CN116521971A

CN116521971A - Content recommendation method, apparatus, device, storage medium, and computer program product

Info

Publication number: CN116521971A
Application number: CN202210059688.5A
Authority: CN
Inventors: 王刘鄞; 段焕中; 路彦雄
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2022-01-19
Filing date: 2022-01-19
Publication date: 2023-08-01

Abstract

The application discloses a content recommendation method, a content recommendation device, content recommendation equipment, a storage medium and a computer program product, and relates to the field of machine learning. The method comprises the following steps: acquiring account basic data and historical interaction data corresponding to a target account, wherein the historical interaction data indicates a historical interaction sequence; acquiring target interaction data in a history interaction sequence; fusing account feature representations corresponding to the account basic data, interactive feature representations corresponding to the historical interactive data and target feature representations corresponding to the target interactive data to obtain fusion features; and carrying out interest distribution prediction on the content in the content library based on the fusion characteristics to obtain an interest distribution prediction result. In the process of predicting the recommended content based on the target account and the corresponding interaction data, feature fusion is carried out on the feature representation corresponding to the target interaction data and the feature representation corresponding to each of the target account and the historical interaction data, prediction is carried out according to the obtained fusion features, and content recommendation accuracy is improved.

Description

Content recommendation method, apparatus, device, storage medium, and computer program product

Technical Field

Embodiments of the present application relate to the field of machine learning, and in particular, to a content recommendation method, apparatus, device, storage medium, and computer program product.

Background

With the continuous development of science and technology, the social mode of people is gradually changed from off-line social mode to on-line social mode, the session refers to interaction data generated between users and items according to time sequence, and the session-based recommendation system determines interest preference of the users according to historical session data corresponding to the users and the items, so that content recommendation is performed.

In the related art, a session-based recommendation model is generally constructed, sample data corresponding to an anonymous user is input into the recommendation model to train model parameters of the recommendation model, and a target recommendation model is obtained and used for recommending contents to the user.

However, in the method, only one recommendation model is trained, so that in actual application of recommending the target user, the recommended content often ignores interest preference of the target user, and the accuracy of content recommendation is low and the human-computer interaction efficiency is poor.

Disclosure of Invention

The embodiment of the application provides a content recommendation method, a device, equipment, a storage medium and a computer program product, which can improve the accuracy of content recommendation. The technical scheme is as follows.

In one aspect, there is provided a method for recognizing an entity word, the method comprising:

acquiring account basic data and historical interaction data corresponding to a target account, wherein the account basic data are used for indicating account basic information of the target account, and the historical interaction data are used for indicating a historical interaction sequence constructed by interaction data of the target account in a historical time period;

acquiring target interaction data in a designated interaction interval in the history interaction sequence, wherein the designated interaction interval is related to the interest time distribution of the target account;

fusing account feature representations corresponding to the account basic data, interactive feature representations corresponding to the historical interactive data and target feature representations corresponding to the target interactive data to obtain fusion features;

and carrying out interest distribution prediction on the content in the content library based on the fusion characteristics to obtain an interest distribution prediction result, wherein the interest distribution prediction result is used for determining the content recommended to the target account.

In another aspect, there is provided a content recommendation apparatus, the apparatus including:

the system comprises an acquisition module, a history interaction module and a history interaction module, wherein the acquisition module is used for acquiring account basic data and history interaction data corresponding to a target account, the account basic data are used for indicating account basic information of the target account, and the history interaction data are used for indicating a history interaction sequence constructed by interaction data of the target account in a history time period;

The acquisition module is further configured to acquire target interaction data in a specified interaction interval in the historical interaction sequence, where the specified interaction interval is related to interest time distribution of the target account;

the fusion module is used for fusing the account feature representation corresponding to the account basic data, the interaction feature representation corresponding to the historical interaction data and the target feature representation corresponding to the target interaction data to obtain fusion features;

and the prediction module is used for predicting the interest distribution of the content in the content library based on the fusion characteristics to obtain an interest distribution prediction result, wherein the interest distribution prediction result is used for determining the content recommended to the target account.

In another aspect, a computer device is provided, where the computer device includes a processor and a memory, where at least one instruction, at least one program, a set of codes, or a set of instructions is stored, where the at least one instruction, the at least one program, the set of codes, or the set of instructions are loaded and executed by the processor to implement a content recommendation method as in any one of the embodiments of the present application.

In another aspect, a computer readable storage medium is provided, in which at least one instruction, at least one program, a set of codes, or a set of instructions is stored, the at least one instruction, the at least one program, the set of codes, or the set of instructions being loaded and executed by a processor to implement a content recommendation method as described in any one of the embodiments of the application.

In another aspect, a computer program product or computer program is provided, the computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to perform the content recommendation method according to any of the above embodiments.

The beneficial effects that technical scheme that this application embodiment provided include at least:

in the process of fusing the account characteristic representation of the account basic data corresponding to the target account and the interactive characteristic representation of the historical interactive data corresponding to the target account, adding the target characteristic representation corresponding to the historical interactive sequence constructed by the interactive data of the target account in the historical time period, fusing the three, determining the fusion characteristic, carrying out interest prediction distribution on the content based on the fusion characteristic, finally determining the content recommended to the target account, and carrying out characteristic fusion by adding the interactive data containing the interactive sequence, so that interest preference of the target account is more easily captured in the content recommendation process, and the accuracy of content recommendation is 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, and it is 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 diagram of a content recommendation method provided in an exemplary embodiment of the present application;

FIG. 2 is a schematic illustration of an implementation environment provided by an exemplary embodiment of the present application;

FIG. 3 is a flow chart of a content recommendation method provided by an exemplary embodiment of the present application;

FIG. 4 is a flow chart of a content recommendation method provided in another exemplary embodiment of the present application;

FIG. 5 is a schematic diagram of a stitching feature provided in an exemplary embodiment of the present application;

FIG. 6 is a global fusion schematic provided by another exemplary embodiment of the present application;

FIG. 7 is a schematic diagram of global fusion and local fusion provided by another exemplary embodiment of the present application;

FIG. 8 is a flowchart of a content recommendation method provided by another exemplary embodiment of the present application;

FIG. 9 is a schematic diagram of a target knowledge graph provided in an exemplary embodiment of the present application;

FIG. 10 is a schematic diagram of a content recommendation method provided in another exemplary embodiment of the present application;

FIG. 11 is a block diagram of a content recommendation device provided in an exemplary embodiment of the present application;

FIG. 12 is a block diagram of a content recommendation device provided in another exemplary embodiment of the present application;

fig. 13 is a schematic structural diagram of a server according to an exemplary embodiment of the present application.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the present application more apparent, the embodiments of the present application will be described in further detail below with reference to the accompanying drawings.

First, a brief description will be given of terms involved in the embodiments of the present application.

Session (Session): is an interactive behavior generated by the target user within a specified time period, and the session-based recommendation is based on the current session recommending the item that the target user clicks next. The historical interaction sequence of the target user is divided into a plurality of sessions according to relevant logic by session recommendation, the current session is modeled, and the current interest preference of the target user is obtained and used for predicting the next interested article of the user.

