CN115203516A

CN115203516A - Information recommendation method, device, equipment and storage medium based on artificial intelligence

Info

Publication number: CN115203516A
Application number: CN202110375387.9A
Authority: CN
Inventors: 王新民; 胡伟龙; 谭莲芝; 袁镱
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2021-04-08
Filing date: 2021-04-08
Publication date: 2022-10-18

Abstract

The application provides an information recommendation method and device based on artificial intelligence, electronic equipment and a computer readable storage medium. The method comprises the following steps: the method comprises the steps of obtaining a plurality of recommendation characteristics of a task to be recommended, wherein the recommendation characteristics comprise at least one article characteristic of information to be recommended and at least one user characteristic of a target user; performing feature projection processing on the plurality of recommended features based on the projection space corresponding to each recommended feature to obtain projection features of the projection space; performing feature interaction processing based on each recommended feature and the projection feature of the corresponding projection space to obtain a feature domain corresponding to each recommended feature; performing index prediction processing based on the feature domains corresponding to the plurality of recommended features to obtain recommended indexes of target users corresponding to the information to be recommended; and executing the task to be recommended based on the recommendation index of the target user corresponding to the information to be recommended. According to the method and the device, the accuracy of content or information recommendation can be improved.

Description

Information recommendation method, device, equipment and storage medium based on artificial intelligence

Technical Field

The present disclosure relates to artificial intelligence technologies, and in particular, to an information recommendation method and apparatus based on artificial intelligence, an electronic device, and a computer-readable storage medium.

Background

Artificial Intelligence (AI) is a comprehensive technique in computer science, and by studying the design principles and implementation methods of various intelligent machines, the machines have the functions of perception, reasoning and decision making. The artificial intelligence technology is a comprehensive subject and relates to a wide range of fields, for example, natural language processing technology and machine learning/deep learning, etc., and along with the development of the technology, the artificial intelligence technology can be applied in more fields and can play more and more important values.

Recommendation systems are one of the important applications in the field of artificial intelligence, and can help users find information that may be of interest to them in an information overload environment and push the information to the users who are interested in them.

Although, the recommendation system in the related art may determine information that may be interested by the user from a large amount of information to be recommended based on the recommendation index of the information to be recommended, and recommend the information that may be interested by the user to the user. However, the accuracy of the recommendation index of the information to be recommended in the related art needs to be improved.

Disclosure of Invention

The embodiment of the application provides an information recommendation method and device based on artificial intelligence, an electronic device and a computer-readable storage medium, and the accuracy of information recommendation can be improved.

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

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

the method comprises the steps of obtaining a plurality of recommended features of a task to be recommended, wherein the recommended features comprise at least one item feature of information to be recommended and at least one user feature of a target user;

performing feature projection processing on the plurality of recommended features based on the projection space corresponding to each recommended feature to obtain projection features of the projection space;

performing feature interaction processing based on each recommended feature and the projection feature corresponding to the projection space to obtain a feature domain corresponding to each recommended feature;

performing index prediction processing based on the feature domains corresponding to the plurality of recommended features to obtain recommended indexes of the target user corresponding to the information to be recommended;

and executing the task to be recommended based on the recommendation index of the target user corresponding to the information to be recommended.

In the above technical solution, the index prediction processing is realized by a first prediction model and a second prediction model;

the index prediction processing is performed based on the feature domains corresponding to the multiple recommended features to obtain the recommended index of the target user corresponding to the information to be recommended, and the index prediction processing includes:

index prediction processing is carried out on the feature domains corresponding to the recommended features through the first prediction model, and the recommended indexes of the first prediction model are obtained;

index prediction processing is carried out on the plurality of recommended features of the task to be recommended through the second prediction model, and recommendation indexes of the second prediction model are obtained;

and obtaining the recommendation index of the information to be recommended corresponding to the target user based on the recommendation index of the first prediction model and the recommendation index of the second prediction model.

In the above technical solution, the obtaining of the recommendation index of the information to be recommended corresponding to the target user based on the recommendation index of the first prediction model and the recommendation index of the second prediction model includes:

adding the recommendation index of the first prediction model and the recommendation index of the second prediction model to obtain a summed recommendation index;

and carrying out normalization processing on the added recommendation index to obtain the recommendation index of the target user corresponding to the information to be recommended.

