CN115545738A

CN115545738A - Recommendation method and related device

Info

Publication number: CN115545738A
Application number: CN202110742728.1A
Authority: CN
Inventors: 陈冲; 张敏; 马为之; 刘奕群; 马少平; 王钊伟; 何秀强
Original assignee: Tsinghua University; Huawei Technologies Co Ltd
Current assignee: Tsinghua University; Huawei Technologies Co Ltd
Priority date: 2021-06-30
Filing date: 2021-06-30
Publication date: 2022-12-30

Abstract

The application discloses an information recommendation method which can be applied to the field of artificial intelligence and is used for generating a target user feature vector for representing the favorite of a target user based on an article with an incidence relation and an operation type and generating a target article feature vector for representing the attraction feature of the target article to the user based on the user with the incidence relation and the operation type to predict the probability of the target user for carrying out multiple operation types on the target article, wherein the probability of the operation types can more accurately depict the operation behavior of the user aiming at the article, and an information recommendation result based on the probability of the operation of the multiple operation types can be more accurate.

Description

Recommendation method and related device

Technical Field

The present application relates to the field of artificial intelligence, and in particular, to a recommendation method and related apparatus.

Background

Artificial Intelligence (AI) is a theory, method, 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. In other words, artificial intelligence is a branch of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.

The selection rate prediction means predicting the selection probability of a user for a certain item under a specific environment. For example, in recommendation systems for applications such as application stores, online advertisements, etc., the prediction of the selection rate plays a key role; the method can achieve the purposes of maximizing the income of enterprises and improving the satisfaction degree of users through selection rate prediction, and a recommendation system needs to consider the selection rate of the users to the articles and article bidding at the same time, wherein the selection rate is obtained through prediction of the recommendation system according to the historical behaviors of the users, and the article bidding represents the income of the system after the articles are selected/downloaded. For example, a function may be constructed that may be calculated based on the predicted user selection rate and the bid for the item to obtain a function value by which the recommender system sorts the items in descending order.

However, the selection rate prediction can only represent the probability of selecting an item by a user, and the item recommendation result based on the information is not accurate.

Disclosure of Invention

In a first aspect, the present application provides a recommendation method, including:

acquiring a first operation information set of a target user, wherein the first operation information set comprises attribute information of a plurality of articles, a plurality of operation types and corresponding relations between the articles and the operation types, and the corresponding relations are used for representing operation types of the target user on the articles;

the attribute information of the target user may be at least one of attributes related to user preference characteristics, sex, age, occupation, income, hobbies and education level, wherein sex may be male or female, age may be a number between 0 and 100, occupation may be a teacher, a programmer, a chef and the like, hobbies may be basketball, tennis, running and the like, education level may be primary school, junior school, high school, university and the like; the specific type of the attribute information of the target user is not limited in the application;

the article may be an entity article or a virtual article, for example, the article may be an article such as APP, audio/video, a webpage, and news information, the attribute information of the article may be at least one of an article name, a developer, an installation package size, a category, and a goodness of evaluation, where, taking the article as an application program as an example, the category of the article may be a chat category, a cool game category, an office category, and the like, and the goodness of evaluation may be a score, a comment, and the like for the article; the application does not limit the specific type of the attribute information of the article;

the operation type may be a behavior operation type of a target user for an article, and on a network platform and an application, the user often has various interaction forms (that is, various operation types) with the article, such as operation types of browsing, clicking, adding to a shopping cart, purchasing, and the like of the user in the e-commerce platform behavior shown in fig. 5. The various behaviors reflect the preference of the user, and are helpful for accurately depicting the characteristics of the user; it should be understood that the type of operation that reflects the user's preference for the item may also be referred to as a forward type of operation;

it should be understood that the data presentation state of the operation type in the embodiment of the present application may be a feature vector (also referred to as an operation type feature vector in the embodiment of the present application), where the operation type feature vector may be updated in a training process of the target recommendation model and obtained after the target recommendation model converges (or meets a data processing precision requirement), in the training process of the target recommendation model, the generalization of the operation type feature vector is continuously enhanced, and after the target recommendation model converges, the operation type feature vector may have a very strong generalization, where the so-called strong generalization means that the operation type feature vector may be suitable for the target recommendation model to perform probability prediction of operations between a new user and a new object, and the precision of the probability prediction is very high;

it should be understood that each operation type may correspond to one operation type feature vector, and when performing model inference of the target recommendation model, the operation type feature vector corresponding to each operation type may be fixed and unchanged;

it should be understood that the operation type feature vector may be used to generate a target user feature vector and a target article feature vector, or may be used as an input of a target recommendation model separately, and the target recommendation model may calculate, according to the target user feature vector, the target article feature vector, and the operation type feature vector, a probability that the target user performs an operation of an operation type corresponding to the operation type feature vector on the target article.

Performing feature extraction according to the first operation information set to determine a target user feature vector;

acquiring a second operation information set of the target article, wherein the second operation information set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the target article operated by the plurality of users;

performing feature extraction according to the second operation information set to determine a target article feature vector;

outputting recommendation information based on a target recommendation model according to the target user feature vector and the target article feature vector of the target user, wherein the recommendation information is used for representing the probability of the target user performing the operations of the multiple operation types on the target article;

and when the recommendation information meets a preset condition, determining to recommend the target article to the target user.

In one possible implementation, when information recommendation is performed on a target user, probabilities of multiple operation types performed on multiple articles (including the target article) by the target user may be calculated, and a recommendation index for the target user for each article may be determined based on the probabilities of the multiple operation types.

In one possible implementation, a maximum probability of the probabilities of the plurality of operation types of the target user on each item may be selected to characterize a recommendation index of each item to the target user;

in a possible implementation, a comprehensive value of the probability of the target user for multiple operation types of each item may be calculated to represent the recommendation index of each item for the target user, the comprehensive value may be based on a weighted summation manner, and specifically, a corresponding weight may be set for each operation type, for example, the weight of a purchase operation is greater than the weight of an operation of joining a shopping cart, and then, the recommendation index of each operation type may be obtained based on the weighted summation by combining the weight corresponding to each operation type and the probability corresponding to each operation type;

after obtaining the recommendation indexes of the respective items for the target user, the recommendation indexes may be sorted, and the M items (including the target item) with the largest recommendation indexes may be recommended to the target user.

In a possible implementation, a probability threshold may also be optionally set, and when the probability corresponding to at least one of the multiple operation types of the target user on the target item is greater than the probability threshold, the target item may be recommended to the target user.

When information recommendation is performed, recommendation information can be recommended to a user in a list page form so as to expect the user to perform behavior action.

According to the method and the device, the target user feature vector representing the preference of the target user is generated based on the article and the operation type with the association relationship, and the second feature vector representing the attraction feature of the target article to the user is generated based on the user and the operation type with the association relationship, so that the probability of the target user performing operation of multiple operation types on the target article is predicted, and the operation probability of the user aiming at the article can be accurately depicted.

In one possible implementation, the determining a feature vector of a target user according to feature extraction performed by the first operation information set includes:

determining a plurality of sub-user characteristic vectors according to the first operation information set, wherein each sub-user characteristic vector is obtained by performing characteristic extraction on attribute information and operation types of the articles with corresponding relations;

and fusing the plurality of sub-user feature vectors to obtain the target user feature vector.

In one possible implementation, the first set of operational information includes: attribute information of a first article, a first operation type and a corresponding relation between the first article and the first operation type; further, a sub-user feature vector (first sub-user feature vector) may be calculated based on the attribute information of the first item and the first operation type. In this way, a plurality of sub-user feature vectors may be obtained, wherein a part of the sub-user feature vectors may be regarded as first-order feature vectors of the target user (based on real operation information, such as attribute information of the first item, the first operation type, and a corresponding relationship between the first item and the first operation type), and a part of the sub-user feature vectors may be regarded as second-order feature vectors of the target user (based on predicted operation information, such as attribute information of the second item, the second operation type, and a corresponding relationship between the second item and the second operation type), and similarly, higher-order feature vectors of the target user may also be obtained.

In a possible implementation, each sub-user feature vector may represent a favorite feature of a target user, so that multiple sub-user feature vectors may be fused, an activation function may be used to fuse the sub-user feature vectors of the same order, one feature vector result that may be obtained by fusing the sub-user feature vectors of the same order may be obtained, multiple feature vector results may be obtained by fusing the sub-user feature vectors of multiple orders, and then multiple feature vector results may be fused (for example, the fusion may be performed based on different weights, for example, a feature vector with a small order may be used to more accurately depict a feature of the target user, and a weight during the fusion may be set to be larger), so as to obtain a target user feature vector, where the fusion may be, but is not limited to, an addition and concatenation operation (concat).

In one possible implementation, the determining a target article feature vector by feature extraction according to the second operation information set includes:

determining a plurality of sub-article feature vectors according to the second operation information set, wherein each sub-article feature vector is obtained by performing feature extraction on attribute information and operation types of users with corresponding relations; and fusing the plurality of sub-article feature vectors to obtain the target article feature vector.

In a possible implementation, feature extraction may be performed on the basis of attribute information and an operation type of a user having a correspondence relationship to obtain a sub-article feature vector, where the sub-article feature vector may represent an attraction feature of a target article for the user, and then multiple sub-article feature vectors may be obtained, and the multiple sub-article feature vectors are fused to obtain the second feature vector.

In this way, a plurality of sub-item feature vectors may be obtained, wherein a part of the sub-item feature vectors may be regarded as first-order feature vectors of the target item (based on real operation information, such as the attribute information of the second user, the fourth forward operation type, and the corresponding relationship between the second user and the fourth forward operation type), and a part of the sub-item feature vectors may be regarded as second-order feature vectors of the target item (based on predicted operation information, such as the attribute information of the third user, the fifth operation type, and the corresponding relationship between the third user and the fifth operation type), and similarly, higher-order feature vectors of the target item may also be obtained.

In a possible implementation, each sub-item feature vector may represent an attractive feature of a target item for a user, so that multiple sub-item feature vectors may be fused, an activation function may be used for fusing the sub-item feature vectors of the same order, one feature vector result that may be obtained for fusing the sub-item feature vectors of the same order, multiple feature vector results that may be obtained for fusing the sub-item feature vectors of multiple orders, and then multiple feature vector results may be fused (for example, the fusion may be performed based on different weights, for example, a feature vector with a small order may be used to more accurately characterize a target item, and the weight during the fusion may be set to be larger), so as to obtain a second feature vector, where the fusion may be, but not limited to, a sum and a concatenation operation (concat).

In one possible implementation, the outputting recommendation information based on a target recommendation model according to the target user feature vector and the target item feature vector includes: and outputting recommendation information based on a target recommendation model according to a plurality of operation type feature vectors of the plurality of operation types, the target user feature vector and the target article feature vector.

In one possible implementation, the target recommendation model may calculate a similarity between a target user feature vector of the target user and a target item feature vector, and then calculate a similarity between the target user feature vector, the target item feature vector, and the operation type feature vector.

In one possible implementation, the obtaining the first set of operational information includes: acquiring a first operation information subset and a second operation information subset; wherein the first operation information subset comprises attribute information of a first article, a first operation type and a corresponding relation between the first article and the first operation type; the first operation type is a real operation type of the target user on the first article; the second operation information subset comprises attribute information of a second article, a second operation type and a corresponding relation between the second article and the second operation type; the second operation type is a predicted operation type of the target user on the second article; the determining a plurality of sub-user feature vectors according to the first operation information set includes: determining a first sub-user feature vector according to the first operation information subset; and determining a second sub-user feature vector according to the second operation information subset.

That is, the attribute information of the first article, the first operation type, and the correspondence between the first article and the first operation type are obtained based on the actual operation record of the target user.

Because the obtained historical operation records related to the user are limited, in order to improve the accuracy of information recommendation, richer information can be found in the limited historical operation data, and then more operation information related to the user is generated, so that the data utilization rate of the historical operation records can be improved.

In one possible implementation, the fusing the plurality of sub-item feature vectors comprises: and fusing the first sub-user feature vector and the second sub-user feature vector according to the first weight of the first sub-user feature vector and the second weight of the second sub-user feature vector.

In the embodiment of the present application, the sub-user feature vectors may be fused based on different weights, and the sub-user feature vector obtained for the real operation data of the user (for example, the first sub-user feature vector) may have a larger weight since the feature of the target user may be more accurately described, and the sub-user feature vector obtained for the predicted operation data of the user (for example, the second sub-user feature vector) may have a smaller weight since the feature of the target user may not be accurately described, and in addition, the sub-user feature vector obtained for the predicted operation data of the user may have a different order for prediction (for example, the predicted operation data obtained based on similarity of preference degrees between 5 users is higher than the predicted operation data obtained based on similarity of preference degrees between 2 users), and may also have a different weight, and the higher order is smaller; in addition, the above-mentioned order can also be used as a part of a feed-forward process to adjust the proportion of the sub-user feature vectors of each order during fusion when training the target recommendation model, and is updated continuously during training, and after the target recommendation model converges, weights for different orders can be obtained, and the weights can adjust the proportion of the sub-user feature vectors of each order during fusion, so as to obtain a target user feature vector capable of accurately characterizing the user preference.

In a possible implementation, each sub-user feature vector may represent a favorite feature of a target user, so that multiple sub-user feature vectors may be fused, an activation function may be used for fusing the sub-user feature vectors of the same order, one feature vector result that may be obtained by fusing the sub-user feature vectors of the same order may be obtained for multiple sub-user feature vectors of multiple orders, and then multiple feature vector results may be fused to obtain the target user feature vector, where the fusion may be, but is not limited to, a sum and splice operation (concat).

In one possible implementation, the obtaining the second subset of operational information includes: acquiring the second operation type; the obtaining of the second operation type specifically includes: acquiring a third operation type of a second article by a first user, wherein the first user is a user whose article preference characteristics of the target user meet preset conditions; obtaining the second operation type based on the third operation type of the first user for the second article.

In a possible implementation, the preset condition may include that the first user and the target user are both users who have operations on the first item.

When different users (for example, the first user and the target user) operate on the same article, it may be characterized that the two users have a greater possibility of having the same or similar preference, and therefore, the target user-related operation information may be generated based on the operation information of the first user on the other article, and this portion of the generated new target user-related operation information may, although not in the historical operation records, generally characterize possible operation associations between the target user and other articles except the first article.