Natural language processing (Nature Language Processing, NLP): is an important direction in the fields of computer science and artificial intelligence. It is studying various theories and methods that enable effective communication between a person and a computer in natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Thus, the research in this field will involve natural language, i.e. language that people use daily, so it has a close relationship with the research in linguistics. Natural language processing techniques typically include text processing, semantic understanding, machine translation, robotic questions and answers, knowledge graph techniques, and the like.

Machine Learning (ML): is a multi-domain interdisciplinary, and relates to a plurality of disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. It is specially studied how a computer simulates or implements learning behavior of a human to acquire new knowledge or skills, and reorganizes existing knowledge structures to continuously improve own performance. Machine learning is the core of artificial intelligence, a fundamental approach to letting computers have intelligence, which is applied throughout various areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, induction learning, teaching learning, and the like.

Heteropatterned neural network (Hypergraph Neural Nerworks, HGNN): the purpose of heterographing representation learning is to seek a meaningful vector representation for each node to facilitate subsequent applications such as link prediction, personalized recommendations, node classification, etc. Therefore, a neural network which comprises the same structure as the main node and is different from the main node is constructed to acquire different types of characteristic information, wherein each node contained in the heterogeneous graph neural network is called a 'neighbor' of the main node, so that the 'neighbor' comprises a isomorphic neighbor and a heterogeneous neighbor, and the weight value allocated by each 'neighbor' is different.

Referring to fig. 1 schematically, a schematic diagram of a content recommendation method provided by an exemplary embodiment of the present application is shown, as shown in fig. 1, account basic data 101 and historical interaction data 102 corresponding to a target account 100 are obtained, a historical interaction sequence (such as interaction data 1, interaction data 2, … … and interaction data n) constructed by interaction data included in a historical time period is obtained in the historical interaction data 102, wherein from the interaction data 1 to the interaction data n are sequentially arranged interaction data), target interaction data 103 in a specified interaction interval is obtained in the historical interaction data 102, account feature representation 104 corresponding to the account basic data 101, interaction feature representation 105 corresponding to the historical interaction data 102 and target feature representation 106 corresponding to the target interaction data 103 are determined, the account feature representation 104, the interaction feature representation 105 and the target feature representation 106 are fused to obtain fusion features, interest distribution prediction is performed on content in a content library based on the fusion features, at least one content (such as content 1, content 2, … … and content n) to be predicted, and a final interest distribution prediction result of the content library is determined to be the target interest distribution prediction result of the target account.

The embodiment of the application provides a content recommendation method, in the process of fusing account characteristic representation of account basic data corresponding to a target account and interactive characteristic representation of historical interaction data corresponding to the target account, adding target characteristic representation corresponding to a historical interaction sequence constructed by the interactive data of the target account in a historical time period, fusing the three, determining fusion characteristics, carrying out interest prediction distribution on content based on the fusion characteristics, finally determining content recommended to the target account, and carrying out characteristic fusion by adding the interactive data containing an interaction sequence, so that interest preference of the target account is easier to capture in the content recommendation process, and the accuracy of content recommendation is improved.

The application scenario of the embodiments of the present application is illustrated by combining the description of the noun introduction and the method.

1.The method is applied to shopping recommendation scenes.Acquiring account basic data corresponding to a shopping account of a user and historical shopping data corresponding to the shopping account, wherein the account basic data comprises shopping preferences corresponding to the user, the historical shopping data comprises historical shopping records arranged according to the distance sequence of the purchase time, one historical shopping record closest to the current time point in the historical shopping records is acquired as target interaction data, characteristic representations corresponding to the account basic data, the target interaction data and the historical shopping data are extracted, the three characteristic representations are subjected to characteristic fusion, and the product with the highest purchase expectation of the current user is presumed to be recommended to the user based on the fusion result;

2.The method is applied to video recommendation scenes.Acquiring account basic data corresponding to an audience account of a user and historical viewing data corresponding to the audience account, wherein the account basic data comprises video viewing preferences corresponding to the user, the historical viewing data comprises historical viewing records corresponding to historical viewing videos which are arranged according to the sequence of the distance of viewing time, one historical viewing video which is closest to the current time point in the historical viewing videos is acquired as target interaction data, the characteristic representations corresponding to the account basic data, the target interaction data and the historical viewing records are extracted, the characteristic fusion is carried out on the three characteristic representations, and the video with the largest viewing probability of the user in the current time period is presumed to be recommended to the user based on the fusion result.

It should be noted that, the content recommendation method provided in the embodiment of the present application may be implemented by a terminal, or may be implemented by a server, or may be implemented by a cooperation of the terminal and the server.

When the terminal and the server cooperatively implement the scheme provided in the embodiments of the present application, the terminal and the server may be directly or indirectly connected through a wired or wireless communication manner, which is not limited in the embodiments of the present application.

Referring to fig. 2, a schematic diagram of an implementation environment provided in an exemplary embodiment of the present application is shown, and as shown in fig. 2, the implementation environment includes a terminal 210 and a server 220, where the terminal 210 and the server 220 are connected through a communication network 230.

In the embodiment of the present application, an application program having a content recommendation function is installed in the terminal 210. The application program providing the content recommendation function may be implemented as a browser, a video playing program, an electronic book reading program, an online shopping application program, an instant messaging application program, etc., which is not limited in this embodiment of the present application. After performing at least one triggering operation on the historical browsing content by the user, the terminal 210 sends at least one historical browsing content triggered by the triggering operation as historical interaction data to the server 220.

After receiving at least one historical browsing content corresponding to a user sent by a terminal, the server 220 obtains account basic data containing user account basic information, constructs a historical interaction sequence according to the historical browsing content, determines target interaction data in a designated interaction interval in the historical interaction sequence, namely target browsing content, obtains account feature representation corresponding to the account basic data, interaction feature representation corresponding to the historical interaction data and target feature representation corresponding to the target interaction data, and fuses the three feature representations to obtain a fusion feature 221.

In addition, the server 220 further includes a content library 222, the content library 222 includes at least one candidate content to be recommended, the fusion feature 221 is input into the content library 222, the candidate content included in the content library 222 is subjected to interest distribution prediction, the interest distribution prediction result corresponding to the candidate content by the user is determined, and the target recommended content 223 recommended to the user is determined based on the interest distribution prediction result corresponding to the at least one candidate content.

The server 220 feeds back the target recommended content to the terminal 210, and the terminal 210 displays the target recommended content, wherein the display mode includes at least one of voice display, text display or interface display.

The terminal 210 includes at least one of a smart phone, a tablet computer, a portable laptop, a desktop computer, an intelligent sound box, an intelligent wearable device, an intelligent voice interaction device, an intelligent home appliance, a vehicle-mounted terminal, and the like.

It should be noted that the above-mentioned communication network 230 may be implemented as a wired network or a wireless network, and the communication network 230 may be implemented as any one of a local area network, a metropolitan area network, or a wide area network, which is not limited in this embodiment.

It should be noted that the server 220 may be implemented as a Cloud server, where Cloud technology (Cloud technology) refers to a hosting technology that unifies serial resources such as hardware, software, and networks in a wide area network or a local area network to implement calculation, storage, processing, and sharing of data. The cloud technology is based on the general names of network technology, information technology, integration technology, management platform technology, application technology and the like applied by the cloud computing business mode, can form a resource pool, and is flexible and convenient as required. Cloud computing technology will become an important support. Background services of technical networking systems require a large amount of computing, storage resources, such as video websites, picture-like websites, and more portals. Along with the high development and application of the internet industry, each article possibly has an own identification mark in the future, the identification mark needs to be transmitted to a background system for logic processing, data with different levels can be processed separately, and various industry data needs strong system rear shield support and can be realized only through cloud computing.