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

the system comprises an acquisition module, a recommendation module and a recommendation module, wherein the acquisition module is used for acquiring a plurality of recommendation characteristics of a task to be recommended, and the recommendation characteristics comprise at least one article characteristic of information to be recommended and at least one user characteristic of a target user;

the feature interaction module is used for performing feature projection processing on the recommended features based on the projection space corresponding to each recommended feature to obtain projection features of the projection space;

the prediction module is used for performing index prediction processing based on the feature domains corresponding to the plurality of recommended features to obtain the recommendation index of the target user corresponding to the information to be recommended;

and the recommending module is used for executing the tasks to be recommended based on the recommending indexes of the information to be recommended, which correspond to the target users.

In the above technical solution, the feature interaction module is further configured to perform nonlinear feature extraction processing on the plurality of recommended features based on a weight of a projection space corresponding to each recommended feature, so as to obtain extracted features of the projection space;

and multiplying the extracted features of the projection space by the projection matrix of the projection space to obtain the projection features of the projection space.

In the above technical solution, the feature interaction module is further configured to scale the plurality of recommended features based on a weight of a projection space corresponding to each recommended feature, so as to obtain the plurality of recommended features after scaling;

and splicing the plurality of zoomed recommended features to obtain the extracted features of the projection space.

In the above technical solution, the feature interaction module is further configured to multiply each recommended feature and the projection feature corresponding to the projection space to obtain a feature vector of each recommended feature;

and mapping the feature vector of each recommended feature to obtain a feature domain corresponding to each recommended feature.

In the above technical solution, the projection space corresponding to each of the recommended features includes a plurality of channels; the feature interaction module is further configured to perform feature projection processing on the plurality of recommended features based on any one of the channels of the projection space corresponding to each recommended feature to obtain a projection feature of any one of the channels;

performing feature interaction processing on each recommended feature and the projection feature corresponding to any channel to obtain a feature domain corresponding to any channel;

and fusing the feature domains respectively corresponding to the plurality of channels of the projection space to obtain the feature domain corresponding to each recommended feature.

In the above technical solution, the feature interaction module is further configured to sum feature domains corresponding to the multiple channels of the projection space, respectively, and use a result of the sum processing as a feature domain corresponding to each recommended feature; or,

and averaging the feature domains respectively corresponding to the multiple channels of the projection space, and taking the result of the averaging as the feature domain corresponding to each recommended feature.

In the above technical solution, the feature interaction network for feature interaction includes a plurality of cascaded feature interaction layers; the feature interaction module is further configured to perform, by any one of the feature interaction layers in the feature interaction network:

performing feature projection processing on the plurality of recommended features input to any one of the feature interaction layers based on the projection space corresponding to each recommended feature to obtain the projection features of any one of the feature interaction layers;

the plurality of recommended features input into the first feature interaction layer are the plurality of recommended features included in the task to be recommended, the plurality of recommended features input into the subsequent feature interaction layer are the plurality of recommended features output by the last feature interaction layer of the subsequent feature interaction layer, and the subsequent feature interaction layer is a feature interaction layer except the first feature interaction layer in the plurality of cascaded feature interaction layers;

performing, by any of the feature interaction layers in the feature interaction network:

performing feature interaction processing based on the recommended features input to any one of the feature interaction layers and the projection features corresponding to any one of the feature interaction layers to obtain a feature domain of any one of the feature interaction layers, and outputting the recommended features included in the feature domain;

and determining the feature domain of the last feature interaction layer as the feature domain corresponding to each recommended feature.

In the above technical solution, the feature interaction module is further configured to perform feature interaction processing based on the recommended features input to any one of the feature interaction layers and the projection features corresponding to any one of the feature interaction layers, so as to obtain hidden features of any one of the feature interaction layers;

and adding the hidden features of any one feature interaction layer and the recommended features input into any one feature interaction layer to obtain a feature domain of any one feature interaction layer.

