CN114996561A

CN114996561A - Information recommendation method and device based on artificial intelligence

Info

Publication number: CN114996561A
Application number: CN202110231593.2A
Authority: CN
Inventors: 徐程程
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2021-03-02
Filing date: 2021-03-02
Publication date: 2022-09-02
Anticipated expiration: 2041-03-02
Also published as: CN114996561B

Abstract

The application provides an information recommendation method and device based on artificial intelligence; relates to the artificial intelligence technology, and the method comprises the following steps: extracting user behavior characteristics from behavior data of a plurality of interactive media accounts corresponding to a user; performing coding processing based on the user behavior characteristics to obtain a user characteristic vector; determining a plurality of recalled media accounts satisfying similar conditions with the user feature vector; and generating information to be recommended based on the plurality of recalled media accounts, and executing recommendation operation corresponding to the user based on the information to be recommended. By the method and the device, the user interest can be sufficiently mined to improve the recommendation accuracy.

Description

Information recommendation method and device based on artificial intelligence

Technical Field

The present application relates to artificial intelligence technologies, and in particular, to an artificial intelligence based information recommendation method and apparatus.

Background

Artificial Intelligence (AI) is a theory, method and technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human Intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results.

Information recommendation is an important application of artificial intelligence, and a plurality of strategies and models are generally processed in parallel in a recall process in a recommendation system, for example, relevant information (also called materials, including articles and videos) is searched based on user figures, similar information is searched based on recent clicks of users, and trending information is searched. In most information flow scenes, different types of information are displayed in a mixed mode, for example, articles and videos appear alternately, different information can be recalled respectively by a recall algorithm in the related technology, namely, the recall complexity is greatly increased when various types of information are faced, and the interests of users in different types of information cannot be fused, so that the information recommended by the method cannot meet the rich interests of the users, and poor experience is caused to the users.

Therefore, an effective scheme for fusing various types of information in which the user is interested to make accurate recommendation is lacking in the related art.

Disclosure of Invention

The embodiment of the application provides an information recommendation method and device based on artificial intelligence, electronic equipment and a computer-readable storage medium, which can fully mine user interest to improve recommendation accuracy.

The technical scheme of the embodiment of the application is realized as follows:

the embodiment of the application provides an information recommendation method based on artificial intelligence, which comprises the following steps:

extracting user behavior characteristics from behavior data of a plurality of interactive media accounts corresponding to a user; coding processing is carried out based on the user behavior characteristics to obtain user characteristic vectors; determining a plurality of recalled media accounts satisfying similar conditions with the user feature vector; and generating information to be recommended based on the plurality of recalled media accounts, and executing recommendation operation corresponding to the user based on the information to be recommended.

The embodiment of the application provides an information recommendation device based on artificial intelligence, includes:

the extraction module is used for extracting user behavior characteristics from behavior data of a plurality of interactive media accounts corresponding to a user; the coding module is used for carrying out coding processing based on the user behavior characteristics to obtain user characteristic vectors; the recall module is used for determining a plurality of recalled media accounts which meet similar conditions with the user feature vector; and the recommending module is used for generating information to be recommended based on the recalling media accounts and executing recommending operation corresponding to the user based on the information to be recommended.

In the above scheme, the extracting module is further configured to map behavior data of the user for a plurality of pieces of information to behavior data of the user for an interactive media account publishing the information; wherein the user's behavior data for a plurality of information characterizes at least one of the following behaviors: and the user pays attention to the behavior of the information published by the interactive media account, and the user subscribes to the behavior of the information published by the interactive media account.

In the above scheme, the encoding module is further configured to determine a user collaborative filtering feature and a user graph feature, sequentially connect the user collaborative filtering feature, the user graph feature and the user behavior feature obtained through extraction, and use a new user behavior feature obtained through connection as the user behavior feature for performing the encoding processing; wherein the user collaborative filtering feature is determined by decomposing a matrix including the behavior data, and the user graph feature is determined based on mapping a directed graph including the behavior data.

In the above scheme, the encoding module is further configured to extract a user attribute feature from the attribute data of the user; rectifying the user behavior characteristics and the user attribute characteristics, or rectifying the user behavior characteristics, normalizing an obtained rectification result, and taking an obtained normalization processing result as a user characteristic vector obtained by coding; wherein the user attribute characteristics include at least one of: the age characteristic of the user and the regional characteristic of the user; the user behavior characteristics include at least one of: the interactive media account characteristics of the user, the tag characteristics of the interactive media account, and the channel characteristics of the interactive media account.

In the above scheme, the recall module is further configured to obtain media account feature vectors of multiple candidate media accounts, determine cosine distances between the media account feature vectors of the multiple candidate media accounts and the user feature vectors, and use the cosine distances as similarity; and determining a plurality of candidate media accounts with similarity exceeding a similarity threshold value with the user feature vector as a plurality of recalled media accounts meeting a similarity condition.

In the foregoing solution, the recall module is further configured to, for each candidate media account of the multiple candidate media accounts, perform the following processing: extracting media account features from the attribute data of the candidate media accounts; rectifying the media account features, normalizing the obtained rectifying result, and taking the obtained normalizing result as a media account feature vector; wherein the media account characteristics include at least one of: the system comprises a media account number identification system, a media account number identification system and a media account number identification system, wherein the media account number identification system comprises a tag characteristic of the media account number, a channel characteristic of the media account number and a fan quantity characteristic of the media account number.

In the above scheme, the recall module is further configured to determine a media account collaborative filtering feature and a media account map feature of the candidate media account, connect the media account collaborative filtering feature, the media account map feature and the extracted media account feature, and use a new media account feature obtained by the connection as a media account feature for performing the rectification processing; the media account collaborative filtering feature is determined by decomposing a matrix including the behavior data, and the media account graph feature is determined by mapping based on a directed graph including the behavior data.

In the above scheme, the recommending module is further configured to generate information to be recommended by at least one of the following methods: generating information to be recommended for recommending the media accounts based on the plurality of recalled media accounts; generating information to be recommended for recommending the information published by the media accounts based on the information published by the plurality of recalled media accounts; predicting to obtain a score of the information to be recommended based on the information feature vector of the information to be recommended, the user feature vector of the user and the cross feature vector of the user and the information to be recommended, wherein the score represents the similarity between the information to be recommended and the user; based on scores of a plurality of candidate information to be recommended, sorting the plurality of candidate information to be recommended in a descending order; and performing diversity ranking on the previously ranked candidate information to be recommended, and executing recommendation operation corresponding to the user based on a diversity ranking result.

In the above scheme, when the information to be recommended is published by the plurality of recalled media accounts, the recommending module is further configured to delete the information from the same media account one by one according to an ascending order of scores when the number of information from the same media account in the plurality of candidate information to be recommended is greater than a first threshold value, until the number of information from the same media account does not exceed the first threshold value; when the number of information from the same channel in the same media account in the plurality of candidate information to be recommended is larger than a second threshold value, deleting the information from the same media account one by one according to the ascending order of scores until the number of information from the same media account does not exceed the second threshold value; and when the number of the non-high-quality information in the candidate information to be recommended is larger than a third threshold value, deleting the non-high-quality information one by one according to the ascending order of scores until the number of the non-high-quality information does not exceed the third threshold value.

An embodiment of the present application provides an electronic device, including:

a memory for storing executable instructions;

and the processor is used for realizing the artificial intelligence based information recommendation method provided by the embodiment of the application when the executable instructions stored in the memory are executed.

The embodiment of the application provides a computer-readable storage medium, which stores executable instructions and is used for realizing the artificial intelligence-based information recommendation method provided by the embodiment of the application when being executed by a processor.