In some embodiments, the server 220 described above may also be implemented as a node in a blockchain system.

In connection with the above description of the noun introduction and the application scenario, the content recommendation method provided in the embodiment of the present application will be described by taking the execution of the method by the server as an example, and referring to fig. 3, a flowchart of the content recommendation method provided in an exemplary embodiment of the present application is shown schematically, where the method includes the following steps.

Step 301, obtaining account basic data and history interaction data corresponding to a target account.

The account basic data are used for indicating account basic information of the target account, and the history interaction data are used for indicating a history interaction sequence constructed by the interaction data of the target account in a history time period.

Illustratively, the basic account information includes account information corresponding to the target account, such as: user gender, age, hobbies and the like corresponding to the target account; or the basic account information also comprises account social information corresponding to the target account, such as: if the associated account exists, the account social information is used for indicating the social relationship between the target account and the associated account; or the basic account information includes the account preference information corresponding to the target account, the account preference information indicates that in a specific period of time, the interest preference corresponding to the target account is determined based on the history operation record corresponding to the target account, for example: when the target account is the corresponding account in the video application program, the video type corresponding to the video content browsed by the target account in a certain appointed historical time period is used as the account preference information corresponding to the target account.

Wherein, the basic account data is updated according to the basic account information of the target account.

Illustratively, the historical interaction data represents interaction data generated between the target account and the historical content or historical items over a historical period of time, such as: taking a target account as a video application program for example, when the target account generates trigger operation on a certain video content or browses the video content, the trigger action/browsing action correspondingly generates interactive data between the target account and the video content.

In some embodiments, the historical interaction sequence represents a series of interaction data corresponding to the interaction behavior generated by the target account (the interaction behavior includes, for example, clicking operation and browsing operation in the video scene, and purchasing operation in the shopping scene, etc., which are not limited herein), wherein the series of interaction data is arranged in sequence, and the shopping scene is exemplified as the example:

in a shopping scene, the target account is a shopping account corresponding to a shopping application program, wherein in a historical time period, the target account purchases at least one article, a historical purchase record generated by the purchase behavior is historical interaction data, and a sequence corresponding to the historical purchase record which is arranged in sequence in the historical time period corresponding to the target account is a historical interaction series.

Optionally, the arranging in sequence includes: the arrangement is not limited herein, and the arrangement is performed according to the time sequence generated by the associated behavior, or the arrangement is performed according to the repetition number corresponding to the interactive behavior (for example, the arrangement is performed according to the repeated browsing number of the user history from high to low in the video scene).

It will be appreciated that in the specific embodiments of the present application, related data such as account base data, historical interaction data, etc. are involved, when the embodiments of the present application are applied to specific products or technologies, user permissions or consents need to be obtained, and the collection, use and processing of related data need to comply with related laws and regulations and standards of related countries and regions.

Step 302, obtaining target interaction data in a specified interaction interval in a history interaction sequence.

Wherein the designated interaction interval is related to the time distribution of interest of the target account.

Optionally, when the history interaction sequence is constructed according to the time sequence, the specified interaction interval includes a specified time period or a specified time point, where the specified time period or the specified time point is an interest time distribution of the history interaction data corresponding to the target account, and the target interaction data is one or more pieces of history interaction data in the specified time period, or one or more pieces of history interaction data corresponding to the specified time point, for example: acquiring one or more pieces of historical interaction data closest to a current time point from historical interaction data arranged in time sequence as target interaction data; when the historical interaction sequence is constructed according to the interaction behavior sequence, the appointed interaction interval comprises the set interaction threshold, namely the interaction threshold is the interest time distribution corresponding to the target account, and when the interaction behavior between the target account and the historical interaction data reaches the interaction threshold, the historical interaction data is used as the target interaction data.

Step 303, fusing account feature representation corresponding to the account basic data, interaction feature representation corresponding to the history interaction data and target feature representation corresponding to the target interaction data to obtain fusion features.

Illustratively, account feature representations corresponding to account basic data, interaction feature representations corresponding to historical interaction data and target feature representations corresponding to target interaction data are extracted, wherein when the target interaction data is one interaction data, the target feature representations correspond to the target interaction data, and when the target interaction data comprises a plurality of interaction data, the target feature representations are common feature representations corresponding to the plurality of target interaction data.

Optionally, the feature representation methods corresponding to the account feature representation, the interaction feature representation and the target feature representation include representation in a vector form, representation in a collection form or representation in an array form, which is not limited herein.

Optionally, the fusion mode includes at least one of the following modes:

1. sequentially splicing the account feature representation, the interaction feature representation and the target feature representation to obtain a splicing result, wherein the splicing result is used as a fusion feature;

2. Establishing a feature fusion network, and inputting the account feature representation, the interaction feature representation and the target feature representation into the feature fusion network simultaneously to obtain an output result as fusion features corresponding to the account feature representation, the interaction feature representation and the target feature representation, wherein the fusion features comprise: a convolutional neural network is established, and a new feature representation is obtained as a fusion feature by performing convolution, pooling, connection and other operations on the input feature representation.

It should be noted that the above description of the fusion mode is only an illustrative example, and the embodiments of the present application are not limited thereto.

And step 304, carrying out interest distribution prediction on the content in the content library based on the fusion characteristics to obtain an interest distribution prediction result.

The interest distribution prediction result is used for determining the content recommended to the target account.

Illustratively, the interest distribution prediction includes calculating the interest probability of each content in the content library corresponding to the target account, that is, predicting the interest probability value of each content in the content library corresponding to the target account, where the greater the interest probability value, the greater the probability that the target account performs interactive operation on the target account.

In some embodiments, the content library is a pre-established library, and the content contained in the content library is recommended content to be predicted, wherein the recommended content contains content with interaction behavior with other accounts or content with interaction behavior with any account is not present; the content of the content library contains the same type or different types of content, and is not limited herein.

Optionally, the method of interest distribution prediction includes at least one of the following prediction methods:

1. obtaining content characteristics corresponding to each content in a content library, comparing the similarity of each content characteristic to the fusion characteristics to obtain a similarity score, taking the similarity score as a prediction probability corresponding to the content, and finally taking the similarity score corresponding to each content in the content library as an interest distribution prediction result;

2. and obtaining fusion vector form representation corresponding to the fusion features, and obtaining content features corresponding to the contents in the content library, wherein the content features are represented in the form of feature vectors, calculating the distance between the fusion vectors and the feature vectors, obtaining a distance result as a prediction probability corresponding to the contents, and finally, each content in the content library is corresponding to a distance result as an interest distribution prediction result.

It should be noted that the above description of the interest distribution prediction mode is merely an illustrative example, and the embodiments of the present application are not limited thereto.

And determining the prediction probability corresponding to each content in the content library, and determining one or the appointed number of the content with the highest prediction probability in the content library as recommended content, or setting a probability threshold value, taking the content with the prediction probability reaching the probability threshold value in the content library as recommended content, and recommending the recommended content to the target account.

