CN116204697A

CN116204697A - Content determination method, content determination device, computer readable storage medium and computer device

Info

Publication number: CN116204697A
Application number: CN202111445120.9A
Authority: CN
Inventors: 高晓沨; 郑作武; 潘军伟; 刘文博; 杨召唤; 李晓波
Original assignee: Shanghai Jiaotong University; Tencent Technology Shenzhen Co Ltd
Current assignee: Shanghai Jiaotong University; Tencent Technology Shenzhen Co Ltd
Priority date: 2021-11-30
Filing date: 2021-11-30
Publication date: 2023-06-02

Abstract

The application discloses a content determination method, a content determination device, a computer readable storage medium and computer equipment, wherein the method comprises the steps of obtaining at least one behavior characteristic corresponding to historical behavior data of a target object; acquiring attribute characteristics of at least two dimensions corresponding to each candidate content in at least two candidate contents, wherein the attribute characteristics are characteristics corresponding to attribute data of the candidate contents; calculating the interest degree of the target object on each candidate content according to the behavior characteristics and the attribute characteristics; and determining target candidate content according to the interest degree. The content determining method provided by the application can improve the accuracy of content determination, and further can improve the accuracy of content data recommended by a recommendation system for a user.

Description

Content determination method, content determination device, computer readable storage medium and computer device

Technical Field

The present invention relates to the field of content recommendation technologies, and in particular, to a content determining method, a content determining device, a computer readable storage medium, and a computer device.

Background

With the continuous development of internet technology, people's life is already indistinct from the internet. In the internet age, with the rapid expansion of content, the pressure of content selection faced by people is increased, which reduces the use efficiency of content by people, thereby causing the problem of information overload.

The recommendation system is a personalized recommendation system for recommending information, products and the like which are interested by a user to the user according to the content requirements, interests and the like of the user. A good recommendation system not only can provide personalized content and service for users, but also can establish close relation with users.

However, in some cases, such as in an internet advertisement recommendation scenario, the content recommended by the recommendation system for the user is not accurate enough.

Disclosure of Invention

The embodiment of the application provides a content determining method, a content determining device, a computer readable storage medium and computer equipment.

A first aspect of the present application provides a content determining method, the method including:

acquiring at least one behavior feature corresponding to historical behavior data of a target object;

acquiring attribute characteristics of at least two dimensions corresponding to each candidate content in at least two candidate contents, wherein the attribute characteristics are characteristics corresponding to attribute data of the candidate contents;

calculating the interest degree of the target object on each candidate content according to the behavior characteristics and the attribute characteristics;

And determining target candidate content according to the interest degree.

A second aspect of the present application provides a content determining apparatus, the apparatus comprising:

the first acquisition unit is used for acquiring at least one behavior characteristic corresponding to the historical behavior data of the target object;

the second acquisition unit is used for acquiring attribute characteristics of at least two dimensions corresponding to each candidate content in at least two candidate contents, wherein the attribute characteristics are characteristics corresponding to attribute data of the candidate content;

a calculating unit, configured to calculate a degree of interest of the target object for each candidate content according to the behavior feature and the attribute feature;

and the determining unit is used for determining target candidate contents according to the interest degree.

In some embodiments, the computing unit comprises:

a calculating subunit, configured to calculate a first weight coefficient of each behavioral feature based on the behavioral feature and the attribute feature;

the processing subunit is used for weighting the at least one behavior characteristic according to the first weight coefficient of each behavior characteristic to obtain an interest characteristic corresponding to each candidate content;

a determining subunit, configured to determine, based on the interest feature, a degree of interest of the target object for each candidate content.

In some embodiments, the computing subunit comprises:

the first acquisition module is used for acquiring a second weight coefficient corresponding to each attribute characteristic;

and the calculation module is used for calculating a first weight coefficient of each behavior feature according to the behavior feature, the attribute feature and the second weight coefficient.

In some embodiments, the acquisition module comprises:

the extraction sub-module is used for extracting target attribute data which is contained in each piece of historical behavior data and is associated with each piece of attribute data from the historical behavior data;

and the determining submodule is used for determining a second weight coefficient corresponding to each attribute data according to the target attribute data, wherein each attribute data is each attribute data in at least two attribute data corresponding to each candidate content.

In some embodiments, the computing module comprises:

the acquisition sub-module is used for acquiring a behavior feature vector corresponding to each behavior feature and an attribute feature vector corresponding to each attribute feature;

the processing sub-module is used for carrying out element level product processing on any target behavior feature and each attribute feature, carrying out weighting processing on a processing result according to the second weighting coefficient, and carrying out first weighting coefficient corresponding to the target behavior feature according to the weighting processing result;

And the computing sub-module is used for traversing each behavior feature based on the computing method of the first weight coefficient corresponding to the target behavior feature and computing to obtain the first weight coefficient of each behavior feature.

In some embodiments, the determining subunit comprises:

the second acquisition module is used for acquiring the trained interested degree prediction model;

and the prediction module is used for inputting the interest characteristic into the interest degree prediction model to obtain the interest degree of the target object on each candidate content.

In some embodiments, the content determining apparatus provided in the embodiments of the present application further includes:

a third obtaining unit, configured to obtain target attribute data of at least one dimension of the target candidate content;

and the recommending unit is used for recommending the target candidate content to the target user based on the target attribute data.

In some embodiments, the first acquisition unit includes:

the first acquisition subunit is used for acquiring at least one historical behavior data of the target object in a preset time period;

and the first mapping subunit is used for mapping the historical behavior data to a vector space to obtain at least one behavior characteristic of the target object.

In some embodiments, the second acquisition unit includes:

the second acquisition subunit is used for acquiring resource attribute data, channel attribute data and associated content attribute data corresponding to each candidate content in at least two candidate contents;

the second mapping subunit is used for mapping the resource attribute data, the channel attribute data and the associated content attribute data to a vector space respectively to obtain resource attribute features, channel attribute features and associated content attribute features corresponding to each candidate content;

the computing unit is further configured to:

and calculating the interest degree of the target object on each candidate content according to the behavior characteristic, the resource attribute characteristic, the channel attribute characteristic and the associated content attribute characteristic.

The third aspect of the present application also provides a computer readable storage medium storing a plurality of instructions adapted to be loaded by a processor for performing the steps in the content determining method provided in the first aspect of the present application.

A fourth aspect of the present application provides a computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps in the content determination method provided in the first aspect of the present application when the computer program is executed.

A fifth aspect of the present application provides a computer program product comprising computer programs/instructions which when executed by a processor implement the steps in the content determination method provided in the first aspect.

