target advertisement are input into the activation unit to calculate the correlation between each behavior and the target advertisement, and the weight corresponding to that behavior is output through the neural network. The higher the correlation, the greater the weight. The user's historical behaviors are then weighted and summed to obtain the user's interest features for the target advertisement. The Deep Interest Evolution Network for Click-Through Rate Prediction (DIEN) is a user interest evolution model. DIEN first models user historical behavior through a Gate Recurrent Unit (GRU) network, thereby capturing the temporal relationships between user historical behaviors. Then, similar to DIN, the correlation between each user's historical behavior and the target advertisement is calculated to obtain the weight of the user's historical behavior. The Deep Session Interest Network for Click-Through Rate Prediction (DSIN) is a user session interest modeling model. DISN first divides the user's historical behavior into multiple sessions according to time windows, and uses Transformer and Long Short-Term Memory (LSTM) units to capture the interest relationships within and between sessions.

[0160] As shown in Table 1, the target model provided in this application exhibits a smaller loss and a higher Area Under Curve (AUC) value compared to other models. Furthermore, compared to the ordinary weighted pooling model that weights user behavior features, the target model provided in this application shows the highest AUC improvement. AUC, or the area under the Receiver Operating Characteristic Curve (ROC), is an important indicator for evaluating model performance. A higher AUC value indicates better model performance.

[0161] like Figure 5 The figure illustrates the impact of a pruning scheme based on certain weighting coefficients on the attribute feature vectors of advertisements in the target model provided in this application. The figure also shows a line graph validating the impact of the pruning rate on model performance on a preset public dataset. As the pruning rate for each attribute feature vector increases, the model's AUC gradually increases. After reaching a preset pruning rate, further increases in the pruning rate lead to a decrease in model performance. In the dataset used in the example, the model performs best when the pruning rate is 0.6. Of course, the dataset used in this example is only one example dataset; different pruning rates may yield the best model performance on other datasets.

[0162] like Figure 6 The image shows a heatmap of the attribute features learned by the model in this application. The intensity of the color represents the weight coefficient of each attribute feature; darker colors correspond to higher weight coefficients. As shown, advertising resource information, placement channel information, and related content information have the highest weight coefficients, while other attribute features have lower weight coefficients. Therefore, it can be determined that advertising resource information, advertising placement channel information, and related content information have larger weight coefficients, while advertising group information, user identity information, and brand information have the next highest weight coefficients, and content management system information and city level information have the lowest weight coefficients. Alternatively, when selecting attribute features, only the attribute features with higher weight coefficients can be selected for calculation.

[0163] in, Figure 6 The heatmaps corresponding to the attribute data shown are trained based on a common dataset. Different datasets may yield different results. That is, the heatmaps corresponding to the attribute data trained on different common datasets may be different, which means that the weight coefficients corresponding to each attribute feature may also be different.

[0164] As described above, the content determination method provided in this application involves a computer device acquiring at least one behavioral feature corresponding to the historical behavioral data of a target object; acquiring at least two attribute features corresponding to each of at least two candidate contents, wherein the attribute features are the features corresponding to the attribute data of the candidate contents; calculating the target object's degree of interest in each candidate content based on the behavioral features and attribute features; and determining the target candidate content based on the degree of interest.

[0165] Therefore, the content determination method provided in this application can bidirectionally calculate the user's level of interest in each candidate content based on the user's historical behavioral characteristics and the multi-dimensional attribute characteristics of the candidate content, and further determine the target content to recommend to the user based on the level of interest. Because the multi-dimensional attribute characteristics of the candidate content are considered when calculating the user's level of interest, the calculated level of interest is more accurate, thereby improving the accuracy of determining the target content and consequently improving the accuracy of the content recommended to the user.

[0166] To better implement the above methods, embodiments of this application also provide a content determination device, which can be integrated into a terminal or server.