In summary, the embodiment of the present application provides a content recommendation method, in the process of fusing the account feature representation of the basic data of the account corresponding to the target account and the interaction feature representation of the historical interaction data corresponding to the target account, adding the target feature representation corresponding to the historical interaction sequence constructed by the interaction data of the target account in the historical time period, fusing the three features, determining the fusion feature, performing interest prediction distribution on the content based on the fusion feature, and finally determining the content recommended to the target account, and performing feature fusion by adding the interaction data including the interaction sequence, so that interest preference of the target account is easier to capture in the content recommendation process, and the accuracy of content recommendation is improved.

In an alternative embodiment, the fusion manner includes a manner of splicing an account feature representation, an interaction feature representation and a target feature representation, please refer to fig. 4, which illustrates a flowchart of a content recommendation method provided in an exemplary embodiment of the present application, as shown in fig. 4, and the method includes the following steps:

step 401, obtaining account basic data and history interaction data corresponding to a target account.

The details of the account basic data and the history interactive data in step 401 are described in the above step 301, and will not be described herein.

Step 402, the last n interactive data in the history interactive sequence are obtained as target interactive data, n is a positive integer.

In some embodiments, the most recent interaction data in the historical interaction sequence is obtained as target interaction data.

The historical interaction sequence comprises a plurality of historical interaction data which are sequentially arranged according to time sequence, wherein the current time point is taken as an anchor point, the last historical interaction data in the sequence in the historical interaction sequence is the interaction data closest to the current time point in time distance, and the last historical interaction data is taken as target interaction data.

The content interest directions of the target account in different time periods are indicated before and after the ordering corresponding to the history interaction sequence, for example: taking video content as an example, when all the first three historical browsing videos in the historical interaction sequence are cartoon videos, the target account is indicated to have higher interest in the cartoon videos in the browsing time period corresponding to the first three historical browsing videos, and the last historical browsing video in the historical interaction sequence is a variety video, the video content with higher interest closest to the current time point of the target account is indicated to be a variety video, so that the latest interaction data in the historical interaction sequence is selected as the target interaction data, and the latest short-term interest of the target account can be better expressed.

And step 403, splicing account feature representations corresponding to the account basic data, interactive feature representations corresponding to the historical interactive data and target feature representations corresponding to the target interactive data to obtain splicing features.

Illustratively, the account feature representation corresponding to the account basic data, the interactive feature representation corresponding to the history interactive data and the target feature representation corresponding to the target interactive data are spliced in sequence, or the account feature representation, the interactive feature representation and the target feature representation are randomly arranged each time and spliced to obtain spliced features, which is not limited herein.

Fig. 5 is a schematic diagram, as shown in fig. 5, showing a stitching feature provided in an exemplary embodiment of the present application, and fig. 5 is a stitching feature 500, where the stitching feature 500 includes an account feature representation 501 corresponding to account basic data, an interaction feature representation 502 corresponding to historical interaction data, and a target feature representation 503 corresponding to target interaction data that are stitched in sequence, where a stitching manner please refer to formula one:

equation one: x= [ f _u ||f _s ||f _short ]

Wherein X represents a splice feature, ||represents splice, f _u Representing account number characteristics, f _s Representing interactive feature representations, f _short Representing a target feature representation.

It should be noted that, in this embodiment, the account feature representation, the interaction feature representation, and the target feature representation are features expressed in the same type of feature representation.

And step 404, performing feature fusion extraction on the spliced features to obtain fusion features.

In some embodiments, inputting the stitching feature into a feedforward neural network to obtain a first feedforward feature vector; determining a first fusion result based on the splice feature and the first feedforward feature vector; determining an attention vector corresponding to the first fusion result through a multi-head self-attention mechanism; determining a second fusion result based on the first fusion result and the attention vector; obtaining a third fusion result based on the second fusion result and a convolution vector corresponding to the second fusion result; and obtaining fusion characteristics based on the third fusion result and the second feedforward characteristic vector corresponding to the third fusion result.

Schematically, in this embodiment, two ways are used to perform fusion extraction on the spliced features, so as to obtain fusion features.

First, a Multi-headed self-attention mechanism (Multi-Head Self Attention, MHSA) is used to globally fuse the splice features.

Referring to fig. 6, a global fusion schematic provided in an exemplary embodiment of the present application is shown, and as shown in fig. 6, a stitching feature 601 is input into a first neural network 602 constructed by a multi-head self-attention mechanism, and a fusion feature 603 is output.

In the neural network 602, the input stitching feature 601 first obtains a first feedforward feature vector through a feedforward neural network (Feedforward Neural Network, FNN), and linearly converts the first feedforward feature vector from Q, K, V to correspond to different parameters through a self-attention networkAnd->Wherein W corresponds to a different superscript indicating a different parameter, i.e. reference is made to formula two below:

formula II:

wherein, Q table Query, K represents Key, V represents Value, and the three parameters represent three sequence vectors in the attention mechanism.

The three sequence vectors after linear transformation determine their corresponding attention values, i.e. the weight values corresponding to Value, through an attention function, and reference is made, illustratively, to the formula three:

and (3) a formula III:

wherein, the liquid crystal display device comprises a liquid crystal display device, ^d q represents a dimension and,in order to prevent the gradient from disappearing.

After the weight Value corresponding to the Value is determined, the weight Value is used as a single self-attention Value, a plurality of self-attention values are connected to obtain a multi-head attention vector, the total consumption is reduced in a dimension reducing mode, and the method is schematically shown in a formula IV and a formula V:

equation four:

formula five: multihead (Q, K, V) = [ head ] ₁ ，head ₂ ，...，head _h ]W _O +b _O ；

After determining the plurality of self-attention values, the fusion characteristics are obtained through feedforward neural network and layer normalization (Layer Normaliztion) and finally output.

Secondly, global fusion is carried out by adopting a multi-head self-attention mechanism, and local fusion is carried out based on a result obtained by the global fusion.

Referring to fig. 7, a global fusion and local fusion schematic diagram provided in an exemplary embodiment of the present application is shown, as shown in fig. 7, a stitching feature 701 is input into a second neural network 702, the second neural network 702 includes a first neural network corresponding to a multi-head self-attention mechanism, and a fusion feature 703 is output.

The method comprises the steps of inputting splicing features into a feedforward neural network to obtain a first feedforward feature vector, and determining a first fusion result according to the splicing features and the first feedforward feature vector, wherein the first fusion result is schematically determined by referring to a formula six:

formula six:

wherein, the liquid crystal display device comprises a liquid crystal display device,the first fusion result is represented, and FFM (X) represents the first feedforward feature vector.

Secondly, inputting the first fusion result into a first neural network corresponding to a multi-head self-attention mechanism, determining a multi-head attention vector, and determining a second fusion result according to the multi-head attention vector and the first fusion result, wherein the schematic way of determining the second fusion result is shown in a formula seven:

formula seven:

Wherein, the liquid crystal display device comprises a liquid crystal display device,representing a multi-headed vector of attention, and X' representing a second fusion result.