The embodiment of the application has the following beneficial effects:

the behavior characteristics of the user are extracted from the behavior data of the user corresponding to the multiple interactive media accounts, and the recalled media accounts are determined based on the user behavior, namely the media accounts meeting the user interest are recalled through the interactive behavior of the user on the media accounts, so that the information published by the user on the media accounts is favorably and fully mined, the information recommendation based on the recalled media accounts can meet the user interest, and the recommendation accuracy is effectively improved.

Drawings

FIG. 1 is a schematic diagram of an architecture of an artificial intelligence-based information recommendation system 100 provided by an embodiment of the present application;

fig. 2A is a schematic structural diagram of a server 200 according to an embodiment of the present application;

FIG. 2B is a schematic structural diagram of a neural network model provided in an embodiment of the present application;

FIG. 3A is a flowchart illustrating an artificial intelligence based information recommendation method according to an embodiment of the present application;

FIG. 3B is a flowchart illustrating an artificial intelligence based information recommendation method according to an embodiment of the present application;

FIG. 3C is a flowchart illustrating an artificial intelligence based information recommendation method according to an embodiment of the present application;

FIG. 4 is a schematic matrix decomposition diagram provided in an embodiment of the present application;

FIG. 5 is a schematic diagram of a user behavior sequence provided by an embodiment of the present application;

fig. 6 is a schematic diagram of a directed graph of an interactive media account provided in an embodiment of the present application;

fig. 7 is a sequence diagram of an interactive media account generated by random walk according to an embodiment of the present application;

FIG. 8 is a schematic structural diagram of a double-tower model for implementing an artificial intelligence-based information recommendation method according to an embodiment of the present application;

fig. 9 is a schematic view of an application scenario of an artificial intelligence based information recommendation method according to an embodiment of the present application.

Detailed Description

In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be described in further detail with reference to the attached drawings, the described embodiments should not be considered as limiting the present application, and all other embodiments obtained by a person of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict.

In the following description, references to the terms "first \ second \ third" are only to distinguish similar objects and do not denote a particular order, but rather the terms "first \ second \ third" are used to interchange specific orders or sequences, where appropriate, so as to enable the embodiments of the application described herein to be practiced in other than the order shown or described herein.

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

Before further detailed description of the embodiments of the present application, terms and expressions referred to in the embodiments of the present application will be described, and the terms and expressions referred to in the embodiments of the present application will be used for the following explanation.

1) Media accounts (CP) refer to accounts for publishing articles or videos in an information stream product, some accounts belong to official accounts for publishing news information, such as a news public number and a newspaper public number, and some accounts belong to self-media accounts for publishing articles in a specific field, such as an entertainment character public number and an artificial intelligence public number.

2) And (4) collaborative filtering, wherein the interested information of the user is recommended by utilizing the preference of a group with mutual interest and common experience.

3) The graph comprises a directed graph and an undirected graph, and the graph formed by directed edges is called the directed graph.

4) The number of edge strips of a vertex in the directed graph is called the out-degree of the vertex.

5) A Linear rectification Function (ReLU), also called a modified Linear Unit, is an Activation Function (AF) commonly used in an artificial neural network, and generally refers to a nonlinear Function represented by a ramp Function and its variants.

6) One-Hot encoding, also known as One-bit-efficient encoding, mainly uses an N-bit state register to encode N states, each state being represented by its own independent register bit and having only One bit active at any time.

Information flow recommendation products in related technologies mainly recommend different information including articles, videos and the like by utilizing a personalized recommendation technology according to different interests of users. Generally, the whole recommendation process is divided into two parts of recall and sorting. The recall aim is to select partial information which is possibly interested by a user from a massive information candidate pool, and the recalled information is combined through various strategies to be used as the input of the sorting module. The sorting module is mainly used for sorting the recalled information, sorting the recalled information by using information characteristics, user characteristics and cross characteristics through an output result of the scoring model, and selecting an article or a video with the highest score to recommend the article or the video to a user. The recalling process generally adopts a plurality of strategies and models for parallel processing, such as searching relevant information based on the user portrait, searching similar information based on the recent click of the user, searching popular information and the like. The recall strategy related to the related art is found in the embodiment of the application to have the following technical problems: 1) when different types of information are displayed in a mixed mode, for example, articles and videos appear alternately, different information can be recalled respectively by a recall algorithm in the related technology, the recall complexity is greatly increased, and the interests of users in different types of information cannot be fused. 2) In the media number-based recall strategy, most of the media numbers concerned by the user are directly recalled by using the concerned behaviors of the user, and when the concerned behaviors of the user are few, for example, the concerned behaviors of a new user are generally few, the recalled media numbers publish less information, and the interest requirements of the user cannot be met. 3) In the related technology, the media numbers which are popular in a current period of time can be recalled and recommended to the user, and all users are covered by the popular media numbers, so that the recommended information deviates from the user interest, and bad experience is brought to the user.

In view of the foregoing technical problems, embodiments of the present application provide an artificial intelligence based information recommendation method, apparatus, electronic device, and computer-readable storage medium, which can merge interests of a user in multiple types of information by recalling a media account to improve recommendation accuracy. In the following, an exemplary application will be explained when the electronic device is implemented as a server.

Referring to fig. 1, fig. 1 is an architectural schematic diagram of an artificial intelligence based information recommendation system 100 according to an embodiment of the present application, where the information recommendation system may be used to support recommendation scenes of various information, such as an application scene for recommending articles, an application scene for recommending videos, an application scene for recommending media numbers of published articles and videos, and the like, and according to different application scenes, the information may be the articles published by media accounts, videos published by media accounts, introduction information of media accounts, and the like, in the information recommendation system, a terminal 400 is connected to a server 200 through a network 300, and the network 300 may be a wide area network or a local area network, or a combination of the two.

In some embodiments, the functions of the information recommendation system are implemented based on each module in the server 200, in a process that a user uses a client, the terminal 400 uses collected behavior data of the user corresponding to a plurality of interactive media accounts as training sample data, the training sample data is collected behavior data of different users of each terminal, a neural network model is trained based on the obtained training data, and the trained neural network model is integrated in the server, wherein the neural network model includes an extraction module 2551 and an encoding module 2552; an extraction module 2551 in the server 200 extracts user behavior characteristics from behavior data of a plurality of interactive media accounts corresponding to a user, obtains a user characteristic vector through an encoding module 2552, and determines a plurality of recall media accounts satisfying similar conditions with the user characteristic vector through a recall module 2553; the recommendation module 2554 generates information to be recommended based on a plurality of recalled media accounts, performs diversified sorting processing on the information to be recommended, and executes recommendation operation of a corresponding user based on a diversified sorting result.

In some embodiments, the server 200 may be an independent physical server, may also be a server cluster or a distributed system formed by a plurality of physical servers, and may also be a cloud server providing basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a network service, cloud communication, a middleware service, a domain name service, a security service, a CDN, and a big data and artificial intelligence platform. The terminal may be, but is not limited to, a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, and the like. The terminal and the server may be directly or indirectly connected through wired or wireless communication, and the embodiment of the present application is not limited.

Next, a structure of an electronic device for implementing an artificial intelligence based information recommendation method according to an embodiment of the present application is described, and as described above, the electronic device according to the embodiment of the present application may be the server 200 in fig. 1. Referring to fig. 2A, fig. 2A is a schematic structural diagram of a server 200 according to an embodiment of the present application, where the server 200 shown in fig. 2A includes: at least one processor 210, memory 250, at least one network interface 220. The various components in server 200 are coupled together by a bus system 240. It is understood that the bus system 240 is used to enable communications among the components. The bus system 240 includes a power bus, a control bus, and a status signal bus in addition to a data bus. For clarity of illustration, however, the various buses are labeled as bus system 240 in fig. 2A.