[0167] For example, such as Figure 7 The diagram shown is a structural schematic of a content determination device provided in an embodiment of this application. The content determination device may include a first acquisition unit 301, a second acquisition unit 302, a calculation unit 303, and a determination unit 304, as follows:

[0168] The first acquisition unit 301 is used to acquire at least one behavioral feature corresponding to the historical behavioral data of the target object;

[0169] The second acquisition unit 302 is used to acquire at least two dimensions of attribute features corresponding to each candidate content in at least two candidate contents, wherein the attribute features are the features corresponding to the attribute data of the candidate content.

[0170] The calculation unit 303 is used to calculate the target object's interest in each candidate content based on behavioral and attribute characteristics.

[0171] The determination unit 304 is used to determine the target candidate content based on the degree of interest.

[0172] In some embodiments, the computing unit includes:

[0173] The computational subunit is used to calculate the first weight coefficient for each behavioral feature based on behavioral features and attribute features;

[0174] The processing subunit is used to weight at least one behavioral feature according to the first weight coefficient of each behavioral feature to obtain the interest feature corresponding to each candidate content.

[0175] The sub-unit is determined to determine the target object's level of interest in each candidate content based on interest features.

[0176] In some embodiments, the computing subunit includes:

[0177] The first acquisition module is used to acquire the second weight coefficient corresponding to each attribute feature;

[0178] The calculation module is used to calculate the first weight coefficient of each behavioral feature based on the behavioral features, attribute features, and the second weight coefficient.

[0179] In some embodiments, the acquisition module includes:

[0180] The extraction submodule is used to extract the target attribute data associated with each attribute data contained in each historical behavior data from the historical behavior data.

[0181] The determination submodule is used to determine the second weight coefficient corresponding to each attribute data based on the target attribute data. Each attribute data is each of the at least two attribute data corresponding to each candidate content.

[0182] In some embodiments, the computing module includes:

[0183] The acquisition submodule is used to acquire the behavior feature vector corresponding to each behavior feature and the attribute feature vector corresponding to each attribute feature.

[0184] The processing submodule is used to perform element-level multiplication on any target behavior feature and each attribute feature, and to perform weighted processing on the processing result according to the second weight coefficient, and to perform weighted processing on the target behavior feature according to the first weight coefficient corresponding to the weighted processing result.

[0185] The calculation submodule is used to traverse each behavior feature based on the calculation method of the first weight coefficient corresponding to the target behavior feature, and calculate the first weight coefficient of each behavior feature.

[0186] In some embodiments, determining the subunit includes:

[0187] The second acquisition module is used to acquire the trained interest level prediction model;

[0188] The prediction module is used to input interest features into the interest level prediction model to obtain the target object's interest level for each candidate content.

[0189] In some embodiments, the content determination apparatus provided in this application further includes:

[0190] The third acquisition unit is used to acquire target attribute data of at least one dimension of the target candidate content;

[0191] The recommendation unit is used to recommend target candidate content to target users based on target attribute data.

[0192] In some embodiments, the first acquisition unit includes:

[0193] The first acquisition subunit is used to acquire at least one historical behavior data of the target object within a preset time period;

[0194] The first mapping subunit is used to map historical behavior data to a vector space to obtain at least one behavioral feature of the target object.

[0195] In some embodiments, the second acquisition unit includes:

[0196] The second acquisition subunit is used to acquire resource attribute data, channel attribute data and associated content attribute data for each of at least two candidate contents;

[0197] The second mapping subunit is used to map resource attribute data, channel attribute data and associated content attribute data to the vector space respectively, so as to obtain the resource attribute features, channel attribute features and associated content attribute features corresponding to each candidate content.

[0198] The computing unit is also used for:

[0199] The degree of interest of the target object in each candidate content is calculated based on behavioral characteristics, resource attribute characteristics, channel attribute characteristics, and related content attribute characteristics.

[0200] In practice, each of the above units can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units, please refer to the previous method embodiments, which will not be repeated here.

[0201] As described above, the content determination device provided in this application embodiment acquires at least one behavioral feature corresponding to the historical behavioral data of the target object through a first acquisition unit 301; a second acquisition unit 302 acquires at least two attribute features corresponding to each of the at least two candidate contents, wherein the attribute features are the features corresponding to the attribute data of the candidate contents; a calculation unit 303 calculates the degree of interest of the target object to each candidate content based on the behavioral features and attribute features; and a determination unit 304 determines the target candidate content based on the degree of interest.