In the above technical solution, the index prediction processing is realized by a first prediction model and a second prediction model; the prediction module is further used for performing index prediction processing on the feature domains corresponding to the recommended features through the first prediction model to obtain recommended indexes of the first prediction model;

In the above technical solution, the prediction module is further configured to perform a splicing process on the feature domains corresponding to the recommended features through the first prediction model to obtain spliced feature domains;

weighting the spliced feature domain based on the weight of the first prediction model to obtain a weighted feature domain;

and adding the bias of the first prediction model and the weighted feature domain to obtain the recommendation index of the first prediction model.

In the above technical solution, the prediction module is further configured to sum the recommendation index of the first prediction model and the recommendation index of the second prediction model to obtain a summed recommendation index;

In the above technical solution, the second prediction model includes a plurality of cascaded hidden layers; the prediction module is further configured to map, through a first hidden layer of the plurality of cascaded hidden layers, a plurality of recommended features of the task to be recommended;

outputting the mapping result of the first hidden layer to a hidden layer of a subsequent cascade, so as to continue mapping processing and outputting the mapping result in the hidden layer of the subsequent cascade until the mapping result is output to a last hidden layer;

and taking the mapping result output by the last hidden layer as a recommendation index of the second prediction model.

In the foregoing technical solution, the prediction module is further configured to execute the following processing by a jth hidden layer of the multiple cascaded hidden layers:

weighting the mapping result of the j-1 th hidden layer based on the weight of the j-th hidden layer to obtain a weighted mapping result;

adding the bias of the jth hidden layer and the weighted mapping result to obtain a mapping result of the jth hidden layer, and outputting the mapping result of the jth hidden layer to a (j + 1) th hidden layer;

wherein j is an increasing natural number and the value range of j is more than or equal to 2 and less than or equal to N-1, and N is the number of the cascaded hidden layers.

An embodiment of the present application provides an electronic device for information recommendation, where the electronic device includes:

a memory for storing executable instructions;

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

The embodiment of the application provides a computer-readable storage medium, which stores executable instructions for causing a processor to execute the computer-readable storage medium to implement the artificial intelligence-based information recommendation method provided by the embodiment of the application.

The embodiment of the application provides a computer program, which is used for causing a processor to execute, so as to realize the artificial intelligence based information recommendation method provided by the embodiment of the application.

The embodiment of the application has the following beneficial effects:

the method comprises the steps of performing projection processing on a plurality of recommended features in different projection spaces, and performing feature interaction on the projection features in each projection space, so that the diversity of feature domains is improved; index prediction is carried out based on various characteristic domains, so that the accuracy of recommendation indexes is improved, and the accuracy of recommendation is improved.

Drawings

Fig. 1 is a schematic view of an application scenario of a recommendation system provided in an embodiment of the present application;

FIG. 2 is a schematic structural diagram of an electronic device for information recommendation provided in an embodiment of the present application;

3-5 are flow diagrams of artificial intelligence based information recommendation methods provided by embodiments of the present application;

FIG. 6 is a schematic structural diagram of a feature interaction network provided in an embodiment of the present application;

FIG. 7 is a diagram illustrating a multi-channel structure of a feature interaction network provided by an embodiment of the present application;

FIG. 8 is a schematic structural diagram of a feature interaction layer provided in an embodiment of the present application;

FIG. 9 is a flowchart illustrating an artificial intelligence based information recommendation method according to an embodiment of the present application;

fig. 10 is a schematic structural diagram of a feature domain aware interaction module provided in an embodiment of the present application.

Detailed Description

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

In the following description, references to the terms "first", "second", and the like are only intended to distinguish similar objects and do not denote a particular order, but rather the terms "first", "second", and the like may be used interchangeably with the specific order or sequence described herein, where permissible, to enable embodiments of the present application to be practiced otherwise than as specifically illustrated or described herein.

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

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

1) The target user: the user currently using the recommendation system, i.e. the current user, for example, user a, is watching news using the text recommendation system, and user a is the target user.

2) User portrait: the method is also called as a user role, and is an effective tool for delineating target users and connecting user appeal and design direction. User images are widely used in various fields, and in the course of actual operations, attributes and behaviors of users are often combined with expectations by words appearing shallowest and living closely to each other to serve as virtual representations of actual users.