The Processor 210 may be an integrated circuit chip having Signal processing capabilities, such as a general purpose Processor, a Digital Signal Processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like, wherein the general purpose Processor may be a microprocessor or any conventional Processor, or the like.

The memory 250 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid state memory, hard disk drives, optical disk drives, and the like. Memory 250 optionally includes one or more storage devices physically located remotely from processor 210.

The memory 250 includes volatile memory or nonvolatile memory, and may include both volatile and nonvolatile memory. The nonvolatile memory may be a Read Only Memory (ROM), and the volatile memory may be a Random Access Memory (RAM). The memory 250 described in embodiments herein is intended to comprise any suitable type of memory.

In some embodiments, memory 250 is capable of storing data, examples of which include programs, modules, and data structures, or a subset or superset thereof, to support various operations, as exemplified below.

An operating system 251 including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks; a network communication module 252 for communicating to other computing devices via one or more (wired or wireless) network interfaces 220, exemplary network interfaces 220 including: bluetooth, wireless compatibility authentication (WiFi), and Universal Serial Bus (USB), among others.

In some embodiments, the artificial intelligence based information recommendation device provided by the embodiments of the present application may be implemented in software, and fig. 2A illustrates an artificial intelligence based information recommendation device 255 stored in a memory 250, which may be software in the form of programs and plug-ins, and the like, and includes the following software modules: extraction module 2551, encoding module 2552, recall module 2553 and recommendation module 2554, which are logical and therefore can be combined arbitrarily or further split depending on the functions implemented. The functions of the respective modules will be explained below.

The artificial intelligence based information recommendation method provided by the embodiment of the present application will be described below in conjunction with an exemplary application and implementation of the server 200 provided by the embodiment of the present application. Referring to fig. 2B, fig. 2B is a schematic structural diagram of a neural network model provided in the embodiment of the present application, and the neural network model may be a two-tower model, and includes a user side, a media account side, and a prediction layer, where structures of the user side and the media account side are similar, both the user side and the media account side include an embedding layer and an encoding layer, and the encoding layer includes a linear rectification layer and a normalization layer; the prediction layer can be used for updating parameters of each layer in a training stage and can be used for calculating the similarity between the user characteristic vector and the media account number vector in an application stage so as to determine the media account number vector meeting similar conditions with the user characteristic vector and further determine the media account number corresponding to the media account number vector. Taking a recommendation task as an example of recommending a media account, completing the recommendation task based on each layer in a trained neural network model, namely responding to the recommendation task initiated by a terminal, acquiring behavior data of a user sent by the terminal aiming at the media account, and extracting user behavior characteristics from the behavior data through an embedded layer at a user side; rectifying the user behavior characteristics through a linear rectifying layer at the user side, and normalizing the rectifying processing result through a normalization layer to obtain a user characteristic vector; similarity prediction is carried out on the user characteristic vectors and a plurality of media accounts stored in an offline media account side through a prediction layer, a plurality of recalled media accounts meeting similar conditions with the user characteristic vectors are determined, and the media accounts to be recommended are sent to a terminal user.

In some embodiments, taking the neural network model as a double-tower model as an example, referring to fig. 8, fig. 8 is a schematic structural diagram of the double-tower model for implementing the artificial intelligence based information recommendation method provided in the embodiments of the present application. The training process of the double-tower model can be realized by the following steps: taking the combination of the user and the media account as a sample set for training a double-tower model; carrying out forward propagation on each layer and a prediction layer in a user side of a double-tower model by taking a user as a sample to obtain a user characteristic vector; performing forward propagation on each layer and a prediction layer in a media account side of a double-tower model by taking a media account corresponding to a user as a sample to obtain a media account feature vector; determining the prediction similarity of the user characteristic vector and the media account characteristic vector; initializing a loss function comprising a prediction similarity for each sample and the corresponding sample; and determining the error between the predicted similarity and the real similarity of each sample, reversely propagating the error in the double-tower model according to the loss function to determine the change value of the double-tower model when the loss function obtains the minimum value, and updating the parameters of the double-tower model according to the change value. The prediction layer may be implemented by a scoring function, for example, the scoring function may be cosine similarity.

The form of the training sample of the model is a combined label of the user and the media account, that is, when the media account is a media account with interactive behavior with the user, the real similarity of the combined sample of the user and the media account is 1, and when the media account is a media account with no interactive behavior with the user, the real similarity of the combined sample of the user and the media account is 0; the loss function may be a cross entropy loss function, a square loss function, or the like.

In some examples, the user-side embedding layer may include only identification features of the user that characterize the user's behavior, and the media account-side embedding layer may include only media account identification features that characterize the user's behavior.

In some examples, the user-side embedding layer may include user behavior characteristics, the user behavior characteristics include at least one of an identification characteristic of a user, an interactive media account characteristic of the user, a tag characteristic of an interactive media account, and a channel characteristic of the interactive media account, and the media account side embedding layer may include media account characteristics, the media account characteristics include at least one of an identification characteristic of a media account, a tag characteristic of a media account, a channel characteristic of a media account, and a fan count characteristic of a media account. Each user's identification feature and media account identification feature have a fixed dimensional embedded representation when initialized. As an example, the embedding layer on the user side may include user behavior characteristics and user attribute characteristics, the user attribute characteristics including at least one of age characteristics of the user and geographic characteristics of the user.

The following describes an artificial intelligence based information recommendation method provided by the embodiment of the present application, taking a method for executing the information recommendation system provided by the embodiment of the present application by the server 200 in fig. 1 as an example, where the information recommendation system includes a training phase and an application phase. First, an application of a model in the artificial intelligence based information recommendation method provided in the embodiment of the present application is explained. Referring to fig. 3A, fig. 3A is a schematic flowchart of an artificial intelligence-based information recommendation method provided by an embodiment of the present application, and will be described with reference to steps 101 to 105 shown in fig. 3A.

In step 101, user behavior features are extracted from behavior data of a user corresponding to a plurality of interactive media accounts. The behavior data of the user corresponding to the multiple interactive media accounts is behaviors of the user such as clicking or watching the media accounts.

In some embodiments, the extracting of the user behavior feature from the behavior data of the user corresponding to the multiple interactive media accounts may be implemented by an embedding layer in the neural network model as shown in fig. 2B, that is, extracting a low-dimensional embedded representation from high-dimensional raw data by the embedding layer in the trained neural network model, where the low-dimensional embedded representation includes at least one dimension data (e.g., a tag feature of an interactive media account, a channel feature of an interactive media account, etc.), when the embedded representation is one dimension data, the embedded representation is the user behavior feature, when the embedded representation is multiple dimension data, the multiple dimension data are spliced, and the spliced embedded representation is the user behavior feature.

In step 102, encoding processing is performed based on the user behavior characteristics to obtain a user characteristic vector.

In some embodiments, the encoding process is performed based on the user behavior characteristics to obtain the user characteristic vector, which may be implemented as follows: rectifying the user behavior characteristics, normalizing the obtained rectifying result, and taking the obtained normalizing result as a user characteristic vector obtained by coding; wherein the user behavior characteristics include at least one of: the interactive media account characteristics of the user, the tag characteristics of the interactive media account, and the channel characteristics of the interactive media account.

In some examples, referring to fig. 2B, the encoding process is implemented by a user-side linear rectification layer and a normalization layer in the neural network model in fig. 2B. The linear rectification layer at the user side is used for carrying out linear rectification processing on the user behavior characteristics output by the embedded layer at the user side through the neural network model, and the linear rectification processing can be realized through a ReLU function; then, normalization processing is carried out through a normalization layer at the user side, normalization factors are the number of the neurons of the normalization layer, and the convergence rate of the neural network model can be improved through normalization.