Then, the second fusion result is input into a Convolution layer (Convolition) for Convolution to obtain a Convolution vector corresponding to the second fusion result, and a third fusion result is determined according to the second fusion result and the Convolution vector corresponding to the second fusion result, and the schematic way of determining the third fusion result please refer to the formula eight:

formula eight: x ' =X ' +ConvM (X ')

Wherein X 'represents the third fusion result, convM (X') represents the convolution vector corresponding to the second fusion result.

Finally, inputting the third fusion result into a feedforward neural network and layer standardization to obtain a second feedforward feature vector corresponding to the third fusion result, and determining fusion features according to the third fusion result and the second feedforward feature vector corresponding to the third fusion result, wherein the schematic way of determining the fusion features is shown in a formula nine:

formula nine:

wherein X is _fusion Representing the fusion feature, FFM (X') represents a second feedforward feature vector corresponding to the third fusion result.

It should be noted that the two fusion modes included in the present embodiment are only illustrative examples, and the embodiments of the present application are not limited thereto.

And step 405, carrying out interest distribution prediction on the content in the content library through the splicing features and the fusion features to obtain an interest distribution prediction result.

After the fusion feature is obtained, the interest distribution prediction is performed on the content in the content library by combining the splicing feature and the fusion feature, and the interest distribution prediction mode is shown in formula ten and formula eleven:

formula ten: h is a _final ＝W _f [X||X _fusion ]

Formula eleven:

in formula ten, h _final Predictive feature representations representing combinations of splice features and fusion features for comparison with content in a content library to determine content pairsAnd (5) corresponding similarity.

In formula eleven, z _i And representing the prediction probability corresponding to the ith content, and finally obtaining a group of prediction probability distribution corresponding to each content in the content library as an interest distribution prediction result.

Schematically, z _i The normalization processing is further needed through the softmax function, so that the accuracy of the obtained prediction probability distribution is higher, wherein the prediction probability corresponding to each content indicates the probability corresponding to the interaction behavior of the target account corresponding to the content at the next time, and the higher the prediction probability indicates the greater the probability of the interaction behavior of the target account at the next time, for example: taking video content as an example, the higher the prediction probability corresponding to the video, the greater the probability that the target account browses the video next time.

In this embodiment, two fusion modes are enumerated, the first is to perform global fusion on the spliced features through a multi-head self-attention mechanism to obtain fusion features, the second is to perform global fusion on the spliced features through the multi-head self-attention mechanism, then perform local fusion on the obtained fusion results through a convolution layer to finally obtain fusion features, and the obtained fusion features can fuse information corresponding to account feature representation, interaction feature representation and target feature representation more comprehensively through the global fusion/global+local fusion mode, so that accuracy of the fusion features is improved.

In this embodiment, the content in the content library is predicted by combining the splicing feature and the fusion feature, so that the prediction accuracy corresponding to the content recommendation degree is improved.

In an alternative embodiment, the stitching process further includes stitching the account feature representation with the interaction vector representation corresponding to the historical interaction data, please refer to fig. 8, which shows a flowchart of a content recommendation method provided in an exemplary embodiment of the present application, as shown in fig. 8, the method includes the following steps:

step 801, obtaining account basic data and history interaction data corresponding to a target account.

The details of the account basic data and the history interaction data in step 801 are described in the above step 301, and will not be described herein.

Step 802, obtaining associated account data and associated interaction data corresponding to the associated account.

The associated account data is used for indicating account basic information of the associated account, and the associated interaction data comprises historical interaction data corresponding to the associated account.

Illustratively, the associated account includes an account having an association relationship with the target account, where the association relationship includes at least one of a friend relationship, a lover relationship, a family relationship, a work partner relationship, and the like, which is not limited herein.

Wherein, the associated account data comprises account information corresponding to the associated account, such as: the gender, age, hobbies and the like of the user corresponding to the associated account; or the basic account information comprises account preference information corresponding to the associated account, wherein the account preference information indicates that the interest preference corresponding to the associated account is determined based on a history operation record corresponding to the associated account within a certain period of time, for example: when the associated account is the corresponding account in the video application program, the associated account is used as the account preference information corresponding to the associated account, wherein the video type corresponds to the video content browsed in a certain appointed historical time period.

Wherein, the related account data is updated according to the basic information of the related account.

Illustratively, the associated interaction data represents interaction data generated between the associated account and the historical content or historical item over a historical period of time, such as: taking the associated account as a video application program for example, when the associated account generates a triggering operation on a certain video content or browses the video content, the triggering action or browsing action correspondingly generates interactive data between the associated account and the video content.

It will be appreciated that in particular embodiments of the present application, related account data, related interaction data, etc. are referred to, and when the embodiments of the present application are applied to particular products or technologies, user permissions or consents need to be obtained, and the collection, use and processing of related data need to comply with related laws and regulations and standards of related countries and regions.

It should be noted that, when the target account and the associated account have the same historical interaction data, the historical interaction data is not only used as the historical interaction data corresponding to the target account, but also used as the associated interaction data corresponding to the associated account, which is not limited herein.

Step 803, determining a target knowledge graph corresponding to the target account based on the account basic data, the history interaction data, the associated account and the associated interaction data.

The target knowledge graph is used for indicating association relations among account basic data, historical interaction data, associated accounts and associated interaction data.

Referring to fig. 9, a schematic diagram of a target knowledge graph provided by an exemplary embodiment is shown, as shown in fig. 9, fig. 9 includes a target knowledge graph 910 corresponding to account basic data 901 (located at the center of the graph), history interaction data 902 (represented by white circles in fig. 9), and associated account data 903 (represented by black circles except for the center of the graph in fig. 9), where the history interaction data 902 and the associated interaction data are shown in fig. 9 as the same interaction data, and the associated interaction data is not shown here.

The account basic data 901 is connected with the history interaction data 902, the account basic data 901 is connected with the associated account data 903, and the history interaction data 902 with a data association relationship is connected, where the data association relationship includes two pieces of history interaction data that are of the same type or corresponding to the same item, and the like, which is not limited herein.

Illustratively, because the data types of the associated account data 903 and the account basic data 901 are all account data, the data structure types are the same, so the associated account data 903 is used as a "isomorphic neighbor" corresponding to the account basic data 901, and because the data type corresponding to the history interaction data 902 is interaction data, the data types of the associated account data 903 and the account basic data 901 are different data structure types, so the history interaction data 902 is used as a "heterogeneous neighbor" corresponding to the account basic data 901.

Step 804, determining account nodes corresponding to the account basic data and interaction nodes corresponding to the history interaction data based on the target knowledge graph.

In some embodiments, determining associated node information corresponding to associated account data and interaction node information corresponding to historical interaction data based on a target knowledge graph; and determining account nodes corresponding to the account basic data and interaction nodes corresponding to the historical interaction data based on the associated node information and the interaction node information.

Illustratively, the account node includes node information corresponding to an account data base, and the interaction node includes node information corresponding to historical interaction data.

Schematically, as shown in fig. 9, in the target knowledge graph 910, each data (including the account basic data 901, the history interaction data 902 and the associated account data 903) is correspondingly represented as a node in the target knowledge graph 910, so that each node corresponds to a node information, after the data is updated, the node information corresponding to the data is updated accordingly, and it is worth noting that the target knowledge graph 910 is a structure diagram overlapped layer by layer, and each layer contains the history interaction data 902 or the associated account data 903, so that the updated node information needs to be finally determined in a layer by layer updating manner in the updating process.