3) Recommendation indexes are as follows: the index is used for guiding the recommendation system to recommend, for example, whether the target user clicks the information to be recommended, whether the target user is interested in the information to be recommended, whether the target user converts the information to be recommended based on the information to be recommended, whether the target user evaluates the information to be recommended, and the like.

4) The task to be recommended is as follows: tasks corresponding to the recommendation indexes, that is, tasks that the target user needs to execute on the information to be recommended based on the recommendation indexes, for example, the target user determines whether the information to be recommended needs to be recommended based on the recommendation indexes, the target user determines whether the recommended position of the information to be recommended needs to be adjusted based on the recommendation indexes, and the like.

5) Click-through rate prediction (click-through rate prediction): the method is used for predicting the click condition of each piece of information to be recommended according to given information to be recommended (such as advertisements), users, context conditions and the like.

6) Factorization Machines (FM, factorization Machines): the training complexity is reduced by introducing a hidden vector mode, the method is a machine learning algorithm based on matrix decomposition, and the method has good learning capability on sparse data.

7) Deep Neural networks (DNN, deep Neural Network): the method is a feedforward neural network with a deep structure, is a technology in the field of Machine Learning (ML), and can represent complex functions by using fewer parameters.

The embodiment of the application provides an information recommendation method and device based on artificial intelligence, electronic equipment and a computer readable storage medium, and the recommendation accuracy can be improved.

The information recommendation method based on artificial intelligence provided by the embodiment of the application can be independently realized by a terminal/a server; the information recommendation method based on artificial intelligence can be executed by the server according to the received information recommendation request aiming at the target user, the recommendation index of the information to be recommended corresponding to the target user is determined, the task to be recommended is executed according to the recommendation index of the information to be recommended corresponding to the target user, and the information recommendation request aiming at the target user is responded.

The electronic device for information recommendation provided by the embodiment of the application can be various types of terminal devices or servers, wherein the server can be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as cloud service, a cloud database, cloud computing, cloud functions, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN (content distribution network), big data and an artificial intelligence platform; the terminal may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, a smart car coupler, etc., but is not limited thereto. The terminal and the server may be directly or indirectly connected through wired or wireless communication, and the application is not limited herein.

Taking a server as an example, the server may be, for example, a server cluster deployed in a cloud, and open an artificial intelligence cloud Service (AI as a Service, AIaaS) to users, the AIaaS platform may split several types of common AI services, and provide an independent or packaged Service in the cloud, this Service mode is similar to an AI theme mall, and all users may access one or more artificial intelligence services provided by the AIaaS platform by using an application programming interface.

For example, one of the artificial intelligence cloud services may be an information recommendation service, that is, a cloud server encapsulates an information recommendation program provided in the embodiments of the present application. A user calls an information recommendation service in a cloud service through a terminal (a client is operated, such as a news client, a video client and the like), so that a server deployed at the cloud calls a packaged information recommendation program, feature projection is carried out on the basis of a projection space corresponding to each recommendation feature of a task to be recommended, feature interaction is carried out on the basis of each recommendation feature and the projection feature (namely projection representation) of the corresponding projection space, prediction is carried out on the basis of feature domains (namely feature representation of feature domain perception) corresponding to a plurality of recommendation features, the task to be recommended is executed on the basis of recommendation indexes, and therefore the information to be recommended is distributed to users meeting interest requirements, user behavior data are obtained quickly, and the information recommendation effect is improved.

Referring to fig. 1, fig. 1 is a schematic view of an application scenario of a recommendation system 10 provided in an embodiment of the present application, where terminals (exemplary shown are a terminal 200-1, a terminal 200-2, and a terminal 200-3) are connected to a server 100 through a network 300, and the network 300 may be a wide area network or a local area network, or a combination of the two.

The terminal (running a client, such as a news client, a video client, etc.) may be used to obtain an information recommendation request for the target user, for example, when the target user opens the video client running on the terminal, the terminal automatically obtains a video recommendation request for the target user.