In some embodiments, the encoding process is performed based on the user behavior characteristics to obtain the user characteristic vector, which may be implemented as follows: when the data volume of the behavior data of the user corresponding to the multiple interactive media accounts is smaller than a data volume threshold value or the service lives of the behavior data of the user corresponding to the multiple interactive media accounts are not within a valid period range, extracting user attribute features from the attribute data of the user; and performing rectification processing on the characteristics obtained after splicing and fusing the user behavior characteristics and the user attribute characteristics, performing normalization processing on the obtained rectification result, and taking the obtained normalization processing result as a user characteristic vector obtained by coding.

In the embodiment of the application, when the behavior data of the plurality of interactive media accounts corresponding to the user is insufficient or exceeds the period, the attribute characteristics are supplemented so as to make up for the defect that the learned behavior characteristics of the user are insufficient based on the behavior data with insufficient or exceeding data amount, and improve the learning precision of the neural network model; and encoding the characteristics obtained after splicing and fusing the user behavior characteristics and the user attribute characteristics, so that the user characteristic vector obtained by encoding can better map the interest and the demand of the user.

In step 103, a plurality of recalled media accounts satisfying similar conditions as the user feature vector are determined.

In some embodiments, referring to fig. 3B, fig. 3B is a flowchart of an artificial intelligence based information recommendation method provided in an embodiment of the present application, which illustrates step 103 in fig. 3A, and may also be implemented by performing steps 1031 to 1033. The description will be made in conjunction with the respective steps.

In step 1031, media account feature vectors of multiple candidate media accounts are obtained.

In some examples, obtaining the media account feature vectors of a plurality of candidate media accounts may be implemented by: for each candidate media account number of the plurality of candidate media account numbers, performing the following: extracting media account features from attribute data of the candidate media accounts; rectifying the media account features, normalizing the obtained rectifying result, and taking the obtained normalizing result as a media account feature vector; wherein the media account characteristics include at least one of: the system comprises the tag characteristics of the media account, the channel characteristics of the media account and the fan number characteristics of the media account.

In some examples, referring to fig. 2B, the processing performed for each of the plurality of candidate media accounts is implemented by an embedding layer, a linear rectification layer, and a normalization layer on the media account side in the neural network model in fig. 2B. The embedded layer at the media account side is used for extracting media account features from the attribute data of the candidate media accounts; the linear rectification layer at the media account side is used for carrying out linear rectification processing on the media account characteristics output by the embedding layer, and the linear rectification processing can be realized by a ReLU function; then, normalization processing is carried out through a normalization layer on the media account side, normalization factors are the number of neurons on the layer, and the convergence speed of the double-tower model can be improved through normalization.

It should be noted here that the candidate media account may only include a media account (i.e., an interactive media account) that has an interaction with the user, and as an example, in a training stage of the neural network model, that is, when a forward operation is performed on the interactive media account, a media account vector of the interactive media account is stored offline and stored in the vector database. In the online application stage, when a media account number which meets similar conditions with the user characteristic vector is determined to be recalled, a vector database is directly utilized to retrieve a plurality of media account numbers with the closest similarity, and relevant recommendation is carried out.

In the embodiment of the application, when the candidate media accounts only include the interactive media account, the media accounts can be stored offline, so that the media accounts with the same interest as the user can be quickly found based on the user characteristic vector.

In some embodiments, the recalled candidate media accounts may include media accounts with which there is interaction with the user (i.e., interactive media accounts) and media accounts with which there is no interaction with the user (i.e., non-interactive media accounts); the recalled candidate media accounts may also include only media accounts for which there is no interaction with the user.

Taking the example that the candidate media accounts only include the non-interactive media accounts, user behavior characteristics are extracted from behavior data for the interactive media accounts 1-10, and during a recall stage, recall is performed from the media accounts 11-100 to obtain a plurality of recalled media accounts. Taking the example that the candidate media accounts include the interactive media account and the non-interactive media account, user behavior characteristics are extracted from behavior data for the interactive media accounts 1-10, and during a recall stage, the media accounts 1-100 are recalled to obtain a plurality of recalled media accounts.

In the embodiment of the application, when the candidate media accounts only include the non-interactive media accounts, the characteristics are learned from the interactive media accounts and reused in the media accounts without interaction with the user, so that fresh media accounts close to the user interest can be effectively mined.

In step 1032, cosine distances between the media account feature vectors of the candidate media accounts and the user feature vectors are determined, and the cosine distances are used as the similarity.

It should be noted that, in some examples, the similarity may also be calculated by a pearson correlation coefficient, a mahalanobis distance, a euclidean distance, or the like.

In step 1033, a plurality of candidate media accounts with similarity exceeding the similarity threshold with the user feature vector are determined as a plurality of recalled media accounts satisfying the similarity condition.

In some examples, multiple recalled media accounts meeting similar conditions may also be determined by: determining the similarity between the candidate media account vectors and the user characteristic vector, and taking the candidate media account corresponding to the candidate media account vector with the highest similarity as a plurality of recalled media accounts meeting the similarity condition.

For example, the candidate media account vectors with the highest similarity rank may be obtained by a top number or a top proportion, for example, 50 top candidate media account vectors are obtained, or two percent of the total number of all the media account vectors are obtained. In step 104, information to be recommended is generated based on the plurality of recalled media accounts.

In some embodiments, the information to be recommended may be generated by at least one of: generating information to be recommended for recommending the media account number based on the plurality of recalled media account numbers; and generating information to be recommended for recommending the published information of the media accounts based on the published information of the plurality of recalled media accounts.

In some examples, the information to be recommended for recommending the media account is generated based on a plurality of recalled media accounts, that is, the recalled media accounts are directly used as the information to be recommended, for example, public numbers for publishing articles and videos. And generating information to be recommended for recommending the information published by the media accounts based on the information published by the recalled media accounts, namely performing diversity ranking processing on the information published by the recalled media accounts, and taking the information published by the recalled media accounts after the diversity ranking processing as the information to be recommended, such as articles, videos and the like.

In step 105, a recommendation operation of the corresponding user is executed based on the information to be recommended.

In some embodiments, referring to fig. 3C, fig. 3C is a flowchart illustrating an artificial intelligence based information recommendation method provided in an embodiment of the present application, which shows step 105 in fig. 3A, and can also be implemented by performing step 1051 to step 1053. The description will be made in conjunction with the respective steps.

In step 1051, a scoring model is called to predict and obtain a score of the information to be recommended based on the information feature vector of the information to be recommended, the user feature vector of the user, and the cross feature vector of the user and the information to be recommended, wherein the score represents the similarity between the information to be recommended and the user.

In some examples, a scoring model used in the artificial intelligence-based information recommendation method provided in the embodiment of the present application and training of the scoring model are described, where the scoring model includes a feature extraction module, a fusion coding module, and a prediction module. The training process of the scoring model can be realized by the following modes: taking the combination of the user and the information to be recommended as a sample set for training a double-tower model, and initializing a loss function comprising the prediction score of each sample and the corresponding sample; extracting information characteristics, user characteristics of the user and cross characteristics of the user and the information to be recommended from the combined sample through a characteristic extraction module; the information characteristic vector of the information to be recommended, the user characteristic vector of the user and the cross characteristic vector of the information to be recommended are obtained by coding the information characteristic of the information to be recommended, the user characteristic vector of the user and the cross characteristic vector of the information to be recommended through a fusion coding module, and the information characteristic vector of the information to be recommended, the user characteristic vector of the user and the cross characteristic vector of the information to be recommended are subjected to fusion processing, for example, the information characteristic vector of the information to be recommended, the user characteristic vector of the user and the cross characteristic vector of the information to be recommended are connected through full connection; and predicting the connection result through a prediction module to obtain the score of the information to be recommended in a prediction mode, determining the error between the predicted score and the real score of each sample, reversely propagating the error in the scoring model according to the loss function to determine the change value of the scoring model when the loss function obtains the minimum value, and updating the parameters of the scoring model according to the change value.