In the target knowledge graph, because each node has a connection (namely, a connection line between the nodes represents the connection between two nodes), when updating the account basic data corresponding to the target account, the importance of an isomorphic neighbor and an isomorphic neighbor corresponding to the account basic data needs to be determined, and the importance of the isomorphic neighbor need to be aggregated, so that the update information corresponding to the nodes in the current layer is determined.

Firstly, the importance of each node in the target knowledge graph needs to be calculated, and the importance is schematically determined by referring to a formula twelve and a formula thirteenth:

formula twelve:

formula thirteen:

wherein alpha is _uv Representing importance scores e _uv Is alpha _uv The attention weight obtained through the normalization processing,andare all learnable parameters, U, v e U, ">Representing nodes u, N on layer 1 _u Representing the "neighbors" (including isomorphic neighbors and heterogeneous neighbors) corresponding to node u.

From the formula twelve, the importance score corresponding to each node is determined by connecting the associated node information and the interaction node information in the target knowledge graph, and the attention weight is obtained by normalizing the importance score through the formula thirteen.

Schematically, as shown in fig. 9, fig. 9 also shows a current layer aggregation manner, which includes determining association node information corresponding to association account data 903, constructing an association graph 911 by using the association node information and account base data 901, determining interaction node information corresponding to history interaction data 902, constructing an interaction graph 912 by using the interaction node information and the account base data 901, and updating the association node information or the interaction node information, that is, updating the association graph 911 or the interaction graph 912 after the association account data 903 or the history interaction data 902 are updated.

Please refer to formula fourteen and formula fifteen for the aggregation method:

formula fourteen:

formula fifteen:/>

wherein in formula fourteen, W ₃ And b is a parameter which can be learned, wherein MLP in the formula fifteen represents a multi-layer sensor, and the node corresponding to each node u in the single layer in the target knowledge graph can be determined through the formula fourteen and the formula fifteen, wherein the node comprises an associated node corresponding to associated account data and an interactive node corresponding to historical interactive data.

Schematically, as shown in fig. 9, the association nodes are used for constructing an association relationship graph 911, the interaction nodes are used for constructing an interaction relationship graph 912, the association relationship graph 911 and the interaction relationship graph 912 are stacked through a stacking layer 913, and finally an updated knowledge graph 914 is obtained, and the updated knowledge graph 914 contains account node representations corresponding to the account basic data and interaction node representations corresponding to the historical interaction data.

Step 805, obtaining target interaction data in a specified interaction interval in the history interaction sequence.

The content of step 805 about the target interaction data is described in detail in step 302 above, and will not be described here again.

It should be noted that, steps 8061 to 8063 are the determination process of the interactive feature representation, steps 8071 to 8072 are the determination process of the target feature representation, and in this embodiment, the two processes are performed simultaneously.

Step 8061, an account feature representation is determined based on the account node, and an interaction vector representation is determined based on the interaction node.

In this embodiment, the account node and the interaction node are input into the embedding layer to obtain an account feature representation corresponding to the account node and an interaction vector representation corresponding to the interaction node.

Step 8062, splicing the account feature representation and the interaction vector representation to obtain an interaction splicing vector.

In this embodiment, the account feature representation and the interaction vector representation are spliced left and right, and the obtained splicing result is used as the interaction splicing vector, wherein the splicing mode refers to formula sixteen, formula seventeen and formula eighteen:

formula sixteen:

seventeenth formula:

equation eighteen:

wherein W is _u 、W _i 、b _u And b _i As a parameter that can be learned,representing account node->Representing interaction vector +.>Is the initial representation of node i, i.e., the inter-splice vector.

And step 8063, extracting features of the interactive splicing vector to obtain an interactive feature representation.

In this embodiment, feature extraction is performed on the interaction stitching vector through a graph attention mechanism (Graph Attention Network, GAT) to obtain an interaction feature representation, where the feature extraction is performed by referring to formula nineteen, formula twenty and formula twenty:

Nineteenth formula:

formula twenty:

formula twenty-one:

wherein W is ₁ ，W ₂ And W is ₃ As a parameter that can be learned,and representing the feature representation in the t-th layer structure, wherein when the graph attention mechanism comprises an L-layer network structure, the sum of the feature representations corresponding to each layer of the L layers is the interaction feature representation (not shown in the formula). />

In step 8071, a target stitching vector located in the specified interaction interval in the interaction stitching vector is obtained.

Illustratively, the last vector in the inter-splice vector is obtained as the target splice vector.

And step 8072, extracting the characteristics of the target splicing vector to obtain the target characteristic representation.

Illustratively, the extraction result is obtained as the target feature representation by extracting the feature representation in the target stitching vector.

Step 808, fusing the account feature representation corresponding to the account basic data, the interactive feature representation corresponding to the history interactive data and the target feature representation corresponding to the target interactive data to obtain the fusion feature.

The details of step 808 are described in step 303 above, and will not be described here again.

And step 809, carrying out interest distribution prediction on the content in the content library based on the fusion characteristics to obtain an interest distribution prediction result.

The details of the interest distribution prediction result in step 809 are described in detail in step 304 and step 405, and are not described herein.

In this embodiment, accuracy of the account basic data and the history interaction data can be improved by constructing a target knowledge graph including isomorphic neighbors and heterogeneous neighbors and updating the account basic data and the history interaction data according to information of each node.

In an alternative embodiment, please refer to fig. 10, which illustrates a schematic diagram of a content recommendation method provided in an exemplary embodiment of the present application, as shown in fig. 10, the method includes four parts:

the first part, the graph embeds the module 1010.

The method comprises the steps of constructing an interaction sequence by virtue of an association relation between a target account and an associated account and historical interaction data 1011 corresponding to the target account and associated interaction data 1012 corresponding to the associated account, and constructing a target knowledge graph 1013 in a graph embedding module 1010, wherein the target knowledge graph 1013 comprises isomorphic neighbors (namely associated account data) and heterogeneous neighbors (namely historical interaction data and associated interaction data) corresponding to account basic data corresponding to the target account, and the isomorphic neighbors (namely historical interaction data and associated interaction data) are used for acquiring node information corresponding to each data node, calculating importance scores corresponding to each neighbor according to a heterogeneous graph neural network (Heterogenous Graph Attention Networks, HGNN), and updating and confirming account nodes 1014 corresponding to the account basic data and interaction nodes 1015 corresponding to the historical interaction data 1011.

The second part, the target account based session embedding module 1020.

After the account node 1014 and the interaction node 1015 are obtained, they are input into a target account based session embedding module 1020. First, the account node 1014 and the interaction node 1015 are input into the embedded layer, and an account feature representation 1021 corresponding to the account node 1014 and an interaction vector representation 1022 corresponding to the interaction node 1015 are obtained.

Secondly, the account feature representation 1021 and the interaction vector representation 1022 are spliced through a neural network 1023 corresponding to the graph attention mechanism to obtain an interaction splice vector 1024, and feature extraction is performed on the interaction splice vector 1024 to obtain an interaction feature representation.

Finally, in the process of obtaining the interactive mosaic vector, the interactive mosaic vector positioned at the last of the sequences is selected as the target mosaic vector 1025, and the feature extraction is performed on the target mosaic vector 1025, so as to obtain the target feature representation.