[0202] Therefore, the content determination method provided in this application can bidirectionally calculate the user's level of interest in each candidate content based on the user's historical behavioral characteristics and the multi-dimensional attribute characteristics of the candidate content, and further determine the target content to recommend to the user based on the level of interest. Because the multi-dimensional attribute characteristics of the candidate content are considered when calculating the user's level of interest, the calculated level of interest is more accurate, thereby improving the accuracy of determining the target content and consequently improving the accuracy of the content recommended to the user.

[0203] This application also provides a computer device, which can be a terminal or a server, such as... Figure 8 The diagram shown is a structural schematic of the computer device provided in this application. Specifically:

[0204] The computer device may include components such as a processing unit 401 with one or more processing cores, a storage unit 402 with one or more storage media, a power module 403, and an input module 404. Those skilled in the art will understand that... Figure 8 The computer device structure shown does not constitute a limitation on the computer device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:

[0205] The processing unit 401 is the control center of the computer device. It connects various parts of the computer device via various interfaces and lines. By running or executing software programs and / or modules stored in the storage unit 402, and by calling data stored in the storage unit 402, it performs various functions of the computer device and processes data, thereby performing overall detection of the computer device. Optionally, the processing unit 401 may include one or more processing cores; preferably, the processing unit 401 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processing unit 401.

[0206] Storage unit 402 can be used to store software programs and modules. Processing unit 401 executes various functional applications and data processing by running the software programs and modules stored in storage unit 402. Storage unit 402 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback, image playback, and web page access), etc.; the data storage area may store data created based on the use of the computer device. In addition, storage unit 402 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, storage unit 402 may also include a memory controller to provide processing unit 401 with access to storage unit 402.

[0207] The computer equipment also includes a power supply module 403 that supplies power to various components. Preferably, the power supply module 403 can be logically connected to the processing unit 401 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply module 403 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0208] The computer device may also include an input module 404, which can be used to receive input numeric or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0209] Although not shown, the computer device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processing unit 401 in the computer device loads the executable files corresponding to the processes of one or more applications into the storage unit 402 according to the following instructions, and the processing unit 401 runs the applications stored in the storage unit 402 to realize various functions, as follows:

[0210] Obtain at least one behavioral feature corresponding to the historical behavioral data of the target object; obtain at least two attribute features corresponding to each of the at least two candidate contents, wherein the attribute features are the features corresponding to the attribute data of the candidate contents; calculate the target object's degree of interest in each candidate content based on the behavioral features and attribute features; determine the target candidate content based on the degree of interest.

[0211] It should be noted that the computer device provided in this application embodiment and the method in the above embodiment belong to the same concept. The specific implementation of each of the above operations can be found in the previous embodiments, and will not be repeated here.

[0212] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0213] Therefore, embodiments of the present invention provide a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute steps in any of the methods provided in the embodiments of the present invention. For example, the instructions can execute the following steps:

[0214] Obtain at least one behavioral feature corresponding to the historical behavioral data of the target object; obtain at least two attribute features corresponding to each of the at least two candidate contents, wherein the attribute features are the features corresponding to the attribute data of the candidate contents; calculate the target object's degree of interest in each candidate content based on the behavioral features and attribute features; determine the target candidate content based on the degree of interest.

[0215] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0216] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0217] Since the instructions stored in the computer-readable storage medium can execute the steps of any of the methods provided in the embodiments of the present invention, the beneficial effects that any of the methods provided in the embodiments of the present invention can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0218] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a storage medium. A processor of a computer device reads the computer instructions from the storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various optional implementations of the above-described content determination method.