It should be noted that the prediction model may be implemented by a logistic regression function softmax or a cosine similarity function, and when implemented by the cosine similarity function, the score may represent the similarity between the information to be recommended and the user. And the cross characteristics of the user and the information to be recommended are obtained by simply splicing the original data of the user and the information to be recommended and carrying out One-Hot coding on the splicing result.

In the embodiment of the application, the scoring model is realized through a neural network model. In the neural network model, the form of the training sample is a combined label of the user and the information to be recommended, that is, when the information to be recommended is information clicked or watched by the user, the real similarity of the combined sample of the user and the recommended information is 1, and when the information to be recommended is information not clicked or watched by the user, the real similarity of the combined sample of the user and the recommended information is 0; the loss function may be a cross entropy loss function, a square loss function, or the like.

In step 1052, the multiple candidate information to be recommended are sorted in descending order based on the scores of the multiple candidate information to be recommended.

In step 1053, the information to be recommended of the plurality of candidates ranked at the top is subjected to diversity ranking, and the recommendation operation of the corresponding user is executed based on the diversity ranking result. Here, the recommendation operation for the user is: and sending a plurality of pieces of information to be recommended which are sorted according to the sorting result to the user terminal.

In some examples, when the information to be recommended is published information of a plurality of recalled media accounts, the diversity ranking of the top candidate objects can be realized by at least one of the following methods: when the number of information from the same media account in the candidate information to be recommended is larger than a first threshold value, deleting the information from the same media account one by one according to the ascending order of scores until the number of the information from the same media account does not exceed the first threshold value; when the number of the information from the same channel in the same media account in the plurality of candidate information to be recommended is larger than a second threshold value, deleting the information from the same media account one by one according to the ascending order of scores until the number of the information from the same media account does not exceed the second threshold value; and when the number of the non-high-quality information in the candidate information to be recommended is larger than a third threshold value, deleting the non-high-quality information one by one according to the ascending order of the scores until the number of the non-high-quality information does not exceed the third threshold value. It should be noted that the first threshold, the second threshold, and the third threshold may be determined according to the number of the total amount of the recalled information to be recommended.

Taking the information to be recommended as an article example, when the number of articles from the same public number in a plurality of candidate articles is more than 50, deleting the articles from the same public number one by one in the ascending order of article scores until the number of articles from the same public number does not exceed 50, and ensuring that the number of articles from the same public number in the plurality of candidate articles is 50 at most.

In the embodiment of the application, information to be recommended is subjected to diversity sequencing, the diversity of the information to be recommended is ensured, the problem that the head of the information to be recommended has an aggregation effect is solved, and the information to be recommended is ensured to be dispersed in channels and labels of different media accounts, namely, the recalled information to be recommended is ensured to be derived from a plurality of media accounts and various media accounts, and the media accounts are given balanced exposure opportunities; and meanwhile, the quality of the information to be recommended is also ensured.

Since the user behavior usually occurs between the user and the media information (e.g., articles, videos, etc.), the embodiment of the present application maps the behavior data of the user for the media information to the behavior data of the user for the media account, for example, if the user u clicks the article d, which is issued by the media account c, then the user u has an interactive behavior with the media account c. In some embodiments, before extracting user behavior characteristics from behavior data of a plurality of interactive media accounts corresponding to a user, behavior data of the plurality of interactive media accounts corresponding to the user may also be acquired, that is, behavior data of the user for a plurality of pieces of information is mapped to behavior data of the user for interactive media accounts publishing information; wherein the behavior data of the user for the plurality of information characterizes at least one of the following behaviors: and the user pays attention to the behavior of the information published by the interactive media account, and the user subscribes to the behavior of the information published by the interactive media account.

In some embodiments, when the obtained data amount of the behavior data of the user corresponding to the multiple interactive media accounts is insufficient, the behavior of the user cannot be accurately characterized by the user behavior characteristics obtained based on the behavior data, so that the following processing may be performed before the encoding processing is performed based on the user behavior characteristics to obtain the user characteristic vector, so as to obtain the user behavior characteristics capable of accurately reflecting the user behavior: determining user collaborative filtering characteristics and user graph characteristics, sequentially connecting the user collaborative filtering characteristics, the user graph characteristics and the user behavior characteristics obtained through extraction, and taking the new user behavior characteristics obtained through connection as the user behavior characteristics for coding; the user collaborative filtering characteristics are determined by decomposing a matrix comprising the behavior data, and the user graph characteristics are determined by mapping based on a directed graph comprising the behavior data.

In some examples, the collaborative filtering characteristics may be determined by: constructing an interaction matrix corresponding to each interactive media account by each user based on behavior data of the user aiming at a plurality of interactive media accounts; and decomposing the interaction matrix to obtain user collaborative filtering characteristics representing user behaviors and media account collaborative filtering characteristics representing user behaviors.

For example, referring to fig. 4, fig. 4 is a schematic matrix decomposition diagram provided in the embodiments of the present application. Wherein 401 is an interaction matrix, 402 is a user collaborative filtering feature, and 403 is a media account collaborative filtering feature. The interaction matrix 401 of the users and the media accounts is a 3 × 5 matrix, representing that the number of the users is 3, the number of the media accounts is 5, and by matrix decomposition, the high-dimensional interaction matrix is decomposed into two low-dimensional matrices, that is, the user matrix 402 is a 3 × 6 matrix, which can be understood as vector representation of 3 users, the vector representation of the 3 users is the user collaborative filtering feature, the media account matrix 403 is a 6 × 5 matrix, which can be understood as vector representation of 5 media accounts, and the vector representation of the 5 media accounts is the media account collaborative filtering feature.

In the embodiment of the application, a high-dimensional interaction matrix is subjected to matrix decomposition, dense user collaborative filtering characteristics and media account collaborative filtering characteristics can be obtained, the problem of data sparseness is well solved, and the user and the media account are in the same vector space, so that recall processing is performed on user behavior characteristics determined based on the user collaborative filtering characteristics and the media account collaborative filtering characteristics in the following process, and the recall precision is higher.

In some examples, the user graph characteristics may be determined by: acquiring a behavior sequence of each user in a time window; the behavior sequence represents the sequence of interaction behaviors of the user and the media account in a time window; the method comprises the steps of taking media accounts with interactive behaviors with users as vertexes, taking turns between the media accounts involved in a behavior sequence as edges between the vertexes, constructing a directed graph of the interactive media accounts of each user, generating a sequence of the interactive media accounts based on the directed graph, obtaining an embedded representation of each interactive media account based on the sequence of each interactive media account, and taking the embedded representation as a user graph feature.

For example, referring to fig. 5, fig. 5 is a schematic diagram of a user behavior sequence provided in the embodiment of the present application. Since the interest of a user may change with time, the behavior of the user within a set time window is taken as a user behavior sequence. The sequence of three users' behavior within a certain time window is shown in fig. 5, with 1-5 representing different media accounts. Then, a directed graph is constructed for the sequence of user behaviors. Referring to fig. 6, fig. 6 is a schematic diagram of an interactive media account directed graph provided in the embodiment of the present application. Two adjacent interactive media accounts within a time window are connected through a directed edge, for example, in fig. 3, the user a sequentially accesses the interactive media account 1 and the interactive media account 2, that is, the interactive media account 1 and the interactive media account 2 have a directed edge, and the directed edge turns to the interactive media account 2 for the interactive media account 1. And distributing corresponding weights to the directed edges in the directed graph through the cooperative behaviors of all the users. The weight of the directed edge from the interactive media account 1 to the interactive media account 2 is equal to the proportion of the frequency of turning the interactive media account 1 to the interactive media account 2 to the out degree of the interactive media account 1. Then, a constructed directed graph of the interactive media accounts generates a plurality of sequences of the interactive media accounts by using a random walk idea, see fig. 7, where fig. 7 is a schematic diagram of a sequence of the interactive media accounts generated by a random walk according to an embodiment of the present application, where 701 shows a sequence of an

interactive media account

1, 702 is a sequence of an

interactive media account

2, and 703 is a sequence of an interactive media account 3. An embedded representation of each interactive media account associated with the user may be derived based on the sequence of each interactive media account, with this embedded representation being a user graph feature.