In a possible implementation, the target user and the first user may both perform an operation for the first item, for example, the target user performs a browsing operation for the first item, and the first user performs a purchasing operation for the first item, and therefore the target user and the first user may be considered as users having similar preferences, so that more operation information related to the target user may be generated based on the historical operation information of the first user for other items, and in particular, based on the historical operation record that the operation information of the first user for the second item is recorded, the operation information of the target user for the second item may be generated.

In order to ensure the accuracy of the generated operation information, the operation type of the second user for the second article in the operation information of the target user for the second article may be set to be consistent with the operation type of the first user for the second article in the history operation record (that is, the second operation type and the third operation type are the same).

In the embodiment of the application, based on the users (the target user and the first user) who operate the same article (the first article), new operation information related to the target user is generated through other operation information of the first user, richer information is found in limited historical operation data, and the data utilization rate of the historical operation record is improved.

In addition, in a possible implementation, the operation information of the target user for the third article may be generated based on the historical operation record further including operation information of the second user for the second article and operation information of the third user for the third article, that is, the second user and the first user may both perform operations for the second article, for example, the first user performs a browsing operation for the second article, while the second user performs a purchasing operation for the second article, the first user and the second user may be considered as users having similar preferences, and since the first user and the second user are users having similar preferences, the target user and the second user may also be considered as users having similar preferences, so that more operation information related to the target user may be generated based on the historical operation information of the second user for the third article, and particularly, based on the historical operation record further including the operation information of the second user for the third article in the historical operation record, the operation information of the target user for the third article may be generated.

In order to ensure the accuracy of the generated operation information, the operation type of the target user for the third article in the operation information of the target user for the third article may be set to be consistent with the operation type of the second user for the third article in the historical operation record.

In the embodiment of the application, new operation information related to the target user is generated through other operation information of the second user based on the user who operates the same article (the second article), richer information is found in limited historical operation data, and the data utilization rate of the historical operation record is improved.

In a possible implementation, the preset condition may include that a degree of difference between user attributes of the first user and the target user is less than a threshold.

In the above manner, the similarity of the preference between the users is reflected by whether the users operate on the same article, and in a possible implementation, the similarity of the preference between the users can also be determined directly through the user attribute difference between the users, wherein the user attribute is an attribute capable of reflecting the preference of the users.

In one possible implementation, the user attributes may be: the gender, the age, the occupation, the income, the hobbies, and the education level, wherein the gender may be a male or a female, the age may be a number between 0 and 100, the occupation may be a teacher, a programmer, a chef, and the like, the hobbies may be basketball, tennis, running, and the like, and the education level may be primary school, junior school, high school, university, and the like.

In this embodiment of the application, users with similar user attributes may consider to have similar preferences, for example, users with similar user attributes may be users with small age difference, users with the same gender, users with the same or similar profession (profession is similar and may be understood as being in the same industry), users with the same or similar hobbies (e.g., users who all enjoy tennis), and users with the same or similar education degree (e.g., users who all enjoy college homework graduation).

In a possible implementation, the degree of difference between the user attributes of the target user and the first user is less than a threshold, that is, the target user and the first user are users with similar preferences, so that more operation information related to the target user can be generated based on the historical operation information of the first user on other articles, and particularly, based on the operation information of the first user on the second article which is also recorded in the historical operation record, the operation information of the target user on the second article can be generated.

In order to ensure the accuracy of the generated operation information, the operation type of the target user for the second article in the operation information of the target user for the second article may be set to be consistent with the operation type of the first user for the second article in the historical operation record (that is, the second operation type is the same as the third operation type).

In the embodiment of the application, based on users (the target user and the first user) with similar user attributes, new operation information related to the target user is generated through other operation information of the first user, richer information is found in limited historical operation data, and the data utilization rate of the historical operation record is improved.

In one possible implementation, the preset condition may include that the first user and the target user are users who operate on an item whose attribute difference is smaller than a threshold value.

In one possible implementation, the item attributes may be: the system comprises at least one of an article name, a developer, an installation package size, a class and a goodness degree, wherein the article is taken as an application program as an example, the class of the article can be a chat class, a cool game, an office class and the like, and the goodness degree can be a score, a comment and the like aiming at the article.

In this embodiment of the application, a user who operates on an article having a similar article attribute may consider that the article having the similar article attribute has a similar preference, for example, the article having the similar article attribute may be an article having a same or similar article name, an article having a same or similar article type, and an article having a same or similar item goodness, and the article attribute of the article may be represented by a weight value, where the more similar the weight value is, the more similar the article attribute between the articles is represented by a feature vector, and the closer the distance between the feature vectors is, the more similar the article attribute between the articles is.

In a possible implementation, the first user operates the first article, the target user operates the fourth article, and the target user and the first user may be considered as users having similar preferences based on that the degree of difference between the article attributes of the first article and the fourth article is less than the threshold, so that more operation information related to the target user may be generated based on historical operation information of the first user on other articles, and in particular, operation information of the target user on the second article may be generated based on operation information of the first user on the second article also recorded in the historical operation record.

In the embodiment of the application, based on users (the target user and the first user) operating on articles with similar article attributes, new operation information related to the target user is generated through other operation information of the first user, richer information is found in limited historical operation data, and the data utilization rate of historical operation records is improved.

In a possible implementation, the second operation type is obtained based on the third operation type of the first user for the second article, specifically, the second operation type is obtained according to the third operation type.

In one possible implementation, the operation type is a forward operation type.

In one possible implementation, the obtaining the second set of operational information includes:

acquiring a third operation information subset and a fourth operation information subset; wherein the content of the first and second substances,

the third operation information subset comprises attribute information of a second user, a fourth forward operation type and a corresponding relation between the second user and the fourth forward operation type; the fourth forward operation type is a real operation type of the second user on the target object.

The fourth operation information subset comprises attribute information of a third user, a fifth operation type and a corresponding relation between the third user and the fifth operation type; the fifth operation type is a predicted operation type of the third user on the target item;

the determining a plurality of sub-item feature vectors according to the second operation information set comprises:

determining a first sub-item feature vector according to the third operation information subset;

and determining a second sub-article feature vector according to the fourth operation information subset.

In one possible implementation, the fusing the plurality of sub-item feature vectors includes:

and fusing the first sub-item feature vector and the second sub-item feature vector according to the third weight of the first sub-item feature vector and the fourth weight of the second sub-user feature vector.

In one possible implementation, the obtaining the fourth subset of operation information includes:

acquiring the fifth operation type;

the obtaining the fifth operation type specifically includes:

acquiring a sixth operation type of a fourth user on the target object, wherein the object preference characteristics of the fourth user and the third user meet preset conditions;

and acquiring the fifth operation type based on a sixth operation type of the target object by the fourth user.

In one possible implementation, the preset condition includes at least one of:

the fourth user and the third user are both users who operate the first article;

the difference degree of the user attributes of the fourth user and the third user is smaller than a threshold value; and

the fourth user and the third user are users who have operations on items whose attribute differences are smaller than a threshold value.

In a possible implementation, the fifth operation type is obtained based on a sixth operation type of the fourth user on the target item, specifically, the fifth operation type is obtained according to the sixth operation type.

In a second aspect, the present application provides a recommendation model training method, including:

acquiring a first operation information sample set of a target sample user, wherein the first operation information sample set comprises attribute information of a plurality of articles, a plurality of operation types and corresponding relations between the articles and the operation types, and the corresponding relations are used for representing operation types of the target sample user on the articles;

performing feature extraction according to the first operation information sample set to determine a target sample user feature vector;

acquiring a second operation information sample set of the target sample article, wherein the second operation information sample set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the plurality of users for operating the target article;

performing feature extraction according to the second operation information sample set to determine a feature vector of the target sample article;

obtaining a sample label according to the actual operation type of the target sample user on the target object;

and performing model training by taking the target sample user characteristic vector and the target sample article characteristic vector as input and the sample label as output to obtain a target recommendation model.

In one possible implementation, the determining a target sample user feature vector by performing feature extraction according to the first operation information sample set includes:

determining a plurality of sub-sample user feature vectors according to the first operation information sample set, wherein each sub-sample user feature vector is obtained by performing feature extraction on attribute information and operation types of the articles with corresponding relations;

and fusing the user characteristic vectors of the plurality of user sub-samples to obtain the user characteristic vector of the target sample.

In one possible implementation, the determining a target sample item feature vector by performing feature extraction according to the second operation information sample set includes:

determining a plurality of subsample article feature vectors according to the second operation information sample set, wherein each second subsample article feature vector is obtained by feature extraction of attribute information and operation types of users with corresponding relations;

and fusing the plurality of sub-sample article feature vectors to obtain the target sample article feature vector.

In one possible implementation, the obtaining the first sample set of operational information includes:

acquiring a first operation information sample subset and a second operation information sample subset; wherein, the first and the second end of the pipe are connected with each other,

the first operation information sample subset comprises attribute information of a first article, a first operation type and a corresponding relation between the first article and the first operation type; the first operation type is a real operation type of the target user on the first article;

the second operation information sample subset comprises attribute information of a second article, a second operation type and a corresponding relation between the second article and the second operation type; the second operation type is a predicted operation type of the target user on the second article;

the determining a plurality of sub-sample user feature vectors from the first set of operation information samples comprises:

determining a first sub-sample user feature vector according to the first operation information sample subset;

and determining a second sub-sample user feature vector according to the second operation information sample subset.

In one possible implementation, the fusing the plurality of user subsample user feature vectors comprises:

and fusing the first sub-sample user feature vector and the second sub-sample user feature vector according to a first weight of the first sub-sample user feature vector and a second weight of the second sub-sample user feature vector.

In one possible implementation, the obtaining the second subset of operation information samples includes:

acquiring the second operation type;

the obtaining of the second operation type specifically includes:

acquiring a third operation type of a second article by a first user, wherein the article preference characteristics of the first user are users meeting preset conditions;

obtaining the second operation type based on the third operation type of the first user for the second article.

In one possible implementation, the preset condition includes at least one of:

the first user and the target user are both users who have operations on the first article;

the difference degree of the user attributes of the first user and the target user is smaller than a threshold value; and

the first user and the target user are users who have operations on the items with the difference of the item attributes smaller than the threshold value.

In one possible implementation, the operation type is a forward operation type.

In one possible implementation, the obtaining the second operation information sample set includes:

acquiring a third operation information sample subset and a fourth operation information sample subset; wherein the content of the first and second substances,

the third operation information sample subset comprises attribute information of a second user, a fourth forward operation type and a corresponding relation between the second user and the fourth forward operation type; the fourth forward operation type is a real operation type of the second user on the target object;

the fourth operation information sample subset comprises attribute information of a third user, a fifth operation type and a corresponding relation between the third user and the fifth operation type; the fifth operation type is a predicted operation type of the third user on the target item;

the determining a plurality of subsample item feature vectors according to the second sample set of operational information comprises:

determining a first subsample item feature vector according to the third operation information sample subset;

determining a second subsample item feature vector according to the fourth subset of operation information samples.

In one possible implementation, the fusing the plurality of sub-sample item feature vectors includes:

fusing the first sub-sample item feature vector and the second sub-sample item feature vector according to a third weight of the first sub-sample item feature vector and a fourth weight of the second sub-sample item feature vector.

acquiring the fifth operation type;

the obtaining the fifth operation type specifically includes:

and acquiring the fifth operation type based on a sixth operation type of the target article by the fourth user.

In one possible implementation, the preset condition includes at least one of:

In a third aspect, the present application provides a training sample construction method, including:

acquiring a first operation type of a first user on a first article, wherein the article preference characteristics of the first user and a target user meet a preset condition;

generating a second operation type of the target user on the first article based on the first operation type of the first user on the first article; the second operation type is a predicted operation behavior of the target user for the first item;

and constructing a training sample according to the attribute information of the target user, the attribute information of the first article and the second operation type.

According to the method and the device, richer information can be found in limited historical operation data, and further more operation information related to the user is generated, so that the data utilization rate of historical operation records can be improved, more training samples are constructed, and the behavior of the user can be predicted more accurately based on the recommendation model trained by the training samples.

In one possible implementation, the preset condition includes at least one of:

the first user and the target user are both users who operate the second article;

the first user and the target user are users who have operations on items whose attribute differences are smaller than a threshold value.

In one possible implementation, the user attribute includes at least one of:

gender, age, occupation, income, hobbies, education level.

In one possible implementation, the item attribute includes at least one of:

item name, developer, installation package size, category and goodness.

In one possible implementation, the first operation type is the same as the second operation type.

In one possible implementation, the first operation type and the second operation type include at least one of:

browsing, joining shopping cart, and purchasing.

In one possible implementation, the generating, based on the first type of operation of the first user with respect to the first item, a second type of operation of the target user with respect to the first item includes: obtaining a second operation type based on the first operation type of the first user for the first article;

the obtaining of the second operation type is specifically to obtain the second operation type according to the first operation type.

In a possible implementation, the obtained operation information may be used to train a target recommendation model, specifically, the attribute information of the target user and the attribute information of the first item may be obtained; determining a target user characteristic vector according to the attribute information of the target user; determining a target article feature vector according to the attribute information of the first article; acquiring a third feature vector of the second operation type; and training a target recommendation model according to the target user feature vector, the target article feature vector and the third feature vector to obtain a trained target recommendation model. Determining a target user feature vector according to the attribute information of the target user; determining a target article feature vector according to the attribute information of the first article; the obtaining of the third feature vector of the second operation type may refer to the description in the foregoing embodiments, and the details of the similarity are not repeated here.

In a possible implementation, a prediction probability may be output through a target recommendation model according to the target user feature vector, the target item feature vector, and the third feature vector, where the prediction probability is used to represent a probability that the target user performs the second operation type on the first item, and according to the probability, a loss is determined, and the target recommendation model is updated according to the loss.