According to the content determination method, at least one behavior feature corresponding to historical behavior data of a target object is obtained; acquiring attribute characteristics of at least two dimensions corresponding to each candidate content in at least two candidate contents, wherein the attribute characteristics are characteristics corresponding to attribute data of the candidate contents; calculating the interest degree of the target object on each candidate content according to the behavior characteristics and the attribute characteristics; and determining target candidate content according to the interest degree.

Therefore, the content determining method provided by the application can be used for bidirectionally calculating the interest degree of the user on each candidate content according to the historical behavior characteristics of the user and the attribute characteristics of multiple dimensions of the candidate content, and further determining the target content recommended to the user according to the interest degree. When the interest degree of the user on the candidate content is calculated, attribute characteristics of multiple dimensions of the candidate content are considered, so that the calculated interest degree is more accurate, the accuracy of determining the target content is improved, and the accuracy of recommending the content for the user can be improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments will be briefly introduced below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic illustration of a scenario for content determination in the present application;

FIG. 2 is a flow chart of a content determination method provided herein;

FIG. 3 is another flow chart of the content determination method provided herein;

FIG. 4 is a schematic diagram of a behavioral characteristic weight coefficient evaluation model provided in the present application;

FIG. 5 is a schematic view of pruning the attribute feature vector in the present application;

FIG. 6 is a thermodynamic diagram of dimension attribute data for targeted advertising content in the present application;

fig. 7 is a schematic structural view of the content determining apparatus provided in the present application;

fig. 8 is a schematic structural diagram of a computer device provided in the present application.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to fall within the scope of the invention.

The embodiment of the invention provides a content determining method, a content determining device, a computer readable storage medium and a terminal. Wherein the content determining method can be used in a content determining apparatus. The content determining means may be integrated in a computer device, which may be a terminal or a server. The terminal can be a mobile phone, a tablet computer, a notebook computer, an intelligent television, a wearable intelligent device, a personal computer (PC, personal Computer), a vehicle-mounted terminal and other devices. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, network acceleration services (Content Delivery Network, CDN), basic cloud computing services such as big data and artificial intelligent platforms, and the like. Wherein the server may be a node in a blockchain.

Referring to fig. 1, a schematic diagram of a scenario of a content determining method provided in the present application is shown. As shown, a computer device a obtains at least one behavior feature corresponding to historical behavior data of a target object; acquiring attribute characteristics of at least two dimensions corresponding to each candidate content in at least two candidate contents, wherein the attribute characteristics are characteristics corresponding to attribute data of the candidate contents; calculating the interest degree of the target object on each candidate content according to the behavior characteristics and the attribute characteristics; and determining target candidate content according to the interest degree.

It should be noted that the schematic view of the content determination scenario shown in fig. 1 is only an example, and the content determination scenario described in the embodiment of the present application is for more clearly describing the technical solution of the present application, and does not constitute a limitation of the technical solution provided in the present application. As one of ordinary skill in the art can know, as the content determines the scene evolution and the appearance of the new service scene, the technical scheme provided by the application is also applicable to similar technical problems.

The following describes the above-described embodiments in detail.

In the related art, when recommending content to a user, a recommendation system may acquire historical behavior data of the user and candidate content data, calculate a degree of interest of the user in each candidate content data, and recommend content to the user based on the degree of interest. However, at present, when the interest degree of the user on the candidate content is calculated, only the content data itself or the attribute of a single dimension of the content data is generally considered, so that the calculation accuracy of the interest degree of the user on the candidate content is poor, and the accuracy of the content recommended to the user by the recommendation system is low. In this regard, the content determining method provided by the present application may calculate the interest degree of the user for each candidate content based on the multidimensional attribute features of the candidate content, so that the calculated interest degree of the user for the candidate content is more accurate, and the accuracy of the content recommended to the user by the recommendation system may be improved.

The embodiments of the present application will be described from the perspective of a content determination apparatus that may be integrated in a computer device. The computer device may be a terminal or a server. The terminal can be a mobile phone, a tablet computer, a notebook computer, an intelligent television, a wearable intelligent device, a personal computer (PC, personal Computer), a vehicle-mounted terminal and other devices. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, network acceleration services (Content Delivery Network, CDN), basic cloud computing services such as big data and artificial intelligent platforms, and the like. As shown in fig. 2, a flow chart of a content determining method provided in the present application includes:

step 101, at least one behavior feature corresponding to the historical behavior data of the target object is obtained.

The target object may be a specific user, an identity (Identity Document, ID) of the user, or a virtual artificial intelligence object.

The historical behavior data of the target object may be data corresponding to historical operation behavior of the target object on the content data. The content data may be content provided by various applications, such as content provided by news information applications, content provided by short video applications, content provided by an e-commerce platform, content provided by an instant messaging application, and the like. The content data may include text information content, video information content, short video content, merchandise content, online advertising content, and the like, in particular. The historical operation behavior of the target object on the content data can be click behavior, reading behavior, viewing behavior, praise behavior, collection behavior, registration behavior, purchase behavior and the like.

In order to calculate the interest degree of the target object for each candidate content by adopting the historical behaviors of the target object, and then recommend the matched target candidate content for the target object according to the interest degree, the historical behaviors of the target object need to be converted into feature data which can be processed by a recommendation system, and the behavior feature corresponding to each historical behavior data is obtained. The historical behavior data of the target object can be one or a plurality of historical behavior data, and each historical behavior data corresponds to one behavior characteristic.

In some embodiments, obtaining at least one behavioral characteristic corresponding to historical behavioral data of the target object includes:

1. acquiring at least one historical behavior data of a target object in a preset time period;

2. and mapping the historical behavior data to a vector space to obtain at least one behavior characteristic of the target object.

In this embodiment of the present application, at least one historical behavior data of the target object in a preset time period may be obtained, and then each historical behavior data is mapped into a vector space to obtain at least one behavior feature of the target object. The historical behavior data can be converted into vector data which can be operated by adopting an Embedding (Embedding) method. Wherein, embedding is a term in the field of deep learning, specifically refers to mapping high-dimensional original data (e.g., images, sentences, etc.) to low-latitude data, so that the high-dimensional original data becomes practically computable. Specifically, for example, word Embedding (Word Embedding), is to map a sentence composed of words to a token vector. When the historical behavior data of the user is expressed in a text form, the text can be mapped to a vector space, and behavior characteristics corresponding to the historical behavior data are obtained.

Step 102, obtaining attribute features of at least two dimensions corresponding to each candidate content in at least two candidate contents.

In determining recommended content data for a target object, it is generally necessary to determine content data that best matches the target object or content data that is most interesting to the target object from a content library. Wherein the content library may comprise a plurality of candidate content, e.g. at least two candidate content. Specifically, when the candidate content is advertisement content, attribute features of at least two dimensions of each candidate advertisement in the advertisement library may be acquired.