In the embodiment of the application, the graph characteristics combined with the time sequence information are obtained in a mode of constructing the directed graph, and the problem of data sparseness is well solved.

In some embodiments, after extracting media account features from the attribute data of the candidate media accounts, the following processes may also be performed to obtain dense media account features: determining the media account number collaborative filtering characteristics and the media account number image characteristics of the candidate media account numbers, connecting the media account number collaborative filtering characteristics, the media account number image characteristics and the extracted media account number characteristics, and taking the new media account number characteristics obtained by connection as the media account number characteristics for rectification processing; the media account collaborative filtering characteristics are determined by decomposing a matrix comprising the behavior data, and the media account graph characteristics are determined by mapping based on a directed graph comprising the behavior data.

It should be noted that the media account collaborative filtering feature is determined by decomposing the interaction matrix in the foregoing; the media map features are determined in the same manner as the user map features.

Next, an exemplary application of the embodiment of the present application in a practical application scenario will be described.

Referring to fig. 9, fig. 9 is a schematic view of an application scenario of the artificial intelligence based information recommendation method according to the embodiment of the present application. A specific implementation scenario of the artificial intelligence based information recommendation method provided by the embodiment of the present application will be described below with reference to fig. 9.

In the embodiment of the present application, taking an information flow product as an example, an information recommendation system maps clicking or watching behaviors of a user on different types of information such as articles, videos, and the like to interaction behaviors of the user on media numbers (i.e., media accounts publishing the articles, videos, and the like, which are hereinafter referred to as CPs for short), so that the media numbers are recalled in combination with interests of the user. That is, the information recommendation system maps the user behavior between the user and the article and the user behavior between the user and the video in the information flow product to the user and the CP uniformly, for example, if the user u clicks the article d, and d is published by the media number c, then u exists the interaction behavior with c.

In some embodiments, recalling of media numbers by the information recommendation system may be accomplished through a two-tower model. Referring to fig. 9, the two-tower model encodes the user-side feature and the CP-side feature respectively, and fits the similarity between the encoding results of the user-side feature and the CP-side feature through a scoring function, where the scoring function may be a cosine function. After the double-tower model training is finished, performing forward calculation on all the CPs in an off-line manner to obtain and store CP codes; in the online calculation, forward calculation is carried out on the user to obtain the code of the user, then the most similar Top N CPs are searched from the CPs stored offline, and the Top N CPs are used as recalled media numbers.

The structure of the double-tower model comprises an embedding layer, a coding layer and a prediction layer, wherein the embedding layer is firstly used for embedding layers and splicing embedding of various dimensions, for example, embedded representation of a user-emb, namely user emboding, is shown in fig. 9, embedded representation of a CP-emb, namely media number feature CP embedding, embedded representation of a tag feature tag embedding of a tag-emb, and embedded representation of a chanl-emb, namely media number feature chan new embedding are shown in fig. 9, and the embedded representations of various dimensions are spliced to obtain the user feature, the media number feature and the media account number feature; and then an encoding encode layer, the characteristics of the embedding layer are captured by using a full-connection network, the encoding layer comprises a linear rectifying layer and a normalization layer, the linear rectifying layer uses Re LU (linear rectifying function) as an activation function to provide nonlinear change after full connection, gradient descent and back propagation are facilitated, and the normalization layer performs normalization processing on the linear rectifying result. It should be noted that in some embodiments, a two-tower network may be constructed only by using embeddings of the user identification feature user-id and the media number identification feature C P-id, but since many users or media numbers have fewer interaction behaviors and cannot accurately learn the corresponding embeddings, the information recommendation system adds various side information to enhance the process of the embeddings. The specific characteristics include: 1) the CP side characteristic: each CP is represented by embedding with a fixed dimension and can be obtained by table lookup in the training or predicting process; tag-id, namely the label characteristic in the CP image, where one CP usually includes a plurality of labels, and when there are a plurality of label characteristics of the interactive media number of the user, the average of the plurality of label characteristics is generally taken as the label characteristic of the interactive media number; chanl-id, i.e. channel characteristics in the CP representation; fans (not shown in the figure), namely the number characteristic of the CP vermicelli, can segment the number of vermicelli as the number of vermicelli is continuous, each magnitude is divided into one segment, and the continuous vermicelli number characteristic is converted into the discrete vermicelli number characteristic. 2) user side characteristics: the user-id characteristic, namely the identification characteristic of the user, can be obtained by looking up a table as with the identification characteristic of C P; the CP-id is the interactive media number characteristic of the user, the CP which has the latest interactive behavior (clicking or watching) with the user is taken, and when the interactive media number characteristic of the user is multiple, the average is taken as the interactive media number characteristic of the user; tag-id, namely the tag characteristics of the interactive media number, counts the tags of the CP of which the user has recently performed interactive behaviors, and takes the average as the tag characteristics of the interactive media number, wherein the average can be used for carrying out average processing on the tags of the interactive media numbers with the top-ranked times of the interactive behaviors in a time period; chanl-id, namely the channel characteristics of the interactive media number, counting the channel chann el, and averaging to be used as the channel characteristics of the interactive media number, wherein the averaging can be used for averaging the channels of the interactive media number with the highest ranking of the times of interactive behaviors in a time period; age (not shown in the figure): an age characteristic of the user; location (not shown): the user region characteristics, here, age and location, can be obtained by One-Hot coding respectively.

The representation of the media number mainly includes tags, channels, and the like, and is extracted from a media number publication or a video. For example, a general article has some static features, such as channels and labels, which are called an image of the article. The information recommendation system counts the static characteristics of the article corresponding to the CP, and the portrait information of the CP can be obtained. However, the article representation is different from the media number representation, and for example, the channel characteristics are taken as an example, one article generally has only one channel, but the text type of the media number is not fixed, so that there may be a plurality of channels, and after statistics is usually performed, a plurality of heads are taken as channels in the media number representation.

In some embodiments, in addition to the embedding of the user and the CP by using a random initialization method before training, the processing of the embedding layer in the information recommendation system introduces more global embedding information to make up for the defect that the user with sparse behavior or the CP has insufficient learning. The method mainly comprises two types of cooperative filtering characteristics cf-e mbedding and graph characteristics graph-embedding. Wherein cf-embedding: embedding of CP behaving similarly should be similar, as should embedding of user behaving similarly. In the related art, the user-based nearest neighbor recommendation UCF and the article-based recommendation ICF respectively construct a user vector space and a media number vector space, and the recommendation can be performed when the similarity is found in any one vector space. In the embodiment of the application, through matrix decomposition, a high-dimensional matrix can be mapped to be a product of two low-dimensional matrices, that is, dense user and CP vectors, that is, user collaborative filtering characteristics and media number collaborative filtering characteristics, are obtained, the problem of data sparseness is well solved, the user and the CP are in the same vector space, and the prediction precision is good. graph-embedding: reflecting the order information of the user's interactive media numbers (which may more accurately affect the user's preferences), the information recommendation system takes the user's behavior within a set time window as a user behavior sequence since the user's interests may drift over time. Deriving a sequence of multiple interactive media numbers as above, the em bellding representation of each CP, i.e. graph features, is learned based on the sequence of each interactive media account. In the embodiment of the application, the cf-imbedding and the graph-imbedding learned in the two modes are used in a splicing mode, and do not participate in updating in the process of double-tower model training.