In a fourth aspect, the present application provides an information recommendation apparatus, comprising:

the system comprises an acquisition module, a processing module and a display module, wherein the acquisition module is used for acquiring a first operation information set of a target user, the first operation information set comprises attribute information of a plurality of articles, a plurality of operation types and corresponding relations between the articles and the operation types, and the corresponding relations are used for representing operation types of the target user for operating the articles; acquiring a second operation information set of the target article, wherein the second operation information set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the operation performed on the target article by the plurality of users;

the characteristic vector generation module is used for extracting characteristics according to the first operation information set to determine a target user characteristic vector; performing feature extraction according to the second operation information set to determine a feature vector of the target article;

the information recommendation module is used for outputting recommendation information based on a target recommendation model according to the target user feature vector and the target article feature vector, wherein the recommendation information is used for representing the probability of the target user performing the operation of the plurality of operation types on the target article; and when the recommendation information meets a preset condition, determining to recommend the target article to the target user.

In a possible implementation, the feature vector generating module is specifically configured to:

and fusing the sub-user feature vectors to obtain the target user feature vector.

determining a plurality of sub-article feature vectors according to the second operation information set, wherein each sub-article feature vector is obtained by performing feature extraction on attribute information and operation types of users with corresponding relations;

and fusing the plurality of sub-article feature vectors to obtain the target article feature vector.

In a possible implementation, the information recommendation module is specifically configured to:

and outputting recommendation information based on a target recommendation model according to a plurality of operation type feature vectors of the plurality of operation types, the target user feature vector and the target article feature vector.

In a possible implementation, the obtaining module is specifically configured to:

acquiring a first operation information subset and a second operation information subset; wherein the content of the first and second substances,

the first operation information subset comprises attribute information of a first article, a first operation type and a corresponding relation between the first article and the first operation type; the first operation type is a real operation type of the target user on the first article;

the second operation information subset comprises attribute information of a second article, a second operation type and a corresponding relation between the second article and the second operation type; the second operation type is a predicted operation type of the target user on the second article;

the feature vector generation module is specifically configured to:

determining a first sub-user feature vector according to the first operation information subset;

and determining a second sub-user characteristic vector according to the second operation information subset.

In a possible implementation, the feature vector generation module is specifically configured to:

and fusing the first sub-user feature vector and the second sub-user feature vector according to the first weight of the first sub-user feature vector and the second weight of the second sub-user feature vector.

In one possible implementation, the obtaining module is further configured to:

acquiring the second operation type;

the obtaining of the second operation type specifically includes:

acquiring a third operation type of a second article by a first user, wherein the first user is a user whose article preference characteristics of the target user meet preset conditions;

In one possible implementation, the preset condition includes at least one of:

In one possible implementation, the operation type is a forward operation type.

the feature vector generation module is specifically configured to:

acquiring the fifth operation type;

the obtaining the fifth operation type specifically includes:

In one possible implementation, the preset condition includes at least one of:

the fourth user and the third user are both users who have operations on the first article;

The application provides an information recommendation device, the device includes: an obtaining module, configured to obtain a first operation information set of a target user, where the first operation information set includes attribute information of a plurality of articles, a plurality of operation types, and correspondence between the plurality of articles and the plurality of operation types, and the correspondence is used to represent operation types of operations performed on the plurality of articles by the target user; acquiring a second operation information set of the target article, wherein the second operation information set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the operation performed on the target article by the plurality of users; the characteristic vector generation module is used for extracting characteristics according to the first operation information set to determine a target user characteristic vector; performing feature extraction according to the second operation information set to determine a feature vector of the target article; the information recommendation module is used for outputting recommendation information based on a target recommendation model according to the target user feature vector and the target article feature vector, wherein the recommendation information is used for representing the probability of the target user performing the operation of the plurality of operation types on the target article; and when the recommendation information meets a preset condition, determining to recommend the target article to the target user. According to the method and the device, the target user feature vector representing the preference of the target user is generated based on the article and the operation type with the association relationship, and the second feature vector representing the attraction feature of the target article to the user is generated based on the user and the operation type with the association relationship, so that the probability of the target user performing operation of multiple operation types on the target article is predicted, and the operation probability of the user aiming at the article can be accurately depicted.

In a fifth aspect, the present application provides a recommendation model training apparatus, including:

the system comprises an acquisition module, a processing module and a display module, wherein the acquisition module is used for acquiring a first operation information sample set of a target sample user, the first operation information sample set comprises attribute information of a plurality of articles, a plurality of operation types and corresponding relations between the articles and the operation types, and the corresponding relations are used for representing operation types of the target sample user on the articles; acquiring a second operation information sample set of the target sample article, wherein the second operation information sample set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the plurality of users for operating the target article;

the characteristic vector generation module is used for extracting characteristics according to the first operation information sample set to determine a target sample user characteristic vector; performing feature extraction according to the second operation information sample set to determine a feature vector of the target sample article;

the obtaining module is further used for obtaining a sample label according to the actual operation type of the target sample user on the target article;

and the model training module is used for performing model training by taking the target sample user characteristic vector and the target sample article characteristic vector as input and the sample label as output to obtain a target recommendation model.

determining a plurality of sub-sample user characteristic vectors according to the first operation information sample set, wherein each sub-sample user characteristic vector is obtained by performing characteristic extraction on attribute information and operation types of the articles with corresponding relations;

the feature vector generation module is specifically configured to:

acquiring the second operation type;

the obtaining of the second operation type specifically includes:

In one possible implementation, the preset condition includes at least one of:

In one possible implementation, the operation type is a forward operation type.

acquiring a third operation information sample subset and a fourth operation information sample subset; wherein, the first and the second end of the pipe are connected with each other,

the feature vector generation module is specifically configured to:

acquiring the fifth operation type;

the obtaining the fifth operation type specifically includes:

In one possible implementation, the preset condition includes at least one of:

In a possible implementation, the fifth operation type is obtained based on a sixth operation type of the target article by the fourth user, specifically, the fifth operation type is obtained according to the sixth operation type.

In a sixth aspect, the present application provides a training sample construction apparatus, the apparatus comprising:

the system comprises an acquisition module, a display module and a display module, wherein the acquisition module is used for acquiring a first operation type of a first user on a first article, and the article preference characteristics of the first user and a target user meet preset conditions;

an operation information generating module, configured to generate a second operation type of the first item by the target user based on the first operation type of the first user for the first item; the second operation type is a predicted operation behavior of the target user for the first item;

and the sample construction module is used for constructing a training sample according to the attribute information of the target user, the attribute information of the first article and the second operation type.

In one possible implementation, the preset condition includes at least one of:

In one possible implementation, the user attribute includes at least one of:

gender, age, occupation, income, hobbies, education level.

In one possible implementation, the item attribute includes at least one of:

item name, developer, installation package size, category and goodness.

In a possible implementation, the operation information generating module is specifically configured to:

obtaining a second operation type based on the first operation type of the first user for the first article;

browsing operation, joining shopping cart operation and purchasing operation.

The embodiment of the application provides a user operation behavior prediction device, which comprises: the system comprises an acquisition module, a display module and a display module, wherein the acquisition module is used for acquiring a first operation type of a first user on a first article, and the article preference characteristics of the first user and a target user meet preset conditions; an operation information generating module, configured to generate a second operation type of the first item by the target user based on the first operation type of the first user for the first item; the second operation type is a predicted operation behavior of the target user for the first item. According to the method and the device, richer information can be found in limited historical operation data, so that more operation information related to the user can be generated, the data utilization rate of historical operation records can be improved, and the behavior of the user can be predicted more accurately based on the recommendation model trained based on the information.

In a seventh aspect, an embodiment of the present application provides a recommendation device, which may include a memory, a processor, and a bus system, where the memory is used to store a program, and the processor is used to execute the program in the memory to perform any one of the methods described in the first aspect.

In an eighth aspect, embodiments of the present application provide a training apparatus, which may include a memory, a processor, and a bus system, where the memory is used for storing programs, and the processor is used for executing the programs in the memory to perform any one of the methods described in the second aspect.

In a ninth aspect, an embodiment of the present application provides a user operation behavior prediction apparatus, which may include a memory, a processor, and a bus system, where the memory is used to store a program, and the processor is used to execute the program in the memory to perform any one of the optional methods described in the third aspect.

In a tenth aspect, embodiments of the present application provide a computer-readable storage medium, where a computer program is stored, and when the computer program runs on a computer, the computer program causes the computer to perform the first aspect and any optional method, the second aspect and any optional method, and the third aspect and any optional method.

In an eleventh aspect, embodiments of the present application provide a computer program product, which includes code, when executed, for implementing the first aspect and any optional method, the second aspect and any optional method, and the third aspect and any optional method.

In a twelfth aspect, the present application provides a chip system, which includes a processor, configured to support an executing device or a training device to implement the functions involved in the foregoing aspects, for example, to transmit or process data involved in the foregoing methods; or, information. In one possible design, the system-on-chip further includes a memory for storing program instructions and data necessary for the execution device or the training device. The chip system may be formed by a chip, or may include a chip and other discrete devices.

The embodiment of the application provides a recommendation method, which comprises the following steps: acquiring a first operation information set of a target user, wherein the first operation information set comprises attribute information of a plurality of articles, a plurality of operation types and corresponding relations between the articles and the operation types, and the corresponding relations are used for representing operation types of the target user for operating the articles; performing feature extraction according to the first operation information set to determine a target user feature vector; acquiring a second operation information set of the target article, wherein the second operation information set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the target article operated by the plurality of users; performing feature extraction according to the second operation information set to determine a target article feature vector; according to the target user feature vector and the target article feature vector, outputting recommendation information based on a target recommendation model, wherein the recommendation information is used for representing the probability of the target user performing the operation of the plurality of operation types on the target article; and when the recommendation information meets a preset condition, determining to recommend the target article to the target user. Through the method, the target user feature vector representing the favorite of the target user is generated based on the article and the operation type with the association relation, and the target article feature vector representing the attraction feature of the target article to the user is generated based on the user and the operation type with the association relation, so that the probability of the target user performing operation of multiple operation types on the target article is predicted, and the operation probability of the user aiming at the article can be accurately depicted.

Drawings

FIG. 1 is a schematic structural diagram of an artificial intelligence body framework;

fig. 2 is a schematic diagram of a system architecture according to an embodiment of the present application;

fig. 3 is a schematic diagram of an information recommendation process according to an embodiment of the present application;

fig. 4 is a schematic flowchart of an information recommendation method according to an embodiment of the present application;

fig. 5 is a schematic diagram of operation information provided in an embodiment of the present application;

fig. 6 is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 7 is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 8 is a schematic diagram of an information recommendation method provided in an embodiment of the present application;

fig. 9 is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 10 is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 11 is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 12 is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 13 is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 14 is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 15 is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 16 is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 17 is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 18 is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 19a is a schematic diagram of an information recommendation method according to an embodiment of the present application;

fig. 19b is a schematic diagram of a user behavior prediction method according to an embodiment of the present application;

FIG. 20 is a schematic diagram of a recommendation model training method provided in an embodiment of the present application;

fig. 21 is a schematic diagram of an information recommendation device according to an embodiment of the present application;

fig. 22 is a schematic diagram of a user behavior prediction apparatus according to an embodiment of the present application;

FIG. 23 is a schematic diagram of a recommendation model training apparatus according to an embodiment of the present application;

FIG. 24 is a schematic diagram of an execution device provided in an embodiment of the present application;

FIG. 25 is a schematic view of a training apparatus provided in an embodiment of the present application;

fig. 26 is a schematic diagram of a chip according to an embodiment of the present disclosure.

Detailed Description

The embodiments of the present invention will be described below with reference to the drawings. The terminology used in the description of the embodiments of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

Embodiments of the present application are described below with reference to the accompanying drawings. As can be known to those skilled in the art, with the development of technology and the emergence of new scenes, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

The terms "first," "second," and the like in the description and claims of this application and in the foregoing drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances and are merely descriptive of the manner in which objects of the same nature are distinguished in the embodiments of the application. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of elements is not necessarily limited to those elements, but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.

The general workflow of the artificial intelligence system will be described first, please refer to fig. 1, which shows a schematic structural diagram of an artificial intelligence body framework, and the artificial intelligence body framework is explained below from two dimensions of "intelligent information chain" (horizontal axis) and "IT value chain" (vertical axis). Where "intelligent information chain" reflects a list of processes processed from the acquisition of data. For example, the general processes of intelligent information perception, intelligent information representation and formation, intelligent reasoning, intelligent decision making and intelligent execution and output can be realized. In this process, the data undergoes a "data-information-knowledge-wisdom" refinement process. The "IT value chain" reflects the value of artificial intelligence to the information technology industry from the underlying infrastructure of human intelligence, information (provision and processing technology implementation) to the industrial ecological process of the system.

(1) Infrastructure arrangement

The infrastructure provides computing power support for the artificial intelligent system, communication with the outside world is achieved, and support is achieved through the foundation platform. Communicating with the outside through a sensor; the computing power is provided by intelligent chips (hardware acceleration chips such as CPU, NPU, GPU, ASIC, FPGA and the like); the basic platform comprises distributed computing framework, network and other related platform guarantees and supports, and can comprise cloud storage and computing, interconnection and intercommunication networks and the like. For example, sensors and external communications acquire data that is provided to smart chips in a distributed computing system provided by the underlying platform for computation.

(2) Data of

Data at the upper level of the infrastructure is used to represent the data source for the field of artificial intelligence. The data relates to graphs, images, voice and texts, and also relates to the data of the Internet of things of traditional equipment, including service data of the existing system and sensing data such as force, displacement, liquid level, temperature, humidity and the like.

(3) Data processing

Data processing typically includes data training, machine learning, deep learning, searching, reasoning, decision making, and the like.

The machine learning and the deep learning can perform symbolized and formalized intelligent information modeling, extraction, preprocessing, training and the like on data.

Inference refers to the process of simulating human intelligent inference mode in a computer or an intelligent system, using formalized information to think and solve problems of a machine according to an inference control strategy, and the typical function is searching and matching.

The decision-making refers to a process of making a decision after reasoning intelligent information, and generally provides functions of classification, sequencing, prediction and the like.

(4) General capabilities

After the above-mentioned data processing, further based on the result of the data processing, some general capabilities may be formed, such as algorithms or a general system, e.g. translation, analysis of text, computer vision processing, speech recognition, recognition of images, etc.

(5) Intelligent product and industrial application

The intelligent product and industry application refers to the product and application of an artificial intelligence system in each field, and is the encapsulation of an artificial intelligence integral solution, the intelligent information decision is commercialized, and the application on the ground is realized, and the application field mainly comprises: intelligent terminal, intelligent transportation, intelligent medical treatment, autopilot, wisdom city etc..