In the related art, for example, in the click rate predicted deep interest network model (Deep Interest Network for Click-Through Rate Prediction, DIN) and the click rate predicted deep interest evolution network model (Deep Interest Evolution Network for Click-Through Rate Prediction), when calculating the correlation between the target advertisement and the user historical behavior, only attribute features of a single dimension of the target advertisement, such as a single feature of an advertiser identity (Identity document, ID) or an advertisement ID, are often considered, and the influence of other features on the user behavior is ignored. Especially in the channel dimension and associated content dimension.

In the embodiment of the present application, the attribute features of at least two dimensions corresponding to each candidate content may be features corresponding to attribute data of at least two dimensions corresponding to each candidate content. Specifically, the attribute data of at least two dimensions may include resource attribute data, channel attribute data, associated content attribute data, and the like of the candidate content. Of course, the attribute data of the candidate content may also include attribute data of other dimensions, such as attribute data of a time dimension, etc., where the resource attribute data, channel attribute data, and associated content attribute data are merely examples.

Specifically, still taking the candidate content as an example of the advertisement content, the resource attribute data of the advertisement content may include advertiser ID data, advertisement ID data, and advertisement content data; the channel attribute data of the advertisement content can be the delivery channel data of the advertisement content, such as delivering advertisements on public numbers, delivering advertisements on friend circles, and the like; the associated content attribute data of the advertisement content may be context data of the advertisement, for example, user dynamic data before and after the advertisement when the advertisement is put in a friend circle, advertisement data, or text data before and after the advertisement when the advertisement is put in a public number, and the like, which may be associated content attribute data of the advertisement. For the same user, advertisements are published in different delivery channels, with different probabilities of being able to be clicked or converted. For example, some users often use a circle of friends with little view to the public number content, and then advertising in the circle of friends may result in a higher probability of clicking or translating relative to advertising in the public number. In addition, for the same user, even if the same delivery channel is adopted for advertisement release, the advertisement release position can also influence the probability of clicking the advertisement by the user; specifically, for example, advertising basketball shoes on a basketball game video-gathering public number may result in a greater probability of being clicked or converted than advertising basketball shoes on a food promoting public number. That is, the corresponding characteristics of the advertisement delivery channel and the associated content of the advertisement delivery position can also influence the behavior of the user.

When the influence of the advertisement characteristics on the user behavior is considered, the advertisement characteristics, namely the advertisement owner ID, the advertisement ID and the like, are considered, and the characteristics corresponding to the advertisement delivery channel, the characteristics corresponding to the associated content of the advertisement and other characteristics related to the advertisement which are not listed are considered. The characteristics of the advertisement are extracted from multiple dimensions, and the relationship between the user behavior and the advertisement characteristics is described in multiple dimensions, so that the predicted interest degree of the user on the advertisement is more accurate.

And step 103, calculating the interest degree of the target object in each candidate content according to the behavior characteristics and the attribute characteristics.

Wherein, the behavior feature is at least one behavior feature of the target object, and the attribute feature is at least two dimension attribute feature corresponding to each candidate content in the at least two candidate contents. In the related art, the interest degree of the target object for each candidate content is often calculated only according to the behavior feature of the target object and the attribute feature of one dimension of the candidate content. For example, when calculating the interest level of the target object in each advertisement content, only the behavior feature of the target object and the resource attribute feature of the advertisement are often considered, that is, only the corresponding attribute features of the advertiser data, the advertisement ID data, and the advertisement content data are considered to calculate the interest level of the target object in each candidate advertisement content, and the attribute features of other dimensions such as the channel attribute data, the associated content attribute data, and the like are not considered.

In the embodiment of the application, the influence of the multi-dimensional attribute characteristics of the candidate contents on the interest degree is comprehensively considered, so that the accuracy of the calculated interest degree of the target object on each candidate content can be effectively improved, and the accuracy of determining the target candidate content can be further improved.

In some embodiments, calculating the interest level of the target object in each candidate content according to the behavior feature and the attribute feature of at least two dimensions comprises:

1. calculating a first weight coefficient of each behavioral feature based on the behavioral features and the attribute features of at least two dimensions;

2. weighting at least one behavior feature according to the first weight coefficient of each behavior feature to obtain an interest feature corresponding to each candidate content;

3. the degree of interest of the target object for each candidate content is determined based on the interest features.

Wherein, here, the first weight coefficient of each behavior feature is the weight coefficient of each behavior feature corresponding to each target candidate content. For example, when there are three behavior features A, B and C, and three candidate contents a, b, and C, then for the candidate content a, the behavior features A, B and C each have a weight coefficient corresponding to the candidate content a; for candidate content b and candidate content C, the behavior characteristics A, B and C also have a weight coefficient corresponding to the candidate content, respectively. To distinguish from the subsequent other weight coefficients, it may be referred to herein as a first weight coefficient.

The first weight coefficient of the behavior feature is a contribution degree of each behavior feature to a degree of interest, where the degree of interest is a degree of interest of the target object to the target content. In general, the higher the first weight coefficient of a behavioral feature, the higher the contribution of that behavioral feature to the degree of interest. Specifically, for example, when the behavior feature a is a feature corresponding to a click behavior of the candidate content a, the behavior feature B is a feature corresponding to a praise behavior of the candidate content a, and the behavior feature C is a payment behavior of the candidate content a, the weight coefficient of the behavior feature a is smaller than that of the behavior feature B, and the weight coefficient of the behavior feature B is smaller than that of the behavior feature C.

In the related art, a conventional click rate prediction model often performs weighted pooling or average pooling processing on historical behavior characteristics of a user. However, in practice, when different historical behaviors of the user predict clicks or conversions of different advertisements, the generated influences are different, and if only the historical behavior features of the user are simply weighted and pooled or averaged and pooled, the association relationship between the different historical behaviors of the user and the interest degree of the user in the advertisements cannot be represented, so that the predicted interest degree of the user in the advertisements is inaccurate. Therefore, in the embodiment of the application, the first weight coefficient of each behavior feature can be calculated firstly based on the behavior feature and the attribute features of at least two dimensions, so that the weight coefficient corresponding to each behavior feature is determined, the influence of different behavior features on the click rate of the target content by the user is differentiated and accurate, and the accuracy of the predicted interest degree of the user on the advertisement can be improved.

In the embodiment of the application, the first weight coefficient of each behavior feature corresponding to each candidate content can be obtained through calculation through the behavior feature and attribute features of a plurality of dimensions corresponding to each candidate content. And then, weighting and summing a plurality of behavior features corresponding to each candidate content by using a first weight coefficient corresponding to each behavior feature to obtain the interest feature corresponding to each candidate content. The interest feature is a feature that characterizes the interest degree of the target object in the candidate content, and may be a feature vector.

Further, a degree of interest of the target object for each candidate content may be determined based on the interest feature, and the degree of interest may be an interest value.