In some embodiments, different embedding and double-tower models are trained in an offline training process, when the double-tower models are trained, forward calculation is performed on all CPs to obtain a media number feature vector, a vector database faiss index is constructed based on the media number feature vector, and the media number feature vector and a corresponding media number are stored offline. In the online calculation process, forward calculation is carried out on the user in real time to obtain a user characteristic vector, and nearest top N (N before ranking) CPs are obtained by means of faiss retrieval. The information recommendation system can directly recommend the obtained CP to the user, and also can recommend articles or videos corresponding to the CP to the user.

In some embodiments, a CP may publish many articles or videos, for example, when the information recommendation system recalls many CPs, many candidate articles are available, and the most suitable article is selected through the scoring model. Features of the scoring model may use article features, user-article intersection features (e.g., media number features of published articles). As an example, specific features used may include: 1) the number of CP vermicelli and the proportion of high-quality articles; 2) issuing time, article quality score and article newness; 3) the intersection of the article portrait and the user portrait is compared.

Here, the article image includes a tag feature, a channel feature, and the like of a media number of a published article, and the user image includes user behavior data and user attribute data, for example, a tag feature, a channel feature, and the like of an interactive media account in the behavior data. The weights of the features can be used for constructing training data to train and learn, and can also be manually specified according to actual services, for example, the information recommendation system can improve the weights of three features, namely, the recall high-quality article, the CP high-quality article proportion and the article quality, so that the recalled article is strongly related to the interest of the user.

In some embodiments, the information recommendation system can obtain a ranked article list through recall of the CP and scoring of the articles, and can also add diversity processing to adjust the ranking, so as to ensure diversity of recall results, and solve the problem of aggregation effect on the heads of the articles caused by the scoring model, that is, the recalled articles all come from one or several media numbers, so that many media numbers have no exposure opportunity all the time. The diversity ordering process specifically comprises the following steps: 1) diversity of CP: ensuring that the article corresponding to each media number does not exceed a certain threshold value, wherein the threshold value is determined according to the actual service recall number; 2) diversity of channels: ensuring that the recalled articles are dispersed in different channels, wherein the article quantity of each channel can be obtained by multiplying the total recalled space quantity by the ratio of each channel in the user portrait; 3) quality control: to ensure exposure of premium media numbers, a ranking module in the information recommendation system limits the number of non-premium media numbers.

In the embodiment of the application, the behaviors of the user on articles and videos are mapped to the CP, the interests of various types of information are comprehensively utilized, and the user interests are more fully mined; the method can cover various application scenes, and can support the recommendation of media numbers, articles and videos, and the simultaneous recommendation of articles and videos; by utilizing various side information and global feature information, learning additional embedding representation of the user and the media number can effectively relieve the problems of sparse recommended scene behaviors, cold start and the like, and the recommendation accuracy is enhanced.

Continuing with the exemplary structure of the artificial intelligence based information recommendation device 255 implemented as software modules provided in the embodiments of the present application, in some embodiments, as shown in fig. 2A, the software modules stored in the artificial intelligence based information recommendation device 255 of the memory 250 may include:

an extracting module 2551, configured to extract user behavior features from behavior data of a plurality of interactive media accounts corresponding to a user; the encoding module 2552 is configured to perform encoding processing based on the user behavior features to obtain user feature vectors; a recall module 2553, configured to determine a plurality of recalled media accounts that satisfy a similar condition as the user feature vector; a recommending module 2554, configured to generate information to be recommended based on the plurality of recalled media accounts, and execute a recommending operation corresponding to the user based on the information to be recommended.

In some embodiments, the extracting module 2551 is further configured to map the behavior data of the user for a plurality of information to the behavior data of the user for an interactive media account publishing the information; wherein the user's behavior data for a plurality of information characterizes at least one of the following behaviors: and the user pays attention to the behavior of the information published by the interactive media account, and the user subscribes to the behavior of the information published by the interactive media account.

In some embodiments, the encoding module 2552 is further configured to determine a user collaborative filtering feature and a user graph feature, sequentially connect the user collaborative filtering feature, the user graph feature and the user behavior feature obtained through extraction, and use a new user behavior feature obtained through connection as the user behavior feature for performing the encoding processing; wherein the user collaborative filtering feature is determined by decomposing a matrix including the behavior data, and the user graph feature is determined based on mapping a directed graph including the behavior data.

In some embodiments, the encoding module 2552 is further configured to extract user attribute features from the attribute data of the user; rectifying the user behavior characteristics and the user attribute characteristics, or rectifying the user behavior characteristics, normalizing an obtained rectification result, and taking an obtained normalization processing result as a user characteristic vector obtained by coding; wherein the user attribute characteristics include at least one of: the age characteristic of the user and the regional characteristic of the user; the user behavior characteristics include at least one of: the interactive media account characteristics of the user, the tag characteristics of the interactive media account, and the channel characteristics of the interactive media account.

In some embodiments, the recall module 2553 is further configured to obtain media account feature vectors of multiple candidate media accounts, determine cosine distances between the media account feature vectors of the multiple candidate media accounts and the user feature vectors, and use the cosine distances as similarities; and determining a plurality of candidate media accounts with similarity exceeding a similarity threshold value with the user feature vector as a plurality of recalled media accounts meeting a similarity condition.

In some embodiments, the recall module 2553 is further configured to, for each candidate media account of the plurality of candidate media accounts, perform the following: extracting media account characteristics from the attribute data of the candidate media accounts; rectifying the media account features, normalizing the obtained rectifying result, and taking the obtained normalizing result as a media account feature vector; wherein the media account characteristics include at least one of: the system comprises the tag characteristics of the media account, the channel characteristics of the media account and the fan number characteristics of the media account.

In some embodiments, the recall module 2553 is further configured to determine a media account collaborative filtering feature and a media account map feature of the candidate media account, connect the media account collaborative filtering feature, the media account map feature and the extracted media account feature, and use a new media account feature obtained by the connection as the media account feature for performing the rectification processing; the media account collaborative filtering feature is determined by decomposing a matrix including the behavior data, and the media account graph feature is determined by mapping based on a directed graph including the behavior data.

In some embodiments, the recommending module 2554 is further configured to generate the information to be recommended by at least one of: generating information to be recommended for recommending the media account number based on the plurality of recalled media account numbers; generating information to be recommended for recommending the published information of the media accounts based on the published information of the recalled media accounts; predicting to obtain a score of the information to be recommended based on the information feature vector of the information to be recommended, the user feature vector of the user and the cross feature vector of the user and the information to be recommended, wherein the score represents the similarity between the information to be recommended and the user; based on scores of a plurality of candidate information to be recommended, sorting the plurality of candidate information to be recommended in a descending order; and performing diversity ranking on the previously ranked candidate information to be recommended, and executing recommendation operation corresponding to the user based on a diversity ranking result.

In some embodiments, when the information to be recommended is published information of the plurality of recalled media accounts, the recommending module 2554 is further configured to delete, when the number of information from the same media account in the plurality of candidate information to be recommended is greater than a first threshold, the information from the same media account one by one in an ascending order of scores until the number of information from the same media account does not exceed the first threshold; when the number of information from the same channel in the same media account in the candidate information to be recommended is larger than a second threshold value, deleting the information from the same media account one by one according to the ascending order of scores until the number of the information from the same media account does not exceed the second threshold value; and when the number of the non-high-quality information in the candidate information to be recommended is larger than a third threshold value, deleting the non-high-quality information one by one according to the ascending order of scores until the number of the non-high-quality information does not exceed the third threshold value.