[0219] The content determination method, apparatus, storage medium, and computer equipment provided in the embodiments of the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for determining content, characterized in that, The method includes: Obtain at least one behavioral feature corresponding to the historical behavioral data of the target object; Obtain at least two dimensions of attribute features corresponding to each of at least two candidate contents, wherein the attribute features are the features corresponding to the attribute data of the candidate contents; Calculating the target object's interest level in each candidate content based on the behavioral features and the at least two-dimensional attribute features includes: obtaining a second weight coefficient corresponding to each attribute feature in the at least two-dimensional attribute features; obtaining a behavioral feature vector corresponding to each behavioral feature and an attribute feature vector corresponding to each attribute feature; performing element-level multiplication on any target behavioral feature and each attribute feature, and weighting the processing result according to the second weight coefficient, and determining a first weight coefficient corresponding to the target behavioral feature based on the weighted processing result; traversing each behavioral feature based on the calculation method of the first weight coefficient corresponding to the target behavioral feature to calculate the first weight coefficient of each behavioral feature; weighting the at least one behavioral feature according to the first weight coefficient of each behavioral feature to obtain an interest feature corresponding to each candidate content; and determining the target object's interest level in each candidate content based on the interest feature. Target candidate content is determined based on the level of interest.

2. The method according to claim 1, characterized in that, The step of obtaining the second weight coefficient corresponding to each attribute feature in the at least two dimensions of attribute features includes: Extract the target attribute data associated with each attribute data contained in each historical behavior data from the historical behavior data. The second weight coefficient corresponding to each attribute data is determined based on the target attribute data, wherein each attribute data is each of the at least two attribute data corresponding to each candidate content.

3. The method according to claim 1, characterized in that, Determining the target object's level of interest in each candidate content based on the interest features includes: Obtain the trained interest prediction model; The interest features are input into the interest prediction model to obtain the target object's interest level for each candidate content.

4. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Obtain target attribute data in at least two dimensions of the target candidate content; Based on the target attribute data, the target candidate content is recommended to the target user.

5. The method according to claim 1, characterized in that, The acquisition of at least one behavioral feature corresponding to the historical behavioral data of the target object includes: Obtain at least one historical behavior data of the target object within a preset time period; The historical behavior data is mapped to a vector space to obtain at least one behavioral feature of the target object.

6. The method according to claim 1, characterized in that, The step of obtaining at least two dimensions of attribute features corresponding to each of the at least two candidate contents includes: Obtain resource attribute data, channel attribute data, and associated content attribute data for each of at least two candidate contents; The resource attribute data, channel attribute data, and associated content attribute data are mapped to a vector space to obtain the resource attribute features, channel attribute features, and associated content attribute features corresponding to each candidate content. The step of calculating the target object's interest level in each candidate content based on the behavioral characteristics and attribute characteristics includes: The degree of interest of the target object in each candidate content is calculated based on the behavioral characteristics, resource attribute characteristics, channel attribute characteristics, and associated content attribute characteristics.

7. A content determining device, characterized in that, The device includes: The first acquisition unit is used to acquire at least one behavioral feature corresponding to the historical behavioral data of the target object; The second acquisition unit is used to acquire at least two dimensions of attribute features corresponding to each candidate content in at least two candidate contents, wherein the attribute features are features corresponding to the attribute data of the candidate content. A calculation unit is configured to calculate the target object's interest level in each candidate content based on the behavioral features and the attribute features of at least two dimensions, including: obtaining a second weight coefficient corresponding to each attribute feature in the at least two dimensions; obtaining a behavioral feature vector corresponding to each behavioral feature and an attribute feature vector corresponding to each attribute feature; performing element-level multiplication on any target behavioral feature and each attribute feature, and weighting the processing result according to the second weight coefficient, and determining a first weight coefficient corresponding to the target behavioral feature based on the weighted processing result; traversing each behavioral feature based on the calculation method of the first weight coefficient corresponding to the target behavioral feature to calculate the first weight coefficient of each behavioral feature; weighting the at least one behavioral feature according to the first weight coefficient of each behavioral feature to obtain an interest feature corresponding to each candidate content; and determining the target object's interest level in each candidate content based on the interest feature. A determining unit is used to determine target candidate content based on the degree of interest.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to perform the steps of the content determination method according to any one of claims 1 to 6.

9. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the content determination method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps in the content determination method according to any one of claims 1 to 6.

Images