The embodiment of the application can be applied to the field of information recommendation, and particularly can be applied to application markets, music playing recommendation, video playing recommendation, reading recommendation, news information recommendation, information recommendation in webpages and the like. The application can be applied to a recommendation system, and the recommendation system can determine a recommendation object based on the recommendation method provided by the application, where the recommendation object may be, for example and without limitation, an Application (APP), an audio/video, a webpage, news information, and other items.

In the recommendation system, information recommendation can include processes such as prediction and recommendation. What needs to be solved for prediction is to predict the preference degree of the user for each item, and the preference degree can be reflected by the probability of the user selecting the item. The recommendation may be to sort the recommendation objects according to the predicted result, for example, according to the predicted preference degree, sorting in the order of preference degrees from high to low, and recommend information to the user based on the sorted result.

For example, in a scenario of an application market, the recommendation system may recommend an application program to the user based on the result of the ranking, in a scenario of a music recommendation, the recommendation system may recommend music to the user based on the result of the ranking, and in a scenario of a video recommendation, the recommendation system may recommend a video to the user based on the result of the ranking.

Next, an application architecture of the embodiment of the present application is described.

The system architecture provided by the embodiment of the present application is described in detail below with reference to fig. 2. Fig. 2 is a schematic diagram of a system architecture according to an embodiment of the present application. As shown in FIG. 2, the system architecture 500 includes an execution device 510, a training device 520, a database 530, a client device 540, a data storage system 550, and a data collection system 560.

The execution device 510 includes a computation module 511, an I/O interface 512, a pre-processing module 513, and a pre-processing module 514. The target model/rule 501 may be included in the calculation module 511, with the pre-processing module 513 and the pre-processing module 514 being optional.

The data acquisition device 560 is used to collect training samples. In an embodiment of the present application, the training sample may be a historical operation record of the user, the historical operation record may be a behavior log (logs) of the user, the historical operation record may include operation information of the user for an item, where the operation information may include an operation type, an identifier of the user, and an identifier of the item, where the item is an e-commerce product, the operation type may include, but is not limited to, clicking, purchasing, returning, joining a shopping cart, and the like, where the item is an application program, the operation type may be, but is not limited to, clicking, downloading, and the like, and the training sample is data used when training an initialized recommendation model. After the training samples are collected, the data collection device 560 stores the training samples in the database 530.

It should be understood that, in the embodiment of the present application, the data collecting device 560 may mine the collected historical operation records to obtain more operation information of the user for the article (for example, the first operation type of the first user on the first article in the embodiment of the present application);

training device 520 may train the initialized recommendation model based on training samples maintained in database 530 to arrive at target model/rules 501. In this embodiment of the application, the target model/rule 501 may be a recommendation model, and the recommendation model may predict, based on the operation information of the user for the article, a probability that the user performs an operation corresponding to the operation type for the article, where the probability may be used to perform information recommendation.

It should be noted that, in practical applications, the training samples maintained in the database 530 do not necessarily all come from the collection of the data collection device 560, and may also be received from other devices, or may be obtained by performing data expansion based on the data collected by the data collection device 560 (for example, the second operation type of the target user on the first item in the embodiment of the present application). It should be noted that, the training device 520 does not necessarily perform the training of the target model/rule 501 based on the training samples maintained by the database 530, and may also obtain the training samples from the cloud or other places for performing the model training, and the above description should not be taken as a limitation to the embodiments of the present application.

The target model/rule 501 obtained by training according to the training device 520 may be applied to different systems or devices, for example, the executing device 510 shown in fig. 2, where the executing device 510 may be a terminal, such as a mobile phone terminal, a tablet computer, a notebook computer, an Augmented Reality (AR)/Virtual Reality (VR) device, a vehicle-mounted terminal, or a server or a cloud.

In fig. 2, the execution device 510 configures an input/output (I/O) interface 512 for data interaction with an external device, and a user may input data (for example, operation information of the second user with respect to the first item, operation information of the second user with respect to the second item, operation information of the first user with respect to the first item, and the like in this embodiment of the application) to the I/O interface 512 through the client device 540.

The pre-processing module 513 and the pre-processing module 514 are configured to perform pre-processing according to input data received by the I/O interface 512. It should be understood that there may be no pre-processing module 513 and pre-processing module 514 or only one pre-processing module. When the pre-processing module 513 and the pre-processing module 514 are not present, the input data may be directly processed by the calculation module 511.

During the process of preprocessing the input data by the execution device 510 or performing the calculation and other related processes by the calculation module 511 of the execution device 510, the execution device 510 may call the data, the code and the like in the data storage system 550 for corresponding processes, or store the data, the instruction and the like obtained by corresponding processes in the data storage system 550.

Finally, the I/O interface 512 presents the processing results to the client device 540 for presentation to the user.

In this embodiment of the present application, the execution device 510 may obtain a code stored in the data storage system 550 to implement the recommendation method in this embodiment of the present application.

In this embodiment, the execution device 510 may include a hardware circuit (e.g., an Application Specific Integrated Circuit (ASIC), a field-programmable gate array (FPGA), a general-purpose processor, a Digital Signal Processor (DSP), a microprocessor, a microcontroller, or a combination of these hardware circuits), for example, the execution device 510 may be a hardware system with an instruction execution function, such as a CPU, a DSP, or a hardware system without an instruction execution function, such as an ASIC, an FPGA, or a combination of the above hardware systems without an instruction execution function and the hardware system with an instruction execution function.

Specifically, the execution device 510 may be a hardware system having a function of executing instructions, the information recommendation method provided in the embodiment of the present application may be a software code stored in the data storage system 550, and the execution device 510 may acquire the software code from the data storage system 550 and execute the acquired software code to implement the recommendation method provided in the embodiment of the present application.

It should be understood that the executing device 510 may be a combination of a hardware system without a function of executing instructions and a hardware system with a function of executing instructions, and some steps of the recommended method provided by the embodiment of the present application may also be implemented by a hardware system without a function of executing instructions in the executing device 510, which is not limited herein.

In the case shown in fig. 2, the user can manually give input data, and this "manual giving of input data" can be operated through an interface provided by the I/O interface 512. Alternatively, the client device 540 may automatically send the input data to the I/O interface 512, and if the client device 540 is required to automatically send the input data to obtain authorization from the user, the user may set the corresponding permissions in the client device 540. The user can view the result output by the execution device 510 at the client device 540, and the specific presentation form can be display, sound, action, and the like. The client device 540 may also serve as a data collection terminal, collecting input data of the input I/O interface 512 and output results of the output I/O interface 512 as new sample data, as shown, and storing the new sample data in the database 530. Of course, the input data inputted to the I/O interface 512 and the output result outputted from the I/O interface 512 as shown in the figure may be directly stored in the database 530 as new sample data by the I/O interface 512 without being collected by the client device 540.

It should be noted that fig. 2 is only a schematic diagram of a system architecture provided in an embodiment of the present application, and the position relationship between the devices, modules, and the like shown in the diagram does not constitute any limitation, for example, in fig. 2, the data storage system 550 is an external memory with respect to the execution device 510, and in other cases, the data storage system 550 may be disposed in the execution device 510. It is understood that the execution device 510 described above may be deployed in the client device 540.

Since the embodiments of the present application relate to the application of a large number of neural networks, for the sake of understanding, the related terms and related concepts such as neural networks related to the embodiments of the present application will be described first.

1. Click probability (click-through, CTR)

The click probability may also be referred to as a click rate, and refers to a ratio of the number of times that recommended information (e.g., recommended articles) on a website or an application is clicked to the number of times that recommended articles are exposed, and the click rate is generally an important index for measuring a recommendation system in the recommendation system.

2. Personalized recommendation system

The personalized recommendation system is a system which analyzes by using a machine learning algorithm according to historical data (such as operation information in the embodiment of the application) of a user, predicts a new request according to the analysis, and gives a personalized recommendation result.

3. Off-line training (offflintraining)

The offline training refers to a module that iteratively updates recommendation model parameters according to an algorithm learned by a recommendation machine in a personalized recommendation system according to historical data (such as operation information in the embodiment of the present application) of a user until a set requirement is met.

4. Online prediction (onlineinterference)

The online prediction means that the preference degree of a user to recommended articles in the current context environment is predicted according to the characteristics of the user, the articles and the context and the probability of selecting the recommended articles by the user is predicted based on an offline trained model.

For example, fig. 3 is a schematic diagram of a recommendation system provided in an embodiment of the present application. As shown in FIG. 3, when a user enters the system, a request for recommendation is triggered, and the recommendation system inputs the request and its related information (e.g., operational information in the embodiments of the present application) into the recommendation model, and then predicts the user's selection rate of items in the system. Further, the items may be sorted in descending order according to the predicted selection rate or based on some function of the selection rate, i.e., the recommendation system may present the items in different locations in sequence as a result of the recommendation to the user. The user browses various located items and takes user actions such as browsing, selecting, and downloading. Meanwhile, the actual behavior of the user can be stored in the log as training data, and the parameters of the recommended model are continuously updated through the offline training module, so that the prediction effect of the model is improved.

For example, a user opening an application market in a smart terminal (e.g., a cell phone) may trigger a recommendation system in the application market. The recommendation system of the application market predicts the probability of downloading each recommended candidate APP by the user according to the historical behavior log of the user, for example, the historical downloading record and the user selection record of the user, and the self characteristics of the application market, such as the environmental characteristic information of time, place and the like. According to the calculated result, the recommendation system of the application market can display the candidate APPs in a descending order according to the predicted probability value, so that the downloading probability of the candidate APPs is improved.

For example, the APP with the higher predicted user selection rate may be presented at the front recommended position, and the APP with the lower predicted user selection rate may be presented at the rear recommended position.

The recommended model may be a neural network model, and the following describes terms and concepts related to a neural network that may be involved in embodiments of the present application.

(1) Neural network

The neural network may be composed of neural units, and the neural units may refer to operation units with xs (i.e. input data) and intercept 1 as inputs, and the output of the operation units may be:

wherein s =1, 2, \8230, n is a natural number greater than 1, ws is the weight of xs, and b is the bias of the neural unit. f is an activation function (activation functions) of the neural unit for introducing a nonlinear characteristic into the neural network to convert an input signal in the neural unit into an output signal. The output signal of the activation function may be used as an input for the next convolutional layer, and the activation function may be a sigmoid function. A neural network is a network formed by a plurality of the above-mentioned single neural units being joined together, i.e. the output of one neural unit may be the input of another neural unit. The input of each neural unit can be connected with the local receiving domain of the previous layer to extract the characteristics of the local receiving domain, and the local receiving domain can be a region composed of a plurality of neural units.

(2) Deep neural network

Deep Neural Networks (DNNs), also known as multi-layer Neural networks, can be understood as Neural networks having many hidden layers, where "many" has no particular metric. From the division of DNNs by the location of different layers, neural networks inside DNNs can be divided into three categories: input layer, hidden layer, output layer. Generally, the first layer is an input layer, the last layer is an output layer, and the middle layers are hidden layers. The layers are all connected, that is, any neuron of the ith layer is necessarily connected with any neuron of the (i + 1) th layer. Although DNN appears complex, it is not really complex in terms of the work of each layer, simply the following linear relational expression:

wherein the content of the first and second substances,

is a function of the input vector or vectors,

is the output vector of the output vector,

is an offset vector, W is a weight matrix (also called coefficient), and α () is an activation function. Each layer is only for the input vector

Obtaining the output vector through such simple operation

Due to the large number of DNN layers, the coefficient W and the offset vector

The number of the same is large. The definition of these parameters in DNN is as follows: taking the coefficient W as an example: assume that in a three-layer DNN, the linear coefficients of the 4 th neuron of the second layer to the 2 nd neuron of the third layer are defined as

Superscript 3 represents the number of layers in which the coefficient W lies, and the subscripts correspond to the third layer index 2 at the output and the second layer index 4 at the input. The summary is that: the coefficients of the kth neuron of the L-1 th layer to the jth neuron of the L-1 th layer are defined as

Note that the input layer is without the W parameter. In deep neural networks, more hidden layers make the network more able to depict complex situations in the real world. Theoretically, the more parameters the higher the model complexity, the larger the "capacity", which means that it can accomplish more complex learning tasks. Training depth spiritThe final goal of the process of learning the weight matrix through the network is to obtain the weight matrix (formed by the vectors W of many layers) of all layers of the trained deep neural network.

(3) Loss function

In the process of training the deep neural network, because the output of the deep neural network is expected to be as close to the value really expected to be predicted as possible, the weight vector of each layer of the neural network can be updated according to the difference between the predicted value of the current network and the really expected target value by comparing the predicted value of the current network and the really expected target value (of course, an initialization process is usually carried out before the first update, namely parameters are pre-configured for each layer in the deep neural network), for example, if the predicted value of the network is high, the weight vector is adjusted to be lower in prediction, and the adjustment is carried out continuously until the deep neural network can predict the really expected target value or the value which is very close to the really expected target value. Therefore, it is necessary to define in advance "how to compare the difference between the predicted value and the target value", which are loss functions (loss functions) or objective functions (objective functions), which are important equations for measuring the difference between the predicted value and the target value. Taking the loss function as an example, if the higher the output value (loss) of the loss function indicates the larger the difference, the training of the deep neural network becomes the process of reducing the loss as much as possible.

(4) Back propagation algorithm

The size of the parameters in the initial model can be corrected in the training process by adopting a Back Propagation (BP) algorithm, so that the error loss of the model is smaller and smaller. Specifically, an error loss occurs when an input signal is transmitted in a forward direction until an output signal is output, and parameters in an initial model are updated by back-propagating error loss information, so that the error loss converges. The back propagation algorithm is an error-loss dominated back propagation motion aimed at obtaining optimal model parameters, such as a weight matrix.

Next, a model inference phase is taken as an example to explain an information recommendation method provided by the embodiment of the present application.

Referring to fig. 4, fig. 4 is a schematic diagram of an embodiment of a recommendation method provided in an embodiment of the present application, and as shown in fig. 4, the recommendation method provided in the embodiment of the present application includes:

401. the method comprises the steps of obtaining a first operation information set of a target user, wherein the first operation information set comprises attribute information of a plurality of articles, a plurality of operation types and corresponding relations between the articles and the operation types, and the corresponding relations are used for representing operation types of the target user for operating the articles.