Embodiments of the present application 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 executes the computer instructions, so that the computer device executes the artificial intelligence based information recommendation method described in the embodiment of the present application.

Embodiments of the present application provide a computer-readable storage medium storing executable instructions, which when executed by a processor, cause the processor to perform an artificial intelligence based information recommendation method provided by embodiments of the present application, for example, the artificial intelligence based information recommendation method shown in fig. 3A, 3B, and 3C.

In some embodiments, the computer-readable storage medium may be memory such as FRAM, ROM, PROM, EP ROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or may be various devices including one or any combination of the above memories.

In some embodiments, the executable instructions may be in the form of a program, software module, script, or code written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and it may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

By way of example, executable instructions may, but need not, correspond to files in a file system, and may be stored in a portion of a file that holds other programs or data, such as in one or more scripts in a hypertext Markup Language (H TML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code).

By way of example, executable instructions may be deployed to be executed on one computing device or on multiple computing devices at one site or distributed across multiple sites and interconnected by a communication network.

In summary, according to the embodiment of the application, when the behavior data of the user corresponding to the plurality of interactive media accounts is insufficient or exceeds the period, the attribute features are supplemented, so that the defect that the learned behavior features of the user are insufficient based on the behavior data with insufficient or exceeding data amount is overcome, and the learning precision of the neural network model is improved; encoding the characteristics obtained by splicing and fusing the user behavior characteristics and the user attribute characteristics so that the user characteristic vectors obtained by encoding can better map the interest and the demand of the user; the high-dimensional interaction matrix is subjected to matrix decomposition, dense user collaborative filtering characteristics and media account collaborative filtering characteristics can be obtained, the problem of data sparseness is well solved, and the user and the media account are in the same vector space, so that recall processing is subsequently performed on user behavior characteristics determined based on the user collaborative filtering characteristics and the media account collaborative filtering characteristics, and the recall precision is higher; by constructing the directed graph, the graph characteristics combined with the time sequence information are obtained, and the problem of data sparseness is well solved.

The above description is only an example of the present application, and is not intended to limit the scope of the present application. Any modification, equivalent replacement, and improvement made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. An information recommendation method based on artificial intelligence is characterized by comprising the following steps:

extracting user behavior characteristics from behavior data of a plurality of interactive media accounts corresponding to a user;

coding processing is carried out based on the user behavior characteristics to obtain user characteristic vectors;

determining a plurality of recalled media accounts satisfying similar conditions as the user feature vector;

and generating information to be recommended based on the plurality of recalled media accounts, and executing recommendation operation corresponding to the user based on the information to be recommended.

2. The method of claim 1, wherein prior to extracting user behavior features from behavior data of a user corresponding to a plurality of interactive media accounts, the method further comprises:

mapping the behavior data of the user aiming at a plurality of information into the behavior data of the user aiming at an interactive media account publishing the information;

wherein the user's behavior data for a plurality of information characterizes at least one of the following behaviors: and the user pays attention to the behavior of the information published by the interactive media account, and the user subscribes to the behavior of the information published by the interactive media account.

3. The method of claim 1, wherein before the encoding based on the user behavior features to obtain user feature vectors, the method further comprises:

determining a user collaborative filtering feature and a user graph feature, sequentially connecting the user collaborative filtering feature, the user graph feature and the user behavior feature obtained through extraction, and taking a new user behavior feature obtained through connection as the user behavior feature for the coding processing;

wherein the user collaborative filtering feature is determined by decomposing a matrix including the behavior data, and the user graph feature is determined based on mapping a directed graph including the behavior data.

4. The method of claim 1, wherein the encoding based on the user behavior feature to obtain a user feature vector comprises:

extracting user attribute features from the attribute data of the user;

rectifying the user behavior characteristics and the user attribute characteristics, or rectifying the user behavior characteristics, normalizing the obtained rectification result, and taking the obtained normalization processing result as a user characteristic vector obtained by coding;

wherein the user attribute characteristics include at least one of: the age characteristic of the user and the regional characteristic of the user; the user behavior characteristics include at least one of: the interactive media account characteristics of the user, the tag characteristics of the interactive media account, and the channel characteristics of the interactive media account.

5. The method of claim 1, wherein the determining a plurality of recalled media accounts that satisfy a similar condition as the user feature vector comprises:

acquiring media account characteristic vectors of a plurality of candidate media accounts, determining cosine distances between the media account characteristic vectors of the candidate media accounts and the user characteristic vectors, and taking the cosine distances as similarity;

and determining a plurality of candidate media accounts with similarity exceeding a similarity threshold value with the user feature vector as a plurality of recalled media accounts meeting a similarity condition.

6. The method of claim 5, wherein the obtaining media account feature vectors for a plurality of candidate media accounts comprises:

for each of the plurality of candidate media accounts, performing the following:

extracting media account characteristics from the attribute data of the candidate media accounts;

rectifying the media account features, normalizing the obtained rectifying result, and taking the obtained normalizing result as a media account feature vector;

wherein the media account characteristics include at least one of: the system comprises the tag characteristics of the media account, the channel characteristics of the media account and the fan number characteristics of the media account.

7. The method of claim 6, wherein after extracting media account features from the attribute data of the candidate media accounts, the method further comprises:

determining the media account number collaborative filtering characteristics and the media account number image characteristics of the candidate media account numbers, connecting the media account number collaborative filtering characteristics, the media account number image characteristics and the extracted media account number characteristics, and taking the new media account number characteristics obtained by connection as the media account number characteristics for rectification processing;

the media account collaborative filtering feature is determined by decomposing a matrix including the behavior data, and the media account graph feature is determined by mapping based on a directed graph including the behavior data.

8. The method of claim 1, wherein the generating information to be recommended based on the plurality of recalled media accounts comprises:

generating information to be recommended by at least one of the following methods:

generating information to be recommended for recommending the media account number based on the plurality of recalled media account numbers;

generating information to be recommended for recommending the information published by the media accounts based on the information published by the plurality of recalled media accounts;

the executing the recommendation operation corresponding to the user based on the information to be recommended comprises:

predicting to obtain a score of the information to be recommended based on the information feature vector of the information to be recommended, the user feature vector of the user and the cross feature vector of the user and the information to be recommended, wherein the score represents the similarity between the information to be recommended and the user;

based on scores of a plurality of candidate information to be recommended, sorting the plurality of candidate information to be recommended in a descending order;

and performing diversity ranking on the plurality of candidate information to be recommended ranked at the top, and executing recommendation operation corresponding to the user based on diversity ranking results.

9. The method according to claim 8, wherein when the information to be recommended is published information of the plurality of recalled media accounts, the diversity ranking of the top ranked candidate information to be recommended includes at least one of:

when the number of information from the same media account in the candidate information to be recommended is larger than a first threshold value, deleting the information from the same media account one by one according to the ascending order of scores until the number of the information from the same media account does not exceed the first threshold value;

when the number of information from the same channel in the same media account in the plurality of candidate information to be recommended is larger than a second threshold value, deleting the information from the same media account one by one according to the ascending order of scores until the number of information from the same media account does not exceed the second threshold value;

and when the number of the non-high-quality information in the candidate information to be recommended is larger than a third threshold value, deleting the non-high-quality information one by one according to the ascending order of scores until the number of the non-high-quality information does not exceed the third threshold value.

10. An artificial intelligence-based information recommendation device, comprising:

the extraction module is used for extracting user behavior characteristics from behavior data of a plurality of interactive media accounts corresponding to a user;

the coding module is used for carrying out coding processing based on the user behavior characteristics to obtain a user characteristic vector;

the recall module is used for determining a plurality of recalled media accounts which meet similar conditions with the user feature vector;

and the recommending module is used for generating information to be recommended based on the plurality of recalled media accounts and executing recommending operation corresponding to the user based on the information to be recommended.