In some embodiments, calculating the first weight coefficient for each behavioral feature from the behavioral feature and the attribute features of the at least two dimensions includes:

1.1, acquiring a second weight coefficient corresponding to each attribute feature in attribute features of at least two dimensions;

and 1.2, calculating a first weight coefficient of each behavior feature according to the behavior feature, the attribute features of at least two dimensions and the second weight coefficient.

Wherein the attribute features of at least two dimensions are the attribute features of at least two dimensions corresponding to each candidate content. In the embodiment of the application, the contribution of each attribute feature to the interest degree may also be different due to the attribute features of multiple dimensions corresponding to each candidate content. Therefore, to further improve accuracy of calculation of the interest degree and further improve accuracy of determination of the target candidate content, a weight coefficient of each attribute feature corresponding to each content may be obtained first, where the weight coefficient is distinguished from the weight coefficient of the foregoing behavioral feature, and may be referred to as a second weight coefficient. The second weight coefficient characterizes the degree of influence, or importance, of the attribute features of each dimension on the user behavior. After the second weight coefficient of each attribute feature corresponding to each candidate content is obtained, the first weight coefficient of each behavior feature in the plurality of behavior features corresponding to each candidate content is obtained through calculation according to the attribute features of the plurality of dimensions corresponding to each candidate content, the second weight coefficient corresponding to each attribute feature and the plurality of behavior features of the target object.

In some embodiments, obtaining a second weight coefficient corresponding to each of the attribute features of the at least two dimensions includes:

1.1.1, extracting target attribute data which are contained in each piece of historical behavior data and are associated with each piece of attribute data from the historical behavior data;

1.1.2, determining a second weight coefficient corresponding to each attribute data according to the target attribute data, wherein each attribute data is each attribute data in at least two attribute data corresponding to each candidate content.

In the embodiment of the present application, the target attribute data associated with each attribute data included in each historical behavior data may be extracted from the historical behavior data of the target object. The target attribute data are data related to a behavior scene corresponding to the historical behavior data. For example, when the candidate content is an advertisement of a sports product, the difference of click rates obtained by putting the advertisement into different channels is much larger than the difference of click rates obtained by putting the advertisement into different context content positions, that is, the influence degree of the channel attribute features on the click rate promotion of the advertisement is larger than the influence degree of the associated content attribute features. Then the weighting factor for the channel attribute feature will generally be higher than the weighting factor for the associated content attribute feature for the advertisement. I.e. the influence of different attribute features on the degree of interest of the target object is different, which different influence can be characterized by the second weight coefficient of each attribute feature. In the embodiment of the application, a weight coefficient can be set for the attribute features of each dimension to realize pruning of the attribute features of different dimensions, and even the attribute features with smaller influence can be removed, so that the contribution of the attribute features corresponding to the attribute data of different dimensions to the interested degree accords with the actual situation, and the result of the interested degree of the target to each candidate content obtained through calculation is more accurate, so that the result of the determined target candidate content is more accurate. The weight coefficient is set for the attribute features of each dimension, an initial weight coefficient can be set first, then the initial weight coefficient is continuously iteratively updated in the deep learning process, and the more accurate weight coefficient of each attribute feature is learned.

In some embodiments, historical behavior data of more objects on the target candidate content can be obtained, and then the historical behavior data is marked to obtain training data. The labeling data may be a weight coefficient corresponding to each attribute feature. Further, a preset neural network model can be trained according to the training data, and a trained neural network model is obtained.

Then, after determining the historical behavior data corresponding to each candidate content, the weight coefficient corresponding to each attribute feature of the candidate content can be determined through the neural network model.

In some embodiments, calculating a first weight coefficient for each behavioral feature from the behavioral feature, the attribute features of the at least two dimensions, and the second weight coefficient comprises:

1.2.1, obtaining a behavior feature vector corresponding to each behavior feature and an attribute feature vector corresponding to each attribute feature;

1.2.2, performing element level product processing on any target behavior feature and each attribute feature, performing weighting processing on a processing result according to a second weighting coefficient, and determining a first weighting coefficient corresponding to the target behavior feature according to the weighting processing result;

1.2.3, traversing each behavior feature based on a calculation method of a first weight coefficient corresponding to the target behavior feature, and calculating to obtain the first weight coefficient of each behavior feature.

In the embodiment of the present application, the behavior feature and the attribute feature may be specifically expressed in the form of vectors, that is, a behavior feature vector corresponding to each behavior feature and an attribute feature vector corresponding to each attribute feature in the attribute features of multiple dimensions corresponding to each candidate content may be obtained.

Wherein, for any target behavior feature vector, the element-level product between the target behavior feature vector and a plurality of attribute feature vectors of each candidate content can be calculated. Specifically, for example, a two-dimensional vector c= (c 1, d 1), a two-dimensional vector d= (c 2, d 2), then the element-level product between the calculated vector c and the vector d may be expressed as < c, d > = (c 1 x c2, d1 x d 2). After the element layer products between the target behavior feature vector and each attribute feature vector of the candidate content are obtained through calculation, carrying out weighted pooling processing on the vectors corresponding to the element layer products according to the second weight coefficient corresponding to each attribute feature of any candidate content, and obtaining interaction vectors between the target behavior feature vector and the candidate content. And carrying out normalization (softmax) processing on the interaction vector to obtain a weight coefficient corresponding to the target behavior feature vector and the candidate content. And traversing each behavior feature to obtain a first weight coefficient corresponding to the candidate content of each behavior feature. Wherein the candidate content is any one of at least two candidate contents. And further traversing each candidate content to obtain a first weight coefficient of each behavior characteristic corresponding to each candidate content.

In some embodiments, determining the level of interest of the target object for each candidate content based on the features of interest includes:

3.1, acquiring a trained interested degree prediction model;

and 3.2, inputting the interest characteristic into an interest degree prediction model to obtain the interest degree of the target object on each candidate content.

In this embodiment of the present application, after determining the interest feature of the target object for each candidate content, the interest feature may be input into a preset interest degree prediction model for processing, so as to obtain the interest degree output by the interest degree prediction model. The interest level may be a value of interest and the interest level prediction model may be a Deep click rate assessment model (Deep CTR model).

In some embodiments, obtaining attribute features of at least two dimensions corresponding to each of at least two candidate content includes:

A. acquiring resource attribute data, channel attribute data and associated content attribute data corresponding to each candidate content in at least two candidate contents;

B. mapping the resource attribute data, the channel attribute data and the associated content attribute data to a vector space respectively to obtain resource attribute features, channel attribute features and associated content attribute features corresponding to each candidate content;

Calculating the interest degree of the target object in each candidate content according to the behavior characteristics and the attribute characteristics, wherein the method comprises the following steps:

C. and calculating according to the behavior characteristics, the resource attribute characteristics, the channel attribute characteristics and the associated content attribute characteristics to obtain the interest degree of the target object on each candidate content.