In embodiments of the present application, the executing subject of step 401 may be a terminal device, which may be a portable mobile device, such as, but not limited to, a mobile or portable computing device (e.g., a smartphone), a personal computer, a server computer, a handheld device (e.g., a tablet) or laptop, a multiprocessor system, a gaming console or controller, a microprocessor-based system, a set top box, a programmable consumer electronics, a mobile phone, a mobile computing and/or communication device with a wearable or accessory form factor (e.g., a watch, glasses, a headset or earpiece), a network PC, a minicomputer, a mainframe computer, a distributed computing environment that includes any of the above systems or devices, and so on.

In this embodiment, the execution main body in step 401 may be a server on the cloud side, and the server may receive the first operation information set of the target user sent by the terminal device, so that the server may obtain the first operation information set of the target user.

For convenience of description, the following description will be made as an execution apparatus without distinguishing the form of the execution main body. In this embodiment, the execution device may obtain a first operation information set of the target user, where the first operation information set of the target user may be obtained based on interaction records (for example, behavior logs of the user) between the target user and the articles, information in the first operation information set may include real operation records of the target user on each article, and the first operation information set may include attribute information of the target user, attribute information of each article, and operation types of operations performed by the target user on the articles.

The attribute information of the target user may be at least one of attributes related to user preference characteristics, gender, age, occupation, income, hobbies and education level, wherein gender may be male or female, age may be a number between 0 and 100, occupation may be a teacher, a programmer, a chef and the like, hobbies may be basketball, tennis, running and the like, education level may be primary school, junior school, high school, university and the like; the present application does not limit the specific type of attribute information of the target user.

The article may be an entity article or a virtual article, for example, the article may be an article such as APP, audio/video, a webpage, and news information, the attribute information of the article may be at least one of an article name, a developer, an installation package size, a category, and a goodness of evaluation, where, taking the article as an application program as an example, the category of the article may be a chat category, a cool game category, an office category, and the like, and the goodness of evaluation may be a score, a comment, and the like for the article; the application does not limit the specific type of attribute information for the article.

The operation type may be a behavior operation type of a target user for an article, and on a network platform and an application, the user often has various interaction forms (that is, various operation types) with the article, for example, the operation types of the user in browsing, clicking, adding to a shopping cart, purchasing and the like in the behavior of an e-commerce platform as shown in fig. 5. These various behaviors reflect the preferences of the user and are of great help to accurately characterize the user.

In one possible implementation, the first operation information set may include attribute information of a first article, a first operation type, and a correspondence between the first article and the first operation type; wherein the first operation type is the real operation behavior of the target user on the first article. That is to say, the attribute information of the first article, the first operation type, and the corresponding relationship between the first article and the first operation type are obtained based on the actual operation record of the target user.

For example, the operation information in fig. 5 may include: a browsing operation by the user 1 for the item 1, a purchase operation by the user 1 for the item 1, a shopping cart joining operation by the user 1 for the item 1, a browsing operation by the user 2 for the item 1, and a purchase operation by the user 2 for the item 1.

In this embodiment of the present application, the operation information may be expressed as a triplet information format, where the triplet may be in a form of < user, relationship, or article >, where a user in the triplet may be information of the user, an article may be information of the article, and the relationship may be an operation type, for example, a purchase operation of the user 1 for the article 2 may be expressed as: < user 1, purchase operation, item 2>.

In one possible implementation, the first set of operational information includes: attribute information of a second article, a second operation type and a corresponding relation between the second article and the second operation type; and the second operation type is the predicted operation behavior of the target user on the second article.

In the embodiment of the application, because the historical operation records related to the user can be acquired in a limited manner, in order to improve the accuracy of information recommendation, richer information can be discovered in the limited historical operation data, so that more operation information related to the user can be generated, and the data utilization rate of the historical operation records can be improved.

Next, how to generate the predicted operation behavior of the target user on the second item is described:

in a possible implementation, a third operation type of a second article by a first user may be obtained, and based on that the third operation type of the second article by the first user and the article preference characteristics of the first user and the target user meet a preset condition, a corresponding relationship between the second article and the second operation type is generated in the first operation information set.

When different users (e.g., the first user and the target user) operate on the same article, it may be characterized that the two users have a higher possibility of having the same or similar preference, and therefore, the target user-related operation information may be generated based on the operation information of the first user on other articles, and this portion of the generated new target user-related operation information may, although not in the historical operation records, generally represent possible operation associations between the target user and other articles except the first article.

In a possible implementation, the target user and the first user may both perform an operation on the first item, for example, the target user performs a browsing operation on the first item, and the first user performs a purchasing operation on the first item, and therefore the target user and the first user may be considered as users having similar preferences, and therefore, more operation information related to the target user may be generated based on the historical operation information of the first user on other items, and in particular, the operation information of the target user on the second item may be generated based on the historical operation record that also records the operation information of the first user on the second item.

Referring to fig. 6, the first user and the target user both operate the first article, the first user performs an operation on the first article, and the operation type is the second operation type, and the generated operation on the second article by the target user is the third operation type.

Illustratively, the operation information may be based on: if < target user, collection operation, first item >, < first user, collection operation, first item > and < first user, purchase operation, second item >, operation information of < target user, purchase operation, second item > may be generated.

In the embodiment of the application, based on users (the target user and the first user) who operate on the same article (the first article), new operation information related to the target user is generated through other operation information of the first user, richer information is found in limited historical operation data, and the data utilization rate of the historical operation record is improved.

Referring to fig. 7, the second user and the first user both operate the second article, and the second user also operates the third article, so that the operation information of the target user with respect to the third article may be generated.

Illustratively, for example, the historical operation information includes < target user, collecting operation, first item >, < first user, purchasing operation, second item >, < second user, browsing operation, second item > and < second user, purchasing operation, third item >, so that the operation information of < target user, purchasing operation, third item > can be generated.

In the above manner, similarity of preference among users is reflected by whether the users operate on the same article, and in a possible implementation, the similarity of preference among users can also be determined directly through user attribute differences among users, wherein the user attribute is an attribute capable of reflecting the preference of users.

In this embodiment of the present application, users having similar user attributes may be considered to have similar preferences, for example, users having similar user attributes may be users with small age differences, users having the same gender, users having the same or similar professions (professionally similar may be understood to be in the same industry), users having the same or similar hobbies (for example, users all hobbies about tennis), and users having the same or similar education degree (for example, users all of college homemade graduates).

For example, the operation information may include < target user, favorite operation, first item >, < first user, purchase operation, second item >, and the operation information of < target user, purchase operation, second item > may be generated based on the user attribute difference between the target user and the first user being less than the threshold value.

In the above manner, similarity of preference among users is determined by user attribute difference among users, and in a possible implementation, similarity of preference among users may also be determined by whether users operate on items having similar item attributes.

In one possible implementation, the preset condition may include that the first user and the target user are users who operate on an article whose article attribute difference is smaller than a threshold value.

In one possible implementation, the item attributes may be: the system comprises at least one of an article name, a developer, an installation package size, a category and a goodness, wherein the article is taken as an application program as an example, the category of the article can be chatting, running and cool games, office games and the like, and the goodness can be scores, comments and the like aiming at the article.

In a possible implementation, the first user operates the first article, the target user operates the fourth article, and the target user and the first user may be considered to be users having similar preferences based on that the degree of difference between the article attributes of the first article and the fourth article is less than the threshold, so that more operation information related to the target user may be generated based on the historical operation information of the first user with respect to other articles, and in particular, the operation information of the target user with respect to the second article may be generated based on that the historical operation record further records the operation information of the first user with respect to the second article.

In order to ensure the accuracy of the generated operation information, the operation type of the target user for the second article in the operation information of the target user for the second article may be set to be consistent with the operation type of the first user for the second article in the historical operation record (that is, the second operation type and the third operation type are the same).

Illustratively, for example, the operation information includes < target user, collecting operation, first item >, < first user, browsing operation, fourth item >, < first user, purchasing operation, second item >, and based on the item attribute difference between the first item and the fourth item being less than the threshold value, the operation information of < target user, purchasing operation, second item > may be generated.

By the above method, the high-order node relationship in the history operation record is mined, more operation information is generated, and the data mining method can be used simultaneously to realize the mining of the maximum degree of the data, and the following describes the above process with an example:

the following historical operating records may be obtained:

to be provided with<User, type of operation, article>The triple represents a multi-behavior ternary heterogeneous graph network, and the historical operation records can be simply represented as:<user 1, purchase, item 1>、<User 1, collection, item 2>、<User 1, collection, item 3>、<User 2, collection, item 2>、<The number of the users 2 is reduced to a certain number,click on, item 3>、<User 3, click, item 3>。

Since the item 2 is collected by the user 2 and is collected by the user 1, new operation information for the user 2 can be generated: < user 2, purchase, item 1>, and so on can obtain operational information: < user 3, purchase, item 1>.

Further, the user attribute information of the user may be:

and combining the operation information of the user, and deducing through the graph network collaborative filtering to obtain the following multi-order ternary data:

to be provided with<User 4, purchase, item 1>Indicate that by analogy, can obtain<User 5, buy, item 1>And (4) data. Further, the item attribute of the item may be: item names, item classes, labels, scores, comments, or the like, e.g.

The same treatment can also be obtained

To be provided with<User 1, buy, item 4>And (4) showing.

Through the data mining process, the operation information which originally only comprises 6 ternary elements is expanded into 10 operation information.

402. And performing feature extraction according to the first operation information set to determine a target user feature vector.

After the first operation information set of the target user is obtained, feature extraction may be performed according to the first operation information set to determine a target user feature vector.

Next, how to determine a target user feature vector by feature extraction according to the first operation information set is described:

in a possible implementation, the first operation information set may include a plurality of groups of articles and operation types having a correspondence, and a sub-user feature vector may be obtained by calculating each group of articles and operation types having a correspondence, and then the calculated sub-user feature vectors are fused to obtain a target user feature vector. And each sub-user feature vector is obtained by extracting the features of the attribute information and the operation type of the article with the corresponding relationship.

In one possible implementation, the first set of operational information includes: attribute information of a first article, a first operation type, and a corresponding relationship between the first article and the first operation type; and calculating to obtain a sub-user feature vector based on the attribute information of the first article and the first operation type.

Wherein the attribute information of the first item can be represented as an embedded vector embedding (e.g., to

Representing attribute information of the item, l may be equal to 1), the first operation type may be represented as an embedded vector (e.g., to

Representing an operation type), the sub-user feature vector may be represented by a similarity between the attribute information of the first item and the first operation type.

Referring to fig. 8 and 9, the sub-user feature vector may be calculated, for example, as follows:

where σ (x) is an activation function, for example the activation function may be a sigmoid function.

Refers to a collection of operation types and items that have an association relationship with a target user. W ^(l) Is the weight of the neurons of the l-th layer,

defined as the bitwise multiplication of elements between two vectors.

In one possible implementation, the first set of operational information includes: attribute information of a second article, a second operation type and a corresponding relation between the second article and the second operation type; and calculating to obtain a sub-user feature vector based on the attribute information of the second object and the second operation type.

Wherein the information of the second item may be represented as an embedded vector (e.g., to

Information representing a second item, l may equal 2), and the second operation type may be represented as an embedded vector (e.g., to

Representing a second operation type), the second feature vector may be represented by a similarity between the information of the second item and the second operation type, and referring to fig. 10, the second feature vector of the second user may be calculated, for example, as follows:

Refers to a collection of operation types and items that have an association with a target user. W ^(l) Is the weight of the neurons of the l-th layer,

defined as the bitwise multiplication of elements between two vectors.

In this way, a plurality of sub-user feature vectors may be obtained, wherein a part of the sub-user feature vectors may be regarded as first-order feature vectors of the target user (based on real operation information, such as attribute information of the first item, the first operation type, and a corresponding relationship between the first item and the first operation type), and a part of the sub-user feature vectors may be regarded as second-order feature vectors of the target user (based on predicted operation information, such as attribute information of the second item, the second operation type, and a corresponding relationship between the second item and the second operation type), and similarly, higher-order feature vectors of the target user may also be obtained.

In a possible implementation, each sub-user feature vector may represent a favorite feature of the target user, so that multiple sub-user feature vectors may be fused, and for the sub-user feature vectors of the same order, an activation function may be used for fusion, for example, σ (x) in the above formula, one feature vector result that may be obtained for the sub-user feature vector fusion of the same order, multiple feature vector results that may be obtained for the sub-user feature vectors of multiple orders, and further, multiple feature vector results may be fused to obtain the target user feature vector, where the fusion may be, but is not limited to, an addition and concatenation operation (concat).

In a possible implementation, referring to fig. 11, because the operation information of the target user for the first article is data in the historical operation record, that is, real data, and the operation information of the target user for the second article is inferred based on other data in the historical operation record, and is not hundred-percent accurate, when fusing is performed, the weight occupied by the operation information of the target user for the first article may be set to be greater than the weight occupied by the operation information of the target user for the second article, and then the favorite feature of the target user represented by the fused operation information may be more accurate.

In one possible implementation, referring to fig. 12, a plurality of sub-user feature vectors may be fused, and for example, the feature vectors may be fused according to the following formula:

where L represents the order of the feature vector.

In one possible implementation, referring to FIG. 13, the initialization embedded vector of the target user may also be the object of the fusion operation.

In one possible implementation, higher order (greater than 2 order) feature vectors of the target user may also be the subject of the fusion operation.

In one possible implementation, a first subset of operational information and a second subset of operational information may be obtained; wherein the first operation information subset comprises attribute information of a first article, a first operation type and a corresponding relation between the first article and the first operation type; the first operation type is a real operation type of the target user on the first article; the second operation information subset comprises attribute information of a second article, a second operation type and a corresponding relation between the second article and the second operation type; the second operation type is a predicted operation type of the target user on the second article; the determining a plurality of sub-user feature vectors according to the first operation information set includes: determining a first sub-user feature vector according to the first operation information subset; and determining a second sub-user characteristic vector according to the second operation information subset.

Because the historical operation records related to the user can be acquired in a limited manner, in order to improve the accuracy of information recommendation, richer information can be discovered in the limited historical operation data, so that more operation information related to the user can be generated, and the data utilization rate of the historical operation records can be improved.