That is, the target account based session embedding module 1020 is configured to obtain an account feature representation 1021, an interaction feature representation, and a target feature representation.

Third, feature fusion module 1030.

The account feature representation 1021, the interaction feature representation and the target feature representation are spliced to obtain a spliced feature 1031, the spliced feature 1031 is fused through a high-level feature fusion network to obtain a fusion feature 1032, wherein the spliced feature 1031 is subjected to global fusion through a multi-head self-attention mechanism in the high-level feature network, and fusion results corresponding to the global fusion are subjected to local fusion through a convolution network to finally obtain the fusion feature 1032.

Fourth, prediction module 1040.

After the fusion feature 1032 is obtained, similarity calculation is performed on the content in the content library and the fusion feature 1032 through the prediction layer, so that the prediction probability corresponding to each content is obtained, the interest distribution prediction result corresponding to the content library is finally obtained, and the recommended content 1041 which is finally recommended to the target account is determined according to the interest distribution prediction result.

The beneficial effect of this scheme still includes:

(1) In the scheme, the node representation corresponding to the target account and the historical interaction data is firstly embedded and generated through the heterogeneous knowledge graph, and the association relationship between accounts and the interaction data can be modeled at the same time. A graphical attention network is then employed to model the session representation based on the corresponding target account number conversions. A high-level feature fusion module is provided for generating high-level fusion features between a target account and interactive data, including global and local fusion mechanisms. The former is the basis and the latter is the supplementary part. And finally, recommending the project by utilizing the splicing characteristics and the fusion characteristics. This helps to improve the accuracy of content recommendation.

(2) The feature fusion module in the scheme is beneficial to improving the performance of the model, and the social relationship is effectively utilized to enhance the recommendation performance.

In addition, the end-to-end training is performed in a graph embedding mode, and the embedded layer characteristic representation of the user interaction data can be calculated offline by other supervised methods. The session embedding module based on the target account uses a graph attention mechanism, and other graph neural network algorithms such as a graph convolution neural network can be used.

Fig. 11 is a block diagram of a content recommendation device according to an exemplary embodiment of the present application, and as shown in fig. 11, the device includes:

an obtaining module 1110, configured to obtain account basic data corresponding to a target account and historical interaction data, where the account basic data is used to indicate account basic information of the target account, and the historical interaction data is used to indicate a historical interaction sequence constructed by interaction data of the target account in a historical time period;

the acquiring module 1110 is further configured to acquire target interaction data in a specified interaction interval in the historical interaction sequence, where the specified interaction interval is related to an interest time distribution of the target account;

The fusion module 1120 is configured to fuse an account feature representation corresponding to the account basic data, an interaction feature representation corresponding to the historical interaction data, and a target feature representation corresponding to the target interaction data, so as to obtain a fusion feature;

and the prediction module 1130 is configured to predict an interest distribution of the content in the content library based on the fusion feature, to obtain an interest distribution prediction result, where the interest distribution prediction result is used to determine content recommended to the target account.

In an optional embodiment, the obtaining module 1110 is further configured to obtain, as the target interaction data, the last n interaction data in the historical interaction sequence, where n is a positive integer.

In an alternative embodiment, the obtaining module 1110 is further configured to obtain, as the target interaction data, the last interaction data in the historical interaction sequence.

In an alternative embodiment, the fusing module 1120 includes:

a splicing unit 1121, configured to splice an account feature representation corresponding to the account basic data, an interaction feature representation corresponding to the historical interaction data, and a target feature representation corresponding to the target interaction data, so as to obtain a spliced feature;

And the extracting unit 1122 is configured to perform feature fusion extraction on the spliced feature to obtain the fusion feature.

In an optional embodiment, the prediction module 1130 is further configured to predict an interest distribution of the content in the content library according to the stitching feature and the fusion feature, so as to obtain the interest distribution prediction result.

In an alternative embodiment, the extracting unit 1122 is further configured to input the stitching feature into a feedforward neural network to obtain a first feedforward feature vector; determining a first fusion result based on the splice feature and the first feedforward feature vector; determining an attention vector corresponding to the first fusion result through a multi-head self-attention mechanism; determining a second fusion result based on the first fusion result and the attention vector; obtaining a third fusion result based on the second fusion result and a convolution vector corresponding to the second fusion result; and obtaining the fusion characteristic based on the third fusion result and a second feedforward characteristic vector corresponding to the third fusion result.

In an alternative embodiment, the fusing module 1120 further includes:

the extracting unit 1122 is further configured to extract an account feature representation corresponding to the account basic data and an interaction vector representation corresponding to the historical interaction data;

The splicing unit 1121 is further configured to splice the account feature representation and the interaction vector representation to obtain an interaction splice vector;

the extracting unit 1122 is further configured to perform feature extraction on the interaction splicing vector to obtain the interaction feature representation.

In an alternative embodiment, the fusing module 1120 further includes:

an obtaining unit 1123, configured to obtain a target splicing vector located in the specified interaction interval in the interaction splicing vector;

the extracting unit 1122 is further configured to perform feature extraction on the target stitching vector to obtain the target feature representation.

In an alternative embodiment, the fusing module 1120 further includes:

the acquiring unit 1123 is further configured to acquire associated account data corresponding to an associated account and associated interaction data, where the associated account data is used to indicate account basic information of the associated account, and the associated interaction data includes historical interaction data corresponding to the associated account;

a determining unit 1124, configured to determine a target knowledge graph corresponding to the target account based on the account basic data, the historical interaction data, the associated account and the associated interaction data, where the target knowledge graph is used to indicate an association relationship among the account basic data, the historical interaction data, the associated account and the associated interaction data;

The determining unit 1124 is further configured to determine, based on the target knowledge graph, the account node corresponding to the account basic data and the interaction node corresponding to the historical interaction data.

In an alternative embodiment, the determining unit 1124 determines the account feature representation based on the account node and the interaction vector representation based on the interaction node.

In an optional embodiment, the determining unit 1124 is further configured to determine, based on the target knowledge graph, association node information corresponding to the association account data and interaction node information corresponding to the historical interaction data.

In summary, the embodiment of the present application provides a content recommendation device, in the process of fusing the account feature representation of the basic data of the account corresponding to the target account and the interaction feature representation of the historical interaction data corresponding to the target account, adding the target feature representation corresponding to the historical interaction sequence constructed by the interaction data of the target account in the historical time period, fusing the three features, determining the fusion feature, performing interest prediction distribution on the content based on the fusion feature, and finally determining the content recommended to the target account, and performing feature fusion by adding the interaction data including the interaction sequence, so that interest preference of the target account is easier to capture in the content recommendation process, and the accuracy of content recommendation is improved.

It should be noted that: the content recommendation apparatus provided in the above embodiment is only exemplified by the division of the above functional modules, and in practical application, the above functional allocation may be performed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules, so as to perform all or part of the functions described above. In addition, the content recommendation device and the content recommendation method provided in the foregoing embodiments belong to the same concept, and specific implementation processes of the content recommendation device and the content recommendation method are detailed in the method embodiments and are not described herein again.