In this embodiment of the present application, the attribute features of at least two dimensions corresponding to each candidate content may specifically include a resource attribute feature, a channel attribute feature, and an associated content attribute feature. Specifically, the resource attribute data, the channel attribute data and the associated content attribute data corresponding to each candidate content may be obtained first, and then these attribute data are mapped to a vector space to obtain a feature vector corresponding to each attribute data, that is, to obtain feature vectors of at least two dimensions corresponding to each candidate content. Thus, the step of calculating the interest degree of the target object for each candidate content according to the behavior feature and the attribute feature may be to calculate the interest degree corresponding to each candidate content according to the behavior feature, the resource attribute feature, the channel attribute feature and the associated content attribute corresponding to each candidate content.

It is understood that the resource attribute data, the channel attribute data, and the associated content attribute data are only attribute data of relatively important dimensions among attribute data of a plurality of dimensions of the candidate content, and do not represent attribute data of all dimensions of the candidate content. The attribute data of the candidate content may further include content object data, region data where the content object is located, content management system data, and the like.

And 104, determining target candidate content according to the interest degree.

Wherein the target candidate content may be determined according to the degree of interest of the target object in each candidate content. The interest level may be specifically an interest value, and in general, the candidate content with the highest interest level is determined as the target candidate content. And sorting at least two candidate contents according to the height of the interested value, and determining the candidate content with the highest interested value as the target candidate content.

In some embodiments, the content determining method provided in the present application may further include:

1. acquiring target attribute data of at least two dimensions of the target candidate content;

2. recommending the target candidate content to the target user based on the target attribute data.

Wherein, since for each candidate content there is attribute data of its corresponding at least two dimensions. Then the target candidate content of most interest to the target object, which likewise has target attribute data of corresponding at least two dimensions.

The target attribute data of at least two dimensions indicates a resource attribute, a channel attribute, and an associated content attribute of the target candidate content. According to the resource attribute of the target candidate content, resource data corresponding to the target candidate content, such as advertiser information, advertisement ID, advertisement content and the like, can be obtained; according to the channel attribute of the target candidate content, a delivery channel of the target candidate content can be obtained, for example, advertisements are delivered in friend circles or advertisements are delivered in public numbers; the content associated with the target candidate content, such as the context information of the advertisement, can be determined according to the associated content attribute of the target candidate content, so as to determine the delivery position of the advertisement.

Thus, after the resource data, the channel data and the associated content data of the target candidate content are obtained, the target candidate content can be accurately recommended based on the data.

As can be seen from the above description, in the content determining method provided by the embodiment of the present application, at least one behavior feature corresponding to historical behavior data of a target object is obtained; acquiring attribute characteristics of at least two dimensions corresponding to each candidate content in at least two candidate contents, wherein the attribute characteristics are characteristics corresponding to attribute data of the candidate contents; calculating the interest degree of the target object on each candidate content according to the behavior characteristics and the attribute characteristics; and determining target candidate content according to the interest degree.

In addition, the application further adds the proper weight coefficient to each attribute feature corresponding to each candidate content, and can identify and remove irrelevant attribute features, so that the calculated weight coefficient of each behavior feature is more accurate. Therefore, the accuracy of determining the target candidate content can be improved, and the accuracy of recommending the content by the recommendation system can be further improved.

The application also provides a content determining method which can be used in computer equipment, wherein the computer equipment can be a terminal or a server. In this embodiment, a detailed description may be given taking content as an example of advertisement. As shown in fig. 3, another flow chart of the content determining method provided in the present application, the method specifically includes:

In step 201, a computer device obtains historical behavioral characteristics of a target user.

The historical behavior characteristics of the target user are characteristics corresponding to the historical behavior data of the target user. The historical behavior of the user may be clicking, browsing, praying, collecting, inserting coins, ordering, paying, etc. on the content data. In the embodiment of the present application, the content data may specifically be advertisement data. The features corresponding to the historical behavior data can be feature vectors corresponding to the historical behavior data, and particularly, word embedding processing can be carried out on descriptive text of the historical behavior data to obtain H behavior feature vectors v corresponding to each historical behavior of the advertisement by a user _i Wherein H is greater than or equal to 1, H is greater than or equal to i is greater than or equal to 1, v _i Is the ith behavior feature vector of the user.

In step 202, a computer device obtains a multi-dimensional attribute feature for each candidate advertisement in an advertisement library.

The advertisement library may be a preset advertisement library, specifically may be all advertisement resources stored in a preset computer device, and the preset computer device may be a server.

The multi-dimensional attribute feature of each candidate advertisement may be a feature vector corresponding to the multi-dimensional attribute data corresponding to each candidate advertisement. Wherein the attribute data for the plurality of dimensions may specifically include resource attribute data, channel attribute data, and associated content attribute data. Similarly, the word embedding method may be used to convert the multi-dimensional attribute data corresponding to each candidate advertisement into a multi-dimensional attribute feature vector corresponding to each candidate advertisement. Specifically, for any target candidate advertisement, M corresponding attribute feature vectors e can be determined _j Wherein M is greater than or equal to 2, M is greater than or equal to j is greater than or equal to1，e _j And j-th attribute feature vector of the target candidate advertisement.

In step 203, the computer device obtains a first weight coefficient corresponding to each attribute feature of each candidate advertisement.

Wherein, for each attribute feature of each candidate advertisement, the corresponding weight coefficient r can be obtained _F(j) Which may be referred to herein as a first weight coefficient. Where F (j) is attribute data corresponding to the jth attribute feature of the target candidate advertisement, r _F(j) A first weight coefficient for a j-th attribute feature of the targeted candidate advertisement.

A random value can be preset for a first weight coefficient corresponding to each attribute characteristic of the target candidate advertisement, and then pruning is carried out in the model training process, so that the characteristics of irrelevant dimensions are automatically identified and removed.

In step 204, the computer device calculates a second weight coefficient corresponding to each candidate advertisement according to the historical behavior feature, the attribute feature of each candidate advertisement and the corresponding first weight coefficient.

Wherein, after determining the behavior feature vector v corresponding to the historical behavior feature of the user _i Attribute feature vector e corresponding to attribute features of multiple dimensions of target candidate advertisement _j And a first weight coefficient r corresponding to each attribute feature vector _F(j) Then, the following formula (1) can be adopted to calculate and obtain a second weight coefficient of each behavior feature corresponding to the target candidate advertisement:

wherein alpha is _{AutoAttention} (v _i ,e ₀ ,e ₁ ,e ₂ ,…,e _M-1 ) The ith behavioral characteristic vector of the user corresponds to the second weight coefficient of the target candidate advertisement,<v _i ,e _j >for vector v _i And e _j Element-level product processing is performed, and α () is normalization (softmax) processing of the vector in brackets.