In one possible implementation, the first sub-user feature vector and the second sub-user feature vector may be fused according to a first weight of the first sub-user feature vector and a second weight of the second sub-user feature vector.

403. Acquiring a second operation information set of the target article, wherein the second operation information set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the target article operated by the plurality of users.

Different from the first operation information set, the second operation information set includes the plurality of users and the plurality of operation types having correspondence with the target item.

In a possible implementation, the second operation information set may include four operation information subsets, and the fourth operation information subset may include attribute information of a third user, a fifth operation type, and a corresponding relationship between the third user and the fifth operation type; wherein the fifth operation type is a predicted operation type of the target item by the third user.

In a possible implementation, a sixth operation type of the target item by a fourth user may be obtained, where the item preference characteristics of the fourth user and the third user meet a preset condition; and acquiring the fifth operation type based on a sixth operation type of the target object by the fourth user.

In one possible implementation, the preset condition includes at least one of:

In a possible implementation, the fifth operation type is obtained based on a sixth operation type of the target article by the fourth user, specifically, the fifth operation type is obtained according to the sixth operation type, for example, the fifth operation type is the same as the sixth operation type.

For the extended policy of the operation information, reference may be made to the description of step 401 in the foregoing embodiment, and similar parts are not described again.

404. And performing feature extraction according to the second operation information set to determine a target article feature vector.

In a possible implementation, feature extraction may be performed based on attribute information and an operation type of a user having a correspondence relationship to obtain a sub-article feature vector, where the sub-article feature vector may represent an attraction feature of a target article for the user, and then a plurality of sub-article feature vectors may be obtained, and the plurality of sub-article feature vectors are fused to obtain the target article feature vector.

In one possible implementation, the second set of operational information includes: attribute information of the second user, a fourth forward operation type, and a corresponding relationship between the second user and the fourth forward operation type, a sub-item feature vector of the target item may be obtained according to the attribute information of the second user and the fourth forward operation type, where the attribute information of the second user may be represented as an embedded vector (for example, to represent the attribute information of the second user as an embedded vector

Attribute information representing the second user, l may be equal to 1), the first operation type may be represented as an embedded vector (e.g., to

Representing the first operation type), the fourth feature vector may be represented by a similarity between the information of the second user and the fourth operation type, and referring to fig. 14, the sub-item feature vector of the target item may be calculated, for example, as follows:

Refers to a set of users and operation types that have an association relationship with a target item. W is a group of ^(l) Is the weight of the neurons of the l-th layer,

defined as the bitwise multiplication of the elements between two vectors.

In one possible implementation, a sub-item feature vector of the target item may be determined based on the operation information of the second user for the second item, wherein the sub-item feature vector may be used to characterize an attraction feature of the target item for the second user.

In one possible implementation, the second set of operational information includes: attribute information of a third user, a fifth operation type and a corresponding relation between the third user and the fifth operation type; the fifth operation type is a predicted operation behavior of the third user on the target item, and a sub-item feature vector of the target user may be obtained according to attribute information of the third user and the fifth operation type, where the information of the first user may be represented as an embedded vector embedding (for example, to embed the embedded vector into the target item

Information representing the first user, l may equal 2), and the fifth operation type may be represented as an embedded vector (e.g., to embed the vector

Representing the second operation type), the sub-item feature vector may be represented by a similarity between the information of the third user and the fifth operation type, and referring to fig. 15, the sub-item feature vector of the target item may be calculated, for example, as follows:

Refers to a collection of operation types and users that have an association relationship with the target item. W ^(l) Is the weight of the neurons of the l-th layer,

defined as the bitwise multiplication of an element between two vectorsAnd (5) operating.

Referring to fig. 16, in the above manner, a plurality of sub-item feature vectors may be obtained, where a part of the sub-item feature vectors may be regarded as first-order feature vectors of the target item (based on real operation information, such as attribute information of the second user, a fourth forward operation type, and a corresponding relationship between the second user and the fourth forward operation type), a part of the sub-item feature vectors may be regarded as second-order feature vectors of the target user (based on predicted operation information, such as attribute information of the third user, a fifth operation type, and a corresponding relationship between the third user and the fifth operation type), and similarly, higher-order feature vectors of the target item may also be obtained.

In one possible implementation, each sub-item feature vector may represent an attraction feature of the target item for the user, so that a plurality of sub-item feature vectors may be fused, and for the sub-item feature vectors of the same order, an activation function may be used for fusion, such as σ (x) in the above formula, one feature vector result may be obtained for the sub-item feature vector fusion of the same order, a plurality of feature vector results may be obtained for the sub-item feature vectors of a plurality of orders, and further, the plurality of feature vector results may be fused to obtain the target item feature vector, where the fusion may be, but is not limited to, an addition and concatenation operation (concat).

In a possible implementation, because the operation information of the second user for the target object is data in the historical operation record, that is, real data, which is hundred percent accurate, and the operation information of the third user for the target object is inferred based on other data in the historical operation record, which is not hundred percent accurate, in the fused operation information, the weight occupied by the operation information of the second user for the target object may be set to be greater than the weight occupied by the operation information of the second user for the target object, and then the attraction characteristic of the target object for the user represented by the fused operation information may be more accurate.

In one possible implementation, referring to fig. 17, a plurality of sub-item feature vectors may be fused, and for example, the feature vectors may be fused according to the following formula:

where L represents the order of the feature vector.

In one possible implementation, referring to FIG. 18, the initialization embedded vector for the target item may also be the subject of the fusion operation.

In one possible implementation, higher order (greater than 2 order) feature vectors of the target item may also be the subject of the fusion operation.

In one possible implementation, an operation type feature vector may be obtained for each of the plurality of operation types; wherein each operation type feature vector may be obtained when training the target recommendation model.

In one implementation, the operation types in the operation information of different orders can all obtain a corresponding operation type sub-feature vector, and the operation type sub-feature vectors of each of the multiple operation types can be obtained by fusing the multiple operation type sub-feature vectors.

For example, the operation type feature vector of the operation type in the operation information of the previous stage may be processed through a fully connected network to obtain an operation type feature vector of the operation type in the operation information of the next stage, which may be calculated through the following formula:

wherein, the first and the second end of the pipe are connected with each other,

is the weight of the neuron or neurons,

an operation type feature vector which is an operation type in the operation information of the previous stage,

and the operation type feature vector is the operation type in the adjacent operation information of the next level.

In one possible implementation, the method can be implemented by

And with

And fusing to obtain the operation type feature vector.

Illustratively, the fusion of feature vectors may be performed with reference to the following formula:

where L represents the order of the feature vector.

405. And outputting recommendation information based on a target recommendation model according to the target user feature vector and the target article feature vector, wherein the recommendation information is used for representing the probability of the target user performing the operations of the plurality of operation types on the target article.

By the method, a target user feature vector of a target user, a target article feature vector of a target article and a plurality of operation type feature vectors of a plurality of operation types can be obtained, recommendation information can be output based on a target recommendation model according to the plurality of operation type feature vectors of the plurality of operation types, the target user feature vector and the target article feature vector, and specifically, the probability of each operation type operation performed on the target article by the target user can be determined based on the similarity among the target user feature vector, the target article feature vector and each operation type feature vector in the plurality of operation type feature vectors.

For example, the predicted probability of the operation type k on the item v by the user u may be calculated based on the following manner:

where d is a model parameter, and is the length of the hidden vector, which may be 128, for example.

For example, referring to fig. 19a and fig. 19a, which illustrate a flow schematic of an information recommendation method, in one possible implementation, a similarity between a feature vector of a target user and a feature vector of a target item may be first calculated, and then a similarity between a feature vector of a second user, a feature vector of a target item of the target item, and a feature vector of an operation type may be calculated.

406. And when the recommendation information meets a preset condition, determining to recommend the target article to the target user.

By the method, the probability of the target user performing the operation corresponding to the plurality of operation types of the target object can be obtained, information recommendation is performed based on the probability, and specifically, when the recommendation information meets the preset condition, the target object can be determined to be recommended to the target user.

The preset conditions are described next:

in a possible implementation, a comprehensive value of probabilities of a plurality of operation types of the target user on each article may be calculated to represent a recommendation index of each article on the target user, the comprehensive value may be based on a weighted summation manner, and specifically, a corresponding weight may be set for each operation type, for example, the weight of a purchase operation is greater than the weight of an operation of joining a shopping cart, and then the recommendation index of each operation type may be obtained based on the weighted summation by combining the weight corresponding to each operation type and the probability corresponding to each operation type;

The following describes the beneficial effects of the embodiments of the present application in conjunction with experiments, and the information recommendation methods provided by the embodiments of the present application are compared with several existing technologies (BPR, NCF, lightGCN, CMF, MC-BPR, NMTR and EHCF) matured in the industry as follows. By using a public e-commerce data set, the statistics of the data set are as follows: user number: 48749; the number of articles: 39493; the number of browsing behaviors: 1548126; adding shopping carts: 193747; number of purchases: 259747. the method is trained with the other data as a training set, except that the last purchase record of the user is taken as a test sample. The invention carries out comparison test by selecting HR (hit rate) and NDCG which are recognized in the industry as evaluation indexes.

Table 1: comparison effect with existing recommendation model in public data set

As shown by the results in table 1, the following conclusions can be drawn: firstly, the result of considering multiple operation types of a user is better than the result of using only one operation type, and the accuracy of the recommendation algorithm can be improved by considering the high-order relationship among the nodes.

The embodiment of the application provides a recommendation method, which comprises the following steps: acquiring a first operation information set of a target user, wherein the first operation information set comprises attribute information of a plurality of articles, a plurality of operation types and corresponding relations between the articles and the operation types, and the corresponding relations are used for representing operation types of the target user for operating the articles; performing feature extraction according to the first operation information set to determine a target user feature vector; acquiring a second operation information set of the target article, wherein the second operation information set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the target article operated by the plurality of users; performing feature extraction according to the second operation information set to determine a feature vector of the target article; outputting recommendation information based on a target recommendation model according to the target user feature vector and the target article feature vector, wherein the recommendation information is used for representing the probability of the target user performing the operations of the plurality of operation types on the target article; and when the recommendation information meets a preset condition, determining to recommend the target article to the target user. By the method, the target user feature vector representing the favorite of the target user is generated based on the articles and the operation types with the association relation, and the target article feature vector representing the attraction feature of the target article to the user is generated based on the users and the operation types with the association relation, so that the probability of the target user performing operations of multiple operation types on the target article is predicted, and the operation probability of the user aiming at the articles can be accurately depicted.

Referring to fig. 19b, fig. 19b is a schematic diagram of a training sample construction testing method provided in the embodiment of the present application, where the method includes:

1901: the method comprises the steps of obtaining a first operation type of a first user on a first article, wherein article preference characteristics of the first user and a target user meet preset conditions.

The first operation type of the first user on the first article can be a real operation type of the first user on the first article.

the following historical operating records may be obtained:

to be provided with<User, type of operation, article>The triple represents a multi-behavior ternary heterogeneous graph network, the historical operation record can be simply represented as:<user 1, purchase, item 1>、<User 1, collection, item 2>、<User 1, collection, item 3>、<User 2, collection, item 2>、<User 2, click, item 3>、<User 3, click, item 3>。

Since the item 2 is collected by the user 2 and is collected by the user 1, new operation information for the user 2 can be generated: < user 2, purchase, item 1>, and so on the operation information: < user 3, purchase, item 1>.

Further, the user attribute information of the user may be:

to be provided with<User 4, buy, item 1>Indicate that by analogy, can obtain<User 5, purchase, item 1>And (4) data. Further, the item attribute of the item may be: item names, categories, labels, scores, comments, etc., e.g.

The same treatment can also be obtained

To be provided with<User 1, buy, item 4>And (4) showing.

Through the data mining process, the operation information originally only containing 6 ternary elements is expanded into 10 operation information.

For more description about step 1901, reference may be made to the description about the data extension policy in the foregoing embodiment, which is not described herein again.

1902. Generating a second operation type of the target user on the first article based on the first operation type of the first user on the first article; the second operation type is a predicted operation behavior of the target user for the first item.

For a description about the step 1902, reference may be made to the description about the data expansion policy in the foregoing embodiment, and details are not described herein again.

1903. And constructing a training sample according to the attribute information of the target user, the attribute information of the first article and the second operation type.

According to the method and the device, richer information can be found in limited historical operation data, and then more operation information related to the user can be generated, so that the data utilization rate of historical operation records can be improved, more training samples can be constructed, and the behavior of the user can be predicted more accurately based on the recommendation model trained by the training samples.

In one possible implementation, the preset condition includes at least one of:

In one possible implementation, the user attribute includes at least one of:

gender, age, occupation, income, hobbies, education level.

In one possible implementation, the item attribute includes at least one of:

item name, developer, installation package size, category, goodness.

browsing operation, joining shopping cart operation and purchasing operation.

In a possible implementation, the obtained operation information may be used to train a target recommendation model, specifically, attribute information of the target user and attribute information of the first item may be obtained; determining a target user characteristic vector according to the attribute information of the target user; determining a target article feature vector according to the attribute information of the first article; acquiring a third feature vector of the second operation type; and training a target recommendation model according to the target user feature vector, the target article feature vector and the third feature vector to obtain a trained target recommendation model. Determining a target user feature vector according to the attribute information of the target user; determining a target article feature vector according to the attribute information of the first article; the obtaining of the third feature vector of the second operation type may refer to the description in the foregoing embodiments, and the details of the similarity are not repeated here.

In a possible implementation, a prediction probability may be output through a target recommendation model according to the target user feature vector, the target item feature vector, and the third feature vector, where the prediction probability is used to represent a probability that the target user performs the operation of the second operation type on the first item, and according to the probability, a loss is determined, and the target recommendation model is updated according to the loss.

The embodiment of the application provides a user operation behavior prediction method, which comprises the following steps: acquiring a first operation type of a first user on a first article, wherein the article preference characteristics of the first user and a target user meet a preset condition; generating a second operation type of the target user on the first item based on the first operation type of the first user on the first item; the second operation type is a predicted operation behavior of the target user for the first item. The information training method has the advantages that richer information can be found in limited historical operation data, so that more operation information related to the user can be generated, the data utilization rate of historical operation records can be improved, and the behavior of the user can be predicted more accurately based on the recommendation model trained on the information.