Fig. 13 shows a schematic structural diagram of a server according to an exemplary embodiment of the present application. Specifically, the following is said:

the server 1300 includes a central processing unit (Central Processing Unit, CPU) 1301, a system Memory 1304 including a random access Memory (Random Access Memory, RAM) 1302 and a Read Only Memory (ROM) 1303, and a system bus 1305 connecting the system Memory 1304 and the central processing unit 1301. The server 1300 also includes a mass storage device 1306 for storing an operating system 1313, application programs 1314, and other program modules 1315.

The mass storage device 1306 is connected to the central processing unit 1301 through a mass storage controller (not shown) connected to the system bus 1305. The mass storage device 1306 and its associated computer-readable media provide non-volatile storage for the server 1300. That is, the mass storage device 1306 may include a computer readable medium (not shown) such as a hard disk or compact disc read only memory (Compact Disc Read Only Memory, CD-ROM) drive.

Computer readable media may include computer storage media and communication media without loss of generality. Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes RAM, ROM, erasable programmable read-only memory (Erasable Programmable Read Only Memory, EPROM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read Only Memory, EEPROM), flash memory or other solid state memory technology, CD-ROM, digital versatile disks (Digital Versatile Disc, DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices. Of course, those skilled in the art will recognize that computer storage media are not limited to the ones described above. The system memory 1304 and mass storage device 1306 described above may be collectively referred to as memory.

According to various embodiments of the application, the server 1300 may also operate by a remote computer connected to the network through a network, such as the Internet. I.e., the server 1300 may be connected to the network 1312 via a network interface unit 1311 coupled to the system bus 1305, or the network interface unit 1311 may be used to connect to other types of networks or remote computer systems (not shown).

The memory also includes one or more programs, one or more programs stored in the memory and configured to be executed by the CPU.

Embodiments of the present application also provide a computer device, where the computer device includes a processor and a memory, where the memory stores at least one instruction, at least one program, a code set, or an instruction set, and where the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the content recommendation method provided by the foregoing method embodiments.

Embodiments of the present application also provide a computer readable storage medium having at least one instruction, at least one program, a code set, or an instruction set stored thereon, where the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the content recommendation method provided by the foregoing method embodiments.

Embodiments of the present application also provide a computer program product, or computer program, comprising computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to perform the content recommendation method according to any of the above embodiments.

Alternatively, the computer-readable storage medium may include: read Only Memory (ROM), random access Memory (RAM, random Access Memory), solid state disk (SSD, solid State Drives), or optical disk, etc. The random access memory may include resistive random access memory (ReRAM, resistance Random Access Memory) and dynamic random access memory (DRAM, dynamic Random Access Memory), among others. The foregoing embodiment numbers of the present application are merely for describing, and do not represent advantages or disadvantages of the embodiments.

It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program for instructing relevant hardware, where the program may be stored in a computer readable storage medium, and the storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.

The foregoing description of the preferred embodiments is merely exemplary in nature and is in no way intended to limit the invention, since it is intended that all modifications, equivalents, improvements, etc. that fall within the spirit and scope of the invention.

Claims

1. A content recommendation method, the method comprising:

2. The method of claim 1, wherein the obtaining target interaction data within a specified interaction interval in the historical interaction sequence comprises:

and acquiring the latest n pieces of interaction data in the history interaction sequence as the target interaction data, wherein n is a positive integer.

3. The method of claim 2, wherein the obtaining the most recent n interaction data in the historical interaction sequence as the target interaction data comprises:

and acquiring the latest interaction data in the historical interaction sequence as the target interaction data.

4. The method according to claim 1, wherein the fusing the account feature representation corresponding to the account base data, the interaction feature representation corresponding to the historical interaction data, and the target feature representation corresponding to the target interaction data to obtain a fused feature includes:

splicing account feature representations corresponding to the account basic data, interactive feature representations corresponding to the historical interactive data and target feature representations corresponding to the target interactive data to obtain splicing features;

and carrying out feature fusion extraction on the spliced features to obtain the fusion features.

5. The method of claim 4, wherein the predicting the interest distribution of the content in the content library based on the fusion feature to obtain the interest distribution prediction result comprises:

and carrying out interest distribution prediction on the content in the content library through the splicing features and the fusion features to obtain an interest distribution prediction result.

6. The method of claim 4, wherein the performing feature fusion extraction on the stitched features to obtain the fused features comprises:

inputting the spliced characteristic into a feedforward neural network to obtain a first feedforward characteristic vector;

determining a first fusion result based on the splice feature and the first feedforward feature vector;

determining an attention vector corresponding to the first fusion result through a multi-head self-attention mechanism;

determining a second fusion result based on the first fusion result and the attention vector;

obtaining a third fusion result based on the second fusion result and a convolution vector corresponding to the second fusion result;

and obtaining the fusion characteristic based on the third fusion result and a second feedforward characteristic vector corresponding to the third fusion result.

7. The method of claim 4, wherein prior to stitching the account feature representation corresponding to the account base data, the interaction feature representation corresponding to the historical interaction data, and the target feature representation corresponding to the target interaction data, further comprising:

extracting account characteristic representations corresponding to the account basic data and interaction vector representations corresponding to the historical interaction data;

Splicing the account feature representation and the interaction vector representation to obtain an interaction splicing vector;

and extracting the characteristics of the interactive splicing vector to obtain the interactive characteristic representation.

8. The method of claim 7, wherein the method further comprises:

acquiring a target splicing vector positioned in the appointed interaction interval in the interaction splicing vector;

and extracting the characteristics of the target splicing vector to obtain the target characteristic representation.

9. The method of claim 7, wherein the extracting account feature representations corresponding to the account base data and interaction vector representations corresponding to the historical interaction data comprises:

acquiring associated account data and associated interaction data corresponding to an associated account, wherein the associated account data is used for indicating account basic information of the associated account, and the associated interaction data comprises historical interaction data corresponding to the associated account;

determining a target knowledge graph corresponding to the target account based on the account basic data, the historical interaction data, the associated account and the associated interaction data, wherein the target knowledge graph is used for indicating the association relationship among the account basic data, the historical interaction data, the associated account and the associated interaction data;

Based on the target knowledge graph, determining an account node corresponding to the account basic data and an interaction node corresponding to the historical interaction data;

the account feature representation is determined based on the account node and the interaction vector representation is determined based on the interaction node.

10. The method according to claim 9, wherein determining, based on the target knowledge graph, an account node corresponding to the account base data and an interaction node corresponding to the historical interaction data includes:

determining associated node information corresponding to the associated account data and interaction node information corresponding to the historical interaction data based on the target knowledge graph;

and determining the account node corresponding to the account basic data and the interaction node corresponding to the historical interaction data based on the associated node information and the interaction node information.

11. A content recommendation device, the device comprising:

12. A computer device comprising a processor and a memory having stored therein at least one instruction, at least one program, code set or instruction set, the at least one instruction, at least one program, code set or instruction set being loaded and executed by the processor to implement a content recommendation method according to any of claims 1 to 10.

13. A computer readable storage medium having stored therein at least one instruction, at least one program, code set, or instruction set, the at least one instruction, the at least one program, the code set, or instruction set being loaded and executed by a processor to implement the content recommendation method of any one of claims 1 to 10.

14. A computer program product comprising a computer program or instructions which, when executed by a processor, implement a content recommendation method as claimed in any one of claims 1 to 10.