Then, each candidate advertisement can be traversed, and a second weight coefficient corresponding to each candidate advertisement of each user characteristic is calculated according to the formula.

In step 205, the computer device performs weighting processing on the historical behavior characteristics of the user according to the second weight coefficient, so as to obtain the interest characteristics of the user for each candidate advertisement.

For the target candidate advertisement, since the second weight coefficient corresponding to each user behavior feature is determined, the second weight coefficient can be used for carrying out weighted pooling on each user behavior feature to obtain the interest feature of the user on the target candidate advertisement. The specific calculation formula is as follows:

wherein v is _u (x) For user u to target candidate advertisement interest feature, v _i Is the ith behavior feature vector of the user, alpha _{AutoAttention} (v _i ,e ₀ ,e ₁ ,e ₂ ,…,e _M-1 ) Can be calculated from the above formula (1).

Likewise, each candidate advertisement may be traversed to derive a user's interest feature v for each candidate advertisement _u 。

In step 206, the computer device inputs the interest feature of the user for each candidate advertisement to a preset interest prediction model, so as to obtain the interest value of the user for each candidate advertisement.

The preset interest prediction model may be a Deep click rate prediction model, i.e., deep CTR model. After feature vectors corresponding to the interest features of the user on each candidate content are obtained, the vectors can be input into a Deep CTR model for prediction, and the interest value of the user on each candidate advertisement is obtained.

In step 207, the computer device determines target candidate advertisements based on the user's value of interest in each candidate advertisement.

Wherein, the value of interest corresponding to each candidate advertisement can be determined, and the candidate advertisements are ordered based on the order of the value of interest from high to low. And further determining the candidate advertisement with the highest interest value as a target candidate advertisement.

In step 208, the computer device recommends the targeted candidate advertisement to the user.

Wherein attribute data for multiple dimensions of the targeted candidate advertisement may be obtained. Specifically, the resource attribute data, the channel attribute data and the associated content attribute data can be included. The resource attribute data may include advertiser information, advertisement ID information, and advertisement content information of the advertisement; the channel attribute data may include channel information for delivering the target candidate advertisement, including web page information, public number information, or friend circle information; the associated content attribute data may include contextual data in the target channel from which the exact placement of the target candidate advertisement may be determined. The attribute information of the multiple dimensions is obtained, and the advertisements can be accurately put according to the multi-dimensional attribute information.

In particular, the scheme can be used for personalized recommendation of advertisements, and online advertisements are the most direct and transparent flow rendering mode for most Internet companies. Taking an information sharing platform of an instant messaging application as an example, when a user opens an information sharing platform refreshing list of the instant messaging application, an advertisement recommendation system receives a refreshing request, obtains historical behavior data of the user and characteristic data of multiple dimensions of multiple candidate advertisements, calculates the interested degree of the user on each candidate advertisement, determines target candidate advertisements according to the interested degree, and then adaptively puts the target candidate advertisements. It will be appreciated that, in the specific embodiments of the present application, related data such as historical behavior data of a user is referred to, and when the embodiments of the present application are applied to specific products or technologies, user permission or consent needs to be obtained, and the collection, use and processing of related data needs to comply with related laws and regulations and standards of related countries and regions.

As shown in fig. 4, a schematic model diagram of a weight coefficient corresponding to each behavioral characteristic is calculated in the content determining method provided in the present application, and the function of the model corresponds to the foregoing formula (1), so that the behavioral characteristic weight coefficient corresponding to each user behavioral characteristic can be calculated. Specifically, the attribute feature vectors of multiple dimensions may be weighted and pooled, that is, the resource attribute feature vector, the channel attribute feature vector and the associated content attribute feature vector are weighted and pooled based on the attribute weight coefficient corresponding to each resource attribute feature vector, so as to obtain the weighted and pooled target attribute feature. And then carrying out vector dot product calculation by adopting the target attribute feature and a plurality of behavior feature vectors to obtain a behavior feature weight coefficient of each behavior feature pair. The behavior feature weight coefficient is a behavior feature weight coefficient for a target candidate advertisement, and the behavior feature vector is weighted and pooled again based on the behavior feature weight coefficient, so that the interest feature of the user for the target candidate advertisement can be obtained.

The improvement in model performance provided herein over the performance of the related art model may be more clearly embodied in a set of data comparisons. As shown in table 1 below, a comparison table of the performance of the object model provided in the present application and several models in the related art.

Model	Loss of	AUC	AUC elevation
				Weighted pooling model	0.1958	0.6312	/
Click rate prediction deep interest network model	0.1956	0.6327	0.24％
				Click rate prediction deep interest evolution network model	0.1955	0.6333	0.33％
Click rate predicted deep session interest network model	0.1955	0.6363	0.81％
				The object model provided by the application	0.1954	0.6372	0.95％

Table 1: model performance comparison table

As shown in table 1, are performance parameters for each type of model running on a common data set. The weighted pooling model is a traditional pooling method, and the method directly carries out weighted pooling on at least one behavior characteristic of the user to obtain the interest characteristic of the user on each candidate advertisement. The click-through rate predicted deep interest network model (Deep interest network for click-through rate prediction, DIN) is a user interest modeling model that takes into account the impact of different targeted advertisements on user behavior modeling and assigns different weights to different historical behaviors of a user. Specifically, when calculating the weight of each user's historical behavior, the behavior feature vector of the historical behavior and the feature vector of the target advertisement are input into the activation unit, the correlation between each behavior and the target advertisement is calculated, and the weight corresponding to the behavior is output through the neural network. The higher the correlation is, the larger the weight is, and the weighted summation is carried out on the historical behaviors of the user to obtain the interest characteristics of the user on the target advertisement. The click rate predicted deep interest evolution network model (Deep interest evolution network for click-through rate prediction, DIEN) is a model of user interest evolution, in which DIEN first models the user's historical behavior through a gated loop unit (Gate Recurrent Unit, GRU) network, capturing the timing relationship between the user's historical behaviors. And further, by adopting a method similar to DIN, calculating the correlation between each user history behavior and the target advertisement to obtain the weight of the user history behavior. The click rate predictive deep session interest network model (Deep session interest network for click-through rate prediction, DSIN) is a user session interest modeling model, where the DISN first divides the user's historical behavior into multiple sessions according to a time window, capturing the intra-session and inter-session interest relationships using a transducer and Long Short-Term Memory (LSTM).

From the experimental data set forth in table 1, the subject model provided herein has smaller losses and higher Area Under Curve (AUC) values relative to other models. Moreover, the AUC improvement ratio of the target model provided by the application is highest compared with a common weighted pooling model for weighted pooling of user behavior characteristics. The AUC is the area under the receiver operating characteristic curve (Receiver Operating Characteristic Curve, ROC), which is an important index for evaluating the performance of the model. The higher the AUC value, the better the model performance.