Referring to fig. 20, fig. 20 is a flowchart illustrating a recommendation model training method provided in an embodiment of the present application, where the method includes:

2001. acquiring a first operation information sample set of a target sample user, wherein the first operation information sample set comprises attribute information of a plurality of articles, a plurality of operation types and corresponding relations between the articles and the operation types, and the corresponding relations are used for representing operation types of the target sample user on the articles;

for the description of step 2001, reference may be made to the description of step 401 in the above embodiment, which is not described herein again,

2002. performing feature extraction according to the first operation information sample set to determine a target sample user feature vector;

the description of step 2002 can refer to the description of step 402 in the above embodiment, which is not repeated here,

2003. acquiring a second operation information sample set of the target sample article, wherein the second operation information sample set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the plurality of users for operating the target article;

for the description of step 2003, reference may be made to the description of step 403 in the above embodiment, which is not repeated here,

2004. performing feature extraction according to the second operation information sample set to determine a feature vector of the target sample article;

the description of step 2004 may refer to the description of step 404 in the above embodiment, and will not be repeated here.

2005. Obtaining a sample label according to the actual operation type of the target sample user on the target article;

2006. and performing model training by taking the target sample user characteristic vector and the target sample article characteristic vector as input and the sample label as output to obtain a target recommendation model.

In one possible implementation, the determining a user feature vector of a target sample according to feature extraction performed on the first operation information sample set includes:

determining a plurality of subsample article feature vectors according to the second operation information sample set, wherein each second subsample article feature vector is obtained by performing feature extraction on attribute information and operation types of users with corresponding relations;

In one possible implementation, the obtaining the first operation information sample set includes:

acquiring a first operation information sample subset and a second operation information sample subset; wherein the content of the first and second substances,

In one possible implementation, the fusing the plurality of user subsample user feature vectors includes:

acquiring the second operation type;

the obtaining of the second operation type specifically includes:

In one possible implementation, the preset condition includes at least one of:

In one possible implementation, the operation type is a forward operation type.

the determining a plurality of subsample item feature vectors from the second sample set of operational information comprises:

acquiring the fifth operation type;

the obtaining the fifth operation type specifically includes:

In one possible implementation, the preset condition includes at least one of:

The embodiment of the application provides a recommendation model training method, which comprises the following steps: acquiring a first operation information sample set of a target sample user, wherein the first operation information sample set comprises attribute information of a plurality of articles, a plurality of operation types and corresponding relations between the articles and the operation types, and the corresponding relations are used for representing operation types of the target sample user on the articles; performing feature extraction according to the first operation information sample set to determine a target sample user feature vector; acquiring a second operation information sample set of the target sample article, wherein the second operation information sample set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the plurality of users for operating the target article; performing feature extraction according to the second operation information sample set to determine a feature vector of the target sample article; obtaining a sample label according to the actual operation type of the target sample user on the target article; and performing model training by taking the target sample user characteristic vector and the target sample article characteristic vector as input and the sample label as output to obtain a target recommendation model. Through the method, the target user feature vector representing the favorite of the target user is generated based on the article and the operation type with the association relation, and the target article feature vector representing the attraction feature of the target article to the user is generated based on the user and the operation type with the association relation, so that the probability of the target user performing operation of multiple operation types on the target article is predicted, and the operation probability of the user aiming at the article can be accurately depicted.

Referring to fig. 21, fig. 21 is a recommendation apparatus 2100 according to an embodiment of the present application, the apparatus including:

an obtaining module 2101, configured to obtain a first operation information set of a target user, where the first operation information set includes attribute information of a plurality of articles, a plurality of operation types, and a correspondence between the plurality of articles and the plurality of operation types, where the correspondence is used to indicate operation types of operations performed on the plurality of articles by the target user; acquiring a second operation information set of the target article, wherein the second operation information set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the operation performed on the target article by the plurality of users;

for a detailed description of the obtaining module 2101, reference may be made to the descriptions of step 401 and step 403, which are not described herein again.

A feature vector generation module 2102, configured to perform feature extraction according to the first operation information set to determine a target user feature vector; performing feature extraction according to the second operation information set to determine a feature vector of the target article;

for a detailed description of the feature vector generation module 2102, reference may be made to the descriptions of step 402 and step 404, which are not described herein again.

An information recommendation module 2103, configured to output recommendation information based on a target recommendation model according to the target user feature vector and the target item feature vector, where the recommendation information is used to indicate probabilities of the target user performing the operations of the multiple operation types on the target item; and when the recommendation information meets a preset condition, determining to recommend the target article to the target user.

For the detailed description of the information recommendation module 2103, reference may be made to the description of step 406, which is not described herein again.

In one possible implementation, the feature vector generation module 2102 is specifically configured to:

In a possible implementation, the information recommendation module 2103 is specifically configured to:

In a possible implementation, the obtaining module 2101 is specifically configured to:

the feature vector generation module 2102 is specifically configured to:

and determining a second sub-user feature vector according to the second operation information subset.

The obtaining module 2101 is further configured to:

acquiring the second operation type;

the obtaining of the second operation type specifically includes:

In one possible implementation, the preset condition includes at least one of:

In one possible implementation, the operation type is a forward operation type.

the feature vector generation module 2102 is specifically configured to:

determining a first sub-article feature vector according to the third operation information subset;

acquiring the fifth operation type;

the obtaining the fifth operation type specifically includes:

In one possible implementation, the preset condition includes at least one of:

the fourth user and the third user are users who have operations on items whose attribute differences are smaller than a threshold.

The application provides an information recommendation device, the device includes: an obtaining module, configured to obtain a first operation information set of a target user, where the first operation information set includes attribute information of a plurality of articles, a plurality of operation types, and correspondence between the plurality of articles and the plurality of operation types, and the correspondence is used to represent operation types of operations performed on the plurality of articles by the target user; acquiring a second operation information set of the target article, wherein the second operation information set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the operation performed on the target article by the plurality of users; the characteristic vector generation module is used for extracting characteristics according to the first operation information set to determine a target user characteristic vector; performing feature extraction according to the second operation information set to determine a target article feature vector; the information recommendation module is used for outputting recommendation information based on a target recommendation model according to the target user feature vector and the target article feature vector, wherein the recommendation information is used for representing the probability of the target user performing the operation of the plurality of operation types on the target article; and when the recommendation information meets a preset condition, determining to recommend the target article to the target user. According to the method and the device, the target user characteristic vector which represents the favorite of the target user is generated based on the article and the operation type with the association relation, and the second characteristic vector which represents the attraction characteristic of the target article to the user is generated based on the user and the operation type with the association relation, so that the probability of the target user performing operation of multiple operation types on the target article is predicted, and the operation probability of the user aiming at the article can be accurately depicted.

Referring to fig. 22, fig. 22 is a schematic structural diagram of a training sample constructing apparatus provided in an embodiment of the present application, where the apparatus 2200 includes:

an obtaining module 2201, configured to obtain a first operation type of a first article by a first user, where article preference characteristics of the first user and an object user meet a preset condition;

for a detailed description of the obtaining module 2201, reference may be made to the description of step 1901, which is not described herein again.

An operation information generating module 2202, configured to generate a second operation type of the target user on the first item based on the first operation type of the first user for the first item; the second operation type is a predicted operation behavior of the target user for the first item.

The detailed description of the operation information generation module 2202 may refer to the description of step 1902, and will not be described herein.

A sample construction module 2203, configured to construct a training sample according to the attribute information of the target user, the attribute information of the first item, and the second operation type.

For a detailed description of the sample construction module 2203, reference may be made to the description of step 1903, which is not described herein again.

In one possible implementation, the preset condition includes at least one of:

In one possible implementation, the user attribute includes at least one of:

gender, age, occupation, income, hobbies, education level.

In one possible implementation, the item attribute includes at least one of:

item name, developer, installation package size, category and goodness.

browsing operation, joining shopping cart operation and purchasing operation.

In a possible implementation, the operation information generating module 2202 is specifically configured to:

obtaining a second operation type based on the first operation type of the first user aiming at the first article;

Referring to fig. 23, fig. 23 is a schematic structural diagram of a recommended model training apparatus provided in an embodiment of the present application, where the apparatus 2300 may include:

an obtaining module 2301, configured to obtain a first operation information sample set of a target sample user, where the first operation information sample set includes attribute information of a plurality of articles, a plurality of operation types, and correspondence between the plurality of articles and the plurality of operation types, and the correspondence is used to represent operation types of operations performed on the plurality of articles by the target sample user; acquiring a second operation information sample set of a target sample article, wherein the second operation information sample set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the target article operated by the plurality of users;

for the detailed description of the obtaining module 2301, reference may be made to the description of step 2001 and step 2003, which are not described herein again.

A feature vector generation module 2302, configured to perform feature extraction according to the first operation information sample set to determine a target sample user feature vector; performing feature extraction according to the second operation information sample set to determine a feature vector of the target sample article;

for a detailed description of the feature vector generation module 2302, reference may be made to the description of step 2002 and step 2004, and details are not repeated here.

The obtaining module 2301 is further configured to obtain a sample label according to an actual operation type of the target object by the target sample user;

for a detailed description of the obtaining module 2301, reference may be made to the description of step 2005, which is not described herein again.

And the model training module 2303 is configured to perform model training by using the target sample user feature vector and the target sample article feature vector as inputs and the sample label as an output, and obtain a target recommendation model.

For a detailed description of the model training module 2303, reference may be made to the description of step 2006, which is not described herein again.

In one possible implementation, the feature vector generation module 2302 is specifically configured to:

In one possible implementation, the taking the target sample user feature vector and the target sample item feature vector as inputs includes:

and taking a plurality of operation type feature vectors of the plurality of operation types, the target sample user feature vector and the target sample article feature vector as input.

In a possible implementation, the obtaining module 2301 is specifically configured to:

the feature vector generation module 2302 is specifically configured to:

acquiring the second operation type;

the obtaining of the second operation type specifically includes:

In one possible implementation, the preset condition includes at least one of:

In one possible implementation, the operation type is a forward operation type.

the feature vector generation module 2302 is specifically configured to:

fusing the first sub-sample item feature vector and the second sub-sample item feature vector according to a third weight of the first sub-sample item feature vector and a fourth weight of the second sub-sample item feature vector. In a possible implementation, the obtaining module 2301 is specifically configured to:

acquiring the fifth operation type;

the obtaining the fifth operation type specifically includes:

In one possible implementation, the preset condition includes at least one of:

Referring to fig. 24, fig. 24 is a schematic structural diagram of an execution device provided in the embodiment of the present application, and the execution device 2400 may be embodied as a mobile phone, a tablet, a notebook computer, an intelligent wearable device, a server, and the like, which is not limited herein. The execution device 2400 may be deployed with the data processing apparatus described in the embodiment corresponding to fig. 10, and is configured to implement the function of data processing in the embodiment corresponding to fig. 10. Specifically, the execution device 2400 includes: a receiver 2401, a transmitter 2402, a processor 2403, and a memory 2404 (where the number of processors 2403 in the execution device 2400 may be one or more), where the processor 2403 may include an application processor 24031 and a communication processor 24032. In some embodiments of the application, the receiver 2401, the transmitter 2402, the processor 2403 and the memory 2404 may be connected by a bus or other means.

Memory 2404 may include both read-only memory and random access memory, and provides instructions and data to processor 2403. A portion of the memory 2404 may also include non-volatile random access memory (NVRAM). The memory 2404 stores processors and operational instructions, executable modules or data structures, or a subset thereof, or an expanded set thereof, wherein the operational instructions may include various operational instructions for performing various operations.

Processor 2403 controls the operation of the performing device. In a particular application, the various components of the execution device are coupled together by a bus system that may include a power bus, a control bus, a status signal bus, etc., in addition to a data bus. For clarity of illustration, the various buses are referred to in the figures as bus systems.

The method disclosed in the embodiments of the present application can be applied to the processor 2403 or implemented by the processor 2403. Processor 2403 may be an integrated circuit chip having signal processing capabilities. In implementation, the steps of the above method may be performed by instructions in the form of hardware, integrated logic circuits, or software in the processor 2403. The processor 2403 may be a general-purpose processor, a Digital Signal Processor (DSP), a microprocessor or a microcontroller, a Vision Processor (VPU), a Tensor Processing Unit (TPU), and other processors suitable for AI operation, and may further include an Application Specific Integrated Circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, and a discrete hardware component. The processor 2403 may implement or perform the methods, steps, and logic blocks disclosed in the embodiments of the present application. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in connection with the embodiments of the present application may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software module may be located in ram, flash memory, rom, prom, or eprom, registers, etc. storage media as is well known in the art. The storage medium is located in the memory 2404, and the processor 2403 reads the information in the memory 2404, and completes the steps 401 to 406, and the

steps

1901 and 1902 in the above embodiment in combination with the hardware thereof.

Receiver 2401 may be used to receive input numeric or character information and generate signal inputs related to performing device related settings and function controls. The transmitter 2402 may be used to output numeric or character information through the first interface; the transmitter 2402 may also be used to send instructions to the disk groups through the first interface to modify data in the disk groups; the transmitter 2402 may also include a display device such as a display screen.

Referring to fig. 25, fig. 25 is a schematic structural diagram of a training device provided in the embodiment of the present application, and specifically, the training device 2500 is implemented by one or more servers, and the training device 2500 may generate a relatively large difference due to different configurations or performances, and may include one or more Central Processing Units (CPUs) 2525 (e.g., one or more processors) and a memory 2532, and one or more storage media 2530 (e.g., one or more mass storage devices) storing an application 2542 or data 2544. Memory 2532 and storage media 2530 can be, among other things, transient or persistent storage. The program stored in storage medium 2530 may include one or more modules (not shown), each of which may include a sequence of instructions for operating on the exercise device. Still further, central processor 2525 may be arranged to communicate with storage medium 2530 to carry out a series of instruction operations on storage medium 2530 on training device 2500.

Training apparatus 2500 may also include one or more power supplies 2526, one or more wired or wireless network interfaces 2550, one or more input-output interfaces 2558; or one or more operating systems 2541, such as Windows Server, mac OS XTM, unixTM, linuxTM, freeBSDTM, etc.

Specifically, the training apparatus may perform the steps from step 2001 to step 2006 in the above embodiments.

Embodiments of the present application also provide a computer program product, which when executed on a computer causes the computer to perform the steps performed by the aforementioned execution device, or causes the computer to perform the steps performed by the aforementioned training device.

In an embodiment of the present application, a computer-readable storage medium is further provided, where a program for signal processing is stored, and when the program runs on a computer, the program causes the computer to execute the steps performed by the foregoing execution device, or causes the computer to execute the steps performed by the foregoing training device.

The execution device, the training device, or the terminal device provided in the embodiment of the present application may specifically be a chip, where the chip includes: a processing unit, which may be, for example, a processor, and a communication unit, which may be, for example, an input/output interface, a pin or a circuit, etc. The processing unit may execute the computer execution instructions stored by the storage unit to cause the chip in the execution device to execute the data processing method described in the above embodiment, or to cause the chip in the training device to execute the data processing method described in the above embodiment. Optionally, the storage unit is a storage unit in the chip, such as a register, a cache, and the like, and the storage unit may also be a storage unit located outside the chip in the radio access device, such as a read-only memory (ROM) or another type of static storage device that may store static information and instructions, a Random Access Memory (RAM), and the like.

Specifically, referring to fig. 26, fig. 26 is a schematic structural diagram of a chip provided in the embodiment of the present application, where the chip may be represented as a neural network processor NPU2600, and the NPU2600 is mounted on a Host CPU (Host CPU) as a coprocessor, and the Host CPU allocates tasks. The core portion of the NPU is an arithmetic circuit 2603, and the controller 2604 controls the arithmetic circuit 2603 to extract matrix data in the memory and perform multiplication.

The NPU2600 may implement the information recommendation method provided in the embodiment described in fig. 4, the training sample construction method provided in the embodiment described in fig. 19b, and the recommendation model training method provided in the embodiment described in fig. 20 through cooperation between internal devices.

More specifically, in some implementations, arithmetic circuit 2603 in NPU2600 includes multiple processing units (PEs) therein. In some implementations, the operational circuit 2603 is a two-dimensional systolic array. The arithmetic circuit 2603 may also be a one-dimensional systolic array or other electronic circuit capable of performing mathematical operations such as multiplication and addition. In some implementations, the arithmetic circuit 2603 is a general-purpose matrix processor.

For example, assume that there is an input matrix A, a weight matrix B, and an output matrix C. The arithmetic circuit fetches the data corresponding to the matrix B from the weight memory 2602 and buffers the data on each PE in the arithmetic circuit. The arithmetic circuit takes the matrix a data from the input memory 2601 and performs matrix operation with the matrix B, and partial or final results of the obtained matrix are stored in an accumulator (accumulator) 2608.

Unified memory 2606 is used to store input data as well as output data. The weight data is directly passed through a Memory Access Controller (DMAC) 2605, and the DMAC is transferred to a weight Memory 2602. The input data is also carried into the unified memory 2606 via the DMAC.

The BIU is a Bus Interface Unit (Bus Interface Unit) 2610, which is used for the interaction of the AXI Bus with the DMAC and an Instruction Fetch memory (IFB) 2609.

A Bus Interface Unit 2610 (Bus Interface Unit, BIU for short) is used for the instruction fetch memory 2609 to obtain instructions from the external memory, and is also used for the memory Unit access controller 2605 to obtain the original data of the input matrix a or the weight matrix B from the external memory.

The DMAC is mainly used to transfer input data in the external memory DDR to the unified memory 2606, to transfer weight data to the weight memory 2602, or to transfer input data to the input memory 2601.

The vector calculation unit 2607 includes a plurality of arithmetic processing units, and further processes the output of the arithmetic circuit 2603 such as vector multiplication, vector addition, exponential operation, logarithmic operation, magnitude comparison, and the like, if necessary. The method is mainly used for non-convolution/full-connection layer network calculation in the neural network, such as Batch Normalization, pixel-level summation, up-sampling of a feature plane and the like.

In some implementations, the vector calculation unit 2607 can store the processed output vector to the unified memory 2606. For example, the vector calculation unit 2607 may calculate a linear function; alternatively, a nonlinear function is applied to the output of the arithmetic circuit 2603, such as linear interpolation of the feature planes extracted from the convolutional layers, and then, such as a vector of accumulated values, to generate the activation values. In some implementations, the vector calculation unit 2607 generates normalized values, pixel-level summed values, or both. In some implementations, the vector of processed outputs can be used as activation inputs to the operational circuitry 2603, e.g., for use in subsequent layers in a neural network.

An instruction fetch buffer 2609 connected to the controller 2604 for storing instructions used by the controller 2604;

the unified memory 2606, the input memory 2601, the weight memory 2602, and the instruction fetch memory 2609 are all On-Chip memories. The external memory is private to the NPU hardware architecture.

The processor mentioned in any of the above may be a general-purpose central processing unit, a microprocessor, an ASIC, or one or more integrated circuits for controlling the execution of the program.

It should be noted that the above-described embodiments of the apparatus are merely schematic, where the units described as separate parts may or may not be physically separate, and the parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. In addition, in the drawings of the embodiments of the apparatus provided in the present application, the connection relationship between the modules indicates that there is a communication connection therebetween, which may be specifically implemented as one or more communication buses or signal lines.

Through the above description of the embodiments, those skilled in the art will clearly understand that the present application can be implemented by software plus necessary general-purpose hardware, and certainly can also be implemented by special-purpose hardware including special-purpose integrated circuits, special-purpose CPUs, special-purpose memories, special-purpose components and the like. Generally, functions performed by computer programs can be easily implemented by corresponding hardware, and specific hardware structures for implementing the same functions may be various, such as analog circuits, digital circuits, or dedicated circuits. However, for the present application, the implementation of a software program is more preferable. Based on such understanding, the technical solutions of the present application may be substantially embodied in the form of a software product, which is stored in a readable storage medium, such as a floppy disk, a usb disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk of a computer, and includes several instructions for enabling a computer device (which may be a personal computer, an exercise device, or a network device) to execute the method according to the embodiments of the present application.

In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, it may be implemented in whole or in part in the form of a computer program product.

The computer program product includes one or more computer instructions. When loaded and executed on a computer, cause the processes or functions described in accordance with the embodiments of the application to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, from one website site, computer, training device, or data center to another website site, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can store or a data storage device, such as a training device, a data center, etc., that incorporates one or more available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid State Disk (SSD)), among others.

Claims

1. A recommendation method, characterized in that the method comprises:

performing feature extraction according to the second operation information set to determine a feature vector of the target article;

according to the target user feature vector and the target article feature vector, outputting recommendation information based on a target recommendation model, wherein the recommendation information is used for representing the probability of the target user performing the operation of the plurality of operation types on the target article;

2. The method of claim 1, wherein the determining a target user feature vector by feature extraction according to the first operation information set comprises:

3. The method according to claim 1 or 2, wherein the determining a target item feature vector by feature extraction according to the second operation information set comprises:

4. The method of claim 2 or 3, wherein the obtaining the first set of operational information comprises: acquiring a first operation information subset and a second operation information subset; wherein, the first and the second end of the pipe are connected with each other,

the determining a plurality of sub-user feature vectors according to the first operation information set includes:

determining a first sub-user feature vector according to the first operation information subset, wherein the first sub-user feature vector is obtained by performing feature extraction on attribute information and operation types of the articles with corresponding relations in the first operation information subset;

determining a second sub-user feature vector according to the second operation information subset, wherein the second sub-user feature vector is obtained by performing feature extraction on attribute information and operation types of the articles with corresponding relations in the second operation information subset;

the fusing the feature vectors of the plurality of sub-users to obtain the feature vector of the target user comprises:

and fusing the first sub-user feature vector and the second sub-user feature vector to obtain the target user feature vector.

5. The method according to claim 4, wherein the fusing the first sub-user feature vector and the second sub-user feature vector comprises:

6. The method of claim 4 or 5, wherein the obtaining the second subset of operation information comprises:

acquiring the second operation type;

the obtaining of the second operation type specifically includes:

7. The method of claim 6, wherein the preset condition comprises at least one of:

8. The method according to claim 6 or 7, wherein the second operation type is obtained based on the third operation type of the first user for the second item, specifically, the second operation type is obtained according to the third operation type.

9. The method according to any of claims 3 to 8, wherein the obtaining the second operation information set comprises:

acquiring a third operation information subset and a fourth operation information subset; wherein, the first and the second end of the pipe are connected with each other,

the third operation information subset comprises attribute information of a second user, a fourth forward operation type and a corresponding relation between the second user and the fourth forward operation type; the fourth forward operation type is a real operation type of the second user on the target object;

the determining a plurality of sub-item feature vectors from the second set of operational information comprises:

determining a first sub-article feature vector according to the third operation information subset, wherein the first sub-article feature vector is obtained by feature extraction of attribute information and operation types of users with corresponding relations of the third operation information subset;

determining a second sub-article feature vector according to the fourth operation information subset, wherein the second sub-article feature vector is obtained by performing feature extraction on attribute information and operation types of users with corresponding relations of the fourth operation information subset;

the fusing the feature vectors of the sub-articles to obtain the feature vector of the target article comprises:

and fusing the first sub-article feature vector and the second sub-article feature vector to obtain the target article feature vector.

10. The method according to claim 9, wherein said fusing the first sub-item feature vector and the second sub-item feature vector comprises:

11. The method according to claim 9 or 10, wherein the obtaining a fourth subset of operation information comprises:

acquiring the fifth operation type;

the obtaining the fifth operation type specifically includes:

12. The method of claim 11, wherein the preset condition comprises at least one of:

13. The method according to claim 11 or 12, wherein the fifth operation type is obtained based on a sixth operation type of the target item by the fourth user, specifically, the fifth operation type is obtained according to the sixth operation type.

14. Method according to any of claims 1 to 13, wherein said operation type is a forward operation type.

15. A recommendation model training method, the method comprising:

acquiring a second operation information sample set of a target sample article, wherein the second operation information sample set comprises attribute information of a plurality of users, a plurality of operation types and corresponding relations between the plurality of users and the plurality of operation types, and the corresponding relations are used for representing operation types of the target article operated by the plurality of users;

obtaining a sample label according to the actual operation type of the target sample user on the target article;

16. The method of claim 15, wherein the determining a target sample user feature vector by feature extraction according to the first operation information sample set comprises:

17. The method according to claim 15 or 16, wherein the determining a target sample item feature vector by performing feature extraction according to the second operation information sample set comprises:

18. The method according to claim 16 or 17, wherein the obtaining the first set of operation information samples comprises:

the determining a plurality of sub-sample user feature vectors according to the first operation information sample set comprises:

determining a first sub-sample user feature vector according to the first operation information sample subset, wherein the first sub-sample user feature vector is obtained by performing feature extraction on attribute information and operation types of the articles with corresponding relations in the first operation information sample subset;

determining a second sub-sample user feature vector according to the second operation information sample subset, wherein the second sub-sample user feature vector is obtained by performing feature extraction on attribute information and operation types of the articles with corresponding relations in the second operation information sample subset;

the fusing the plurality of sub-sample user feature vectors to obtain the target sample user feature vector includes:

and fusing the first sub-sample user feature vector and the second sub-sample user feature vector to obtain the target sample user feature vector.

19. The method of claim 18, wherein the fusing the first sub-sample user feature vector and the second sub-sample user feature vector comprises:

20. The method according to claim 18 or 19, wherein the obtaining a second subset of operation information samples comprises:

acquiring the second operation type;

the obtaining of the second operation type specifically includes:

21. The method of claim 20, wherein the preset condition comprises at least one of:

22. The method according to claim 20 or 21, wherein the second operation type is obtained based on the third operation type of the first user for the second item, specifically, the second operation type is obtained according to the third operation type.

23. The method according to any of claims 17 to 22, wherein said obtaining the second set of operation information samples comprises:

determining a first sub-sample article feature vector according to the third operation information sample subset, wherein the first sub-sample article feature vector is obtained by performing feature extraction on attribute information and operation types of users with corresponding relations in the third operation information sample subset;

determining a second subsample article feature vector according to the fourth operation information sample subset, wherein the second subsample article feature vector is obtained by feature extraction of attribute information and operation types of users with corresponding relations in the fourth operation information sample subset;

the fusing the plurality of sub-sample article feature vectors to obtain the target sample article feature vector includes:

and fusing the first sub-sample article feature vector and the second sub-sample article feature vector to obtain the target sample article feature vector.

24. The method of claim 23, wherein fusing the first sub-sample item feature vector and the second sub-sample item feature vector comprises:

25. The method according to claim 23 or 24, wherein the obtaining a fourth subset of operation information comprises:

acquiring the fifth operation type;

the obtaining the fifth operation type specifically includes:

26. The method of claim 25, wherein the preset condition comprises at least one of:

27. The method according to claim 25 or 26, wherein the fifth operation type is obtained based on a sixth operation type of the target item by the fourth user, specifically, the fifth operation type is obtained according to the sixth operation type.

28. The method according to any of claims 15 to 27, wherein the operation type is a forward operation type.

29. A training sample construction method, the method comprising:

generating a second operation type of the target user on the first item based on the first operation type of the first user on the first item; the second operation type is a predicted operation behavior of the target user for the first item;

30. The method of claim 29, wherein the preset condition comprises at least one of:

31. The method of claim 30, wherein the user attribute comprises at least one of:

gender, age, occupation, income, hobbies, education level.

32. The method of claim 30 or 31, wherein the item attribute comprises at least one of:

item name, developer, installation package size, category, goodness.

33. The method according to any one of claims 30 to 32, wherein generating a second type of operation of the target user on the first item based on the first type of operation of the first user on the first item comprises: obtaining a second operation type based on the first operation type of the first user for the first article;

34. The method of any of claims 29 to 33, wherein the first type of operation and the second type of operation comprise at least one of:

browsing operation, joining shopping cart operation and purchasing operation.

35. A computing device, wherein the computing device comprises a memory and a processor; the memory stores code, and the processor is configured to retrieve the code and perform the method of any of claims 1 to 34.

36. A computer storage medium, characterized in that the computer storage medium stores one or more instructions that, when executed by one or more computers, cause the one or more computers to implement the method of any of claims 1 to 34.

37. A computer program product comprising code for implementing a method as claimed in any one of claims 1 to 34 when executed.