As shown in fig. 5, a schematic diagram of the influence of a pruning scheme on the model effect of the attribute feature vector of the advertisement according to a certain weight coefficient in the target model provided by the application is shown. As shown in the figure, in order to verify the line graph of the pruning rate of the attribute feature vector on the preset public data set, as the pruning rate of each attribute feature vector is improved, the AUC corresponding to the model is gradually improved, and when the pruning rate reaches a preset pruning rate, the continuous improvement of the pruning rate can lead to the reduction of the model performance. In the data set employed in the example, the model performs best when pruning rate is 0.6. Of course, the data set adopted in this example is only one example data set, and the best model performance may be obtained under different pruning rates in other data sets.

As shown in fig. 6, in the thermodynamic diagram corresponding to each dimension attribute feature learned by the model in the present application, the shade of the color represents the height of the weight coefficient of each dimension attribute feature, and the darker the color, the higher the weight coefficient corresponding to the attribute feature. As shown in the figure, the advertisement resource information, the delivery channel information and the associated content information have the highest weight coefficient, and the attribute characteristics of other dimensions have lower weight coefficients. Thus, it is determined that the advertisement resource information, the advertisement delivery channel information, and the associated content information have larger weight coefficient values, while the advertisement group information, the user identity information, and the brand information have sub-weight coefficient values, and the content management system information and the city level information have the smallest weight coefficient values. Alternatively, when selecting the attribute features, only a part of the attribute features having a high weight coefficient may be selected for calculation.

The thermodynamic diagram corresponding to each attribute data shown in fig. 6 is trained based on a common data set, and different results may be trained for different data sets. That is, thermodynamic diagrams corresponding to the attribute data obtained through training according to different public data sets may be different, that is, weight coefficients corresponding to the attribute features may also be different.

According to the content determining method provided by the application, the computer equipment obtains at least one behavior feature corresponding to the historical behavior data of the target object; acquiring attribute characteristics of at least two dimensions corresponding to each candidate content in at least two candidate contents, wherein the attribute characteristics are characteristics corresponding to attribute data of the candidate contents; calculating the interest degree of the target object on each candidate content according to the behavior characteristics and the attribute characteristics; and determining target candidate content according to the interest degree.

In order to better implement the above method, the embodiments of the present application also provide a content determining apparatus, which may be integrated in a terminal or a server.

For example, as shown in fig. 7, a schematic structural diagram of a content determining apparatus provided in an embodiment of the present application may include a first acquiring unit 301, a second acquiring unit 302, a calculating unit 303, and a determining unit 304, as follows:

a first obtaining unit 301, configured to obtain at least one behavior feature corresponding to historical behavior data of a target object;

a second obtaining unit 302, configured to obtain attribute features of at least two dimensions corresponding to each candidate content in at least two candidate contents, where the attribute features are features corresponding to attribute data of the candidate content;

a calculating unit 303, configured to calculate the interest degree of the target object for each candidate content according to the behavior feature and the attribute feature;

a determining unit 304, configured to determine target candidate content according to the interest level.

In some embodiments, the computing unit comprises:

a calculation subunit, configured to calculate a first weight coefficient of each behavior feature based on the behavior feature and the attribute feature;

the processing subunit is used for weighting at least one behavior characteristic according to the first weight coefficient of each behavior characteristic to obtain an interest characteristic corresponding to each candidate content;

And the determining subunit is used for determining the interest degree of the target object on each candidate content based on the interest characteristics.

In some embodiments, the computing subunit comprises:

and the calculation module is used for calculating the first weight coefficient of each behavior feature according to the behavior feature, the attribute feature and the second weight coefficient.

In some embodiments, the acquisition module includes:

In some embodiments, the computing module includes:

the processing sub-module is used for carrying out element level product processing on any target behavior feature and each attribute feature, carrying out weighting processing on the processing result according to the second weighting coefficient, and carrying out first weighting coefficient corresponding to the target behavior feature according to the weighting processing result;

The computing sub-module is used for traversing each behavior feature based on a computing method of the first weight coefficient corresponding to the target behavior feature, and computing to obtain the first weight coefficient of each behavior feature.

In some embodiments, determining the subunit includes:

and the prediction module is used for inputting the interest characteristics into the interest degree prediction model to obtain the interest degree of the target object on each candidate content.

a third acquisition unit configured to acquire target attribute data of at least one dimension of target candidate content;

and a recommending unit for recommending the target candidate content to the target user based on the target attribute data.

In some embodiments, the first acquisition unit comprises:

In some embodiments, the second acquisition unit comprises:

the calculating unit is further used for:

and calculating according to the behavior characteristics, the resource attribute characteristics, the channel attribute characteristics and the associated content attribute characteristics to obtain the interest degree of the target object on each candidate content.

In the implementation, each unit may be implemented as an independent entity, or may be implemented as the same entity or several entities in any combination, and the implementation of each unit may be referred to the foregoing method embodiment, which is not described herein again.

As can be seen from the above description, the content determining apparatus provided in the embodiments of the present application acquires, by the first acquiring unit 301, at least one behavior feature corresponding to historical behavior data of a target object; the second obtaining unit 302 obtains attribute features of at least two dimensions corresponding to each candidate content in the at least two candidate contents, where the attribute features are features corresponding to attribute data of the candidate content; the calculation unit 303 calculates the degree of interest of the target object for each candidate content based on the behavior feature and the attribute feature; the determination unit 304 determines target candidate contents according to the degree of interest.

The embodiment of the application also provides a computer device, which may be a terminal or a server, as shown in fig. 8, which is a schematic structural diagram of the computer device provided in the application. Specifically, the present invention relates to a method for manufacturing a semiconductor device.

The computer device may include one or more processing cores 'processing units 401, one or more storage media's storage units 402, a power module 403, and an input module 404, among other components. Those skilled in the art will appreciate that the computer device structure shown in FIG. 8 is not limiting of the computer device and may include more or fewer components than shown, or may be combined with certain components, or a different arrangement of components. Wherein:

The processing unit 401 is a control center of the computer device, connects respective parts of the entire computer device using various interfaces and lines, and performs various functions of the computer device and processes data by running or executing software programs and/or modules stored in the storage unit 402 and calling data stored in the storage unit 402, thereby performing overall monitoring of the computer device. Optionally, 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 processes an operating system, a user interface, an application program, etc., and the modem processor mainly processes wireless communication. It will be appreciated that the modem processor described above may not be integrated in the processing unit 401.

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

The computer device further comprises a power module 403 for supplying power to the respective components, and preferably, the power module 403 may be logically connected to the processing unit 401 through a power management system, so that functions of managing charging, discharging, and power consumption management are implemented through the power management system. The power module 403 may also include one or more of any components, such as a direct current or alternating current power source, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and the like.

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

Although not shown, the computer device may further include a display unit or the like, which is not described herein. In particular, in this embodiment, the processing unit 401 in the computer device loads executable files corresponding to the processes of one or more application programs into the storage unit 402 according to the following instructions, and the processing unit 401 executes the application programs stored in the storage unit 402, so as to implement various functions as follows:

Acquiring at least one behavior feature corresponding to historical behavior data of a target object; acquiring attribute characteristics of at least two dimensions corresponding to each candidate content in at least two candidate contents, wherein the attribute characteristics are characteristics corresponding to attribute data of the candidate contents; calculating the interest degree of the target object on each candidate content according to the behavior characteristics and the attribute characteristics; and determining target candidate content according to the interest degree.

It should be noted that, the computer device provided in the embodiment of the present application and the method in the foregoing embodiment belong to the same concept, and the specific implementation of each operation above may refer to the foregoing embodiment, which is not described herein.

Those of ordinary skill in the art will appreciate that all or a portion of the steps of the various methods of the above embodiments may be performed by instructions, or by instructions controlling associated hardware, which may be stored in a computer-readable storage medium and loaded and executed by a processor.

To this end, embodiments of the present invention provide a computer readable storage medium having stored therein a plurality of instructions capable of being loaded by a processor to perform the steps of any of the methods provided by the embodiments of the present invention. For example, the instructions may perform the steps of:

The specific implementation of each operation above may be referred to the previous embodiments, and will not be described herein.

Wherein the computer-readable storage medium may comprise: read Only Memory (ROM), random access Memory (RAM, random Access Memory), magnetic or optical disk, and the like.

Since the instructions stored in the computer readable storage medium may perform the steps in 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 be achieved are detailed in the previous embodiments, and are not described herein.

Among other things, according to one aspect of the present application, there is provided a computer program product or computer program comprising computer instructions stored in a storage medium. The computer instructions are read from the storage medium by a processor of a computer device, which executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations of the content determination methods described above.

The foregoing has described in detail the methods, apparatuses, storage medium and computer devices for determining content provided by the embodiments of the present invention, and specific examples have been applied to illustrate the principles and embodiments of the present invention, and the above description of the embodiments is only for aiding in understanding the methods and core ideas of the present invention; meanwhile, as those skilled in the art will vary in the specific embodiments and application scope according to the ideas of the present invention, the present description should not be construed as limiting the present invention in summary.

Claims

1. A content determination method, the method comprising:

calculating the interest degree of the target object on each candidate content according to the behavior characteristics and the attribute characteristics of the at least two dimensions;

and determining target candidate content according to the interest degree.

2. The method of claim 1, wherein said calculating the interest level of the target object for each candidate content based on the behavioral characteristics and the attribute characteristics of the at least two dimensions comprises:

Calculating a first weight coefficient of each behavioral feature based on the behavioral feature and the attribute features of the at least two dimensions;

weighting the at least one behavior feature according to a first weight coefficient of each behavior feature to obtain an interest feature corresponding to each candidate content;

and determining the interest degree of the target object in each candidate content based on the interest features.

3. The method of claim 2, wherein calculating the first weight coefficient for each behavioral feature from the behavioral feature and the attribute feature for the at least two dimensions comprises:

acquiring a second weight coefficient corresponding to each attribute feature in the attribute features of the at least two dimensions;

and calculating a first weight coefficient of each behavior feature according to the behavior feature, the attribute features of the at least two dimensions and the second weight coefficient.

4. A method according to claim 3, wherein the obtaining a second weight coefficient corresponding to each of the attribute features of the at least two dimensions comprises:

extracting target attribute data associated with each attribute data contained in each historical behavior data from the historical behavior data;

And determining a second weight coefficient corresponding to each attribute data according to the target attribute data, wherein each attribute data is each attribute data in at least two attribute data corresponding to each candidate content.

5. A method according to claim 3, wherein said calculating a first weight coefficient for each behavioral characteristic from said behavioral characteristic, said attribute characteristics of said at least two dimensions, and said second weight coefficient comprises:

acquiring a behavior feature vector corresponding to each behavior feature and an attribute feature vector corresponding to each attribute feature;

performing element level product processing on any target behavior feature and each attribute feature, performing weighting processing on a processing result according to the second weight coefficient, and determining a first weight coefficient corresponding to the target behavior feature according to the weighting processing result;

traversing each behavior feature based on a calculation method of the first weight coefficient corresponding to the target behavior feature, and calculating to obtain the first weight coefficient of each behavior feature.

6. The method of claim 2, wherein determining the degree of interest of the target object for each candidate content based on the interest features comprises:

Acquiring a trained interested degree prediction model;

and inputting the interest characteristic into the interest degree prediction model to obtain the interest degree of the target object on each candidate content.

7. The method according to any one of claims 1 to 6, further comprising:

acquiring target attribute data of at least two dimensions of the target candidate content;

recommending the target candidate content to the target user based on the target attribute data.

8. The method according to claim 1, wherein the obtaining at least one behavior feature corresponding to the historical behavior data of the target object includes:

acquiring at least one historical behavior data of a target object in a preset time period;

mapping the historical behavior data to a vector space to obtain at least one behavior feature of the target object.

9. The method of claim 1, wherein the obtaining attribute features of at least two dimensions corresponding to each of the at least two candidate contents comprises:

acquiring resource attribute data, channel attribute data and associated content attribute data corresponding to each candidate content in at least two candidate contents;

Mapping the resource attribute data, the channel attribute data and the associated content attribute data to a vector space respectively to obtain resource attribute characteristics, channel attribute characteristics and associated content attribute characteristics corresponding to each candidate content;

the calculating the interest degree of the target object to each candidate content according to the behavior feature and the attribute feature comprises the following steps:

10. A content determining apparatus, characterized in that the apparatus comprises:

a calculating unit, configured to calculate a degree of interest of the target object for each candidate content according to the behavior feature and the attribute features of the at least two dimensions;

11. The apparatus of claim 10, wherein the computing unit comprises:

a calculating subunit, configured to calculate a first weight coefficient of each behavioral feature based on the behavioral feature and the attribute features of the at least two dimensions;

12. The apparatus of claim 11, wherein the computing subunit comprises:

the acquisition module is used for acquiring a second weight coefficient corresponding to each attribute feature in the attribute features of the at least two dimensions;

13. A computer readable storage medium storing a plurality of instructions adapted to be loaded by a processor to perform the steps in the content determination method of any one of claims 1 to 9.

14. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps in the content determination method of any one of claims 1 to 9 when the computer program is executed.

15. A computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps in the content determination method of any one of claims 1 to 9.