In some embodiments, after the terminal obtains the information recommendation request for the target user, an information recommendation interface (which may be provided in a cloud service form, that is, an information recommendation service) of the server 100 is called, the server 100 performs feature projection based on the projection space corresponding to each recommendation feature of the task to be recommended based on the information recommendation request for the target user, performs feature interaction based on each recommendation feature and the projection feature of the corresponding projection space, performs prediction based on feature domains corresponding to a plurality of recommendation features, executes the task to be recommended based on the recommendation index, distributes the information to be recommended to users meeting interest requirements, quickly obtains user behavior data, and improves the information recommendation effect to respond to the information recommendation request for the target user.

As an application example, for a video application, after a plurality of target users open a video client running on a terminal, the terminal automatically obtains a video recommendation request for the target users, invokes an information recommendation interface of the server 100, the server 100 executes an artificial intelligence-based information recommendation method based on a plurality of recommendation characteristics of a task to be recommended, determines recommendation indexes (e.g., whether to click) of the target users corresponding to the video to be recommended respectively, estimates a click rate of the video to be recommended based on the recommendation indexes of the target users corresponding to the video to be recommended respectively, estimates the click rate of the video to be recommended, for example, if 80 target users are likely to click the video to be recommended among the 100 target users, estimates the click rate of the video to be recommended to be 80%, adjusts a recommendation position of the video to be recommended to enable the video to be recommended to be located at a position with a better exposure, thereby quickly obtaining user behavior data and improving an information recommendation effect to respond to the information recommendation request for the target users.

In some embodiments, an information recommendation plug-in may be implanted in a client running in the terminal, so as to implement the artificial intelligence based information recommendation method locally at the client. For example, after the terminal obtains an information recommendation request for a target user, the terminal calls an information recommendation plug-in to realize an artificial intelligence-based information recommendation method, performs feature projection based on a projection space corresponding to each recommendation feature of a task to be recommended, performs feature interaction based on each recommendation feature and the projection feature of the corresponding projection space, performs prediction based on feature domains corresponding to a plurality of recommendation features, and executes the task to be recommended based on recommendation indexes to respond to the information recommendation request for the target user.

As an application example, for a news application, after a target user opens a news client running on a terminal, the terminal automatically acquires a news recommendation request for the target user, calls an information recommendation plug-in, executes an artificial intelligence-based information recommendation method based on a plurality of recommendation features of a task to be recommended, determines a recommendation index (e.g., whether to click) of the target user corresponding to the news to be recommended, and executes the task to be recommended when the recommendation index of the target user corresponding to the news to be recommended represents that the target user clicks the news to be recommended, so that user behavior data is quickly acquired, and an information recommendation effect is improved, so as to respond to the information recommendation request for the target user.

The structure of the electronic device for information recommendation provided in the embodiment of the present application is described below, referring to fig. 2, fig. 2 is a schematic structural diagram of the electronic device 500 for information recommendation provided in the embodiment of the present application, and taking the electronic device 500 as an example of a server as an illustration, the electronic device 500 for information recommendation shown in fig. 2 includes: at least one processor 510, memory 550, at least one network interface 520, and a user interface 530. The various components in the electronic device 500 are coupled together by a bus system 540. It is understood that the bus system 540 is used to enable communications among the components. The bus system 540 includes a power bus, a control bus, and a status signal bus in addition to a data bus. For clarity of illustration, however, the various buses are labeled as bus system 540 in fig. 2.

The Processor 510 may be an integrated circuit chip having Signal processing capabilities, such as a general purpose Processor, a Digital Signal Processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like, wherein the general purpose Processor may be a microprocessor or any conventional Processor, or the like.

The memory 550 may comprise volatile memory or nonvolatile memory, and may also comprise both volatile and nonvolatile memory. The non-volatile Memory may be a Read Only Memory (ROM), and the volatile Memory may be a Random Access Memory (RAM). The memory 550 described in embodiments herein is intended to comprise any suitable type of memory. Memory 550 optionally includes one or more storage devices physically located remote from processor 510.

In some embodiments, memory 550 can store data to support various operations, examples of which include programs, modules, and data structures, or subsets or supersets thereof, as exemplified below.

An operating system 551 including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;

a network communication module 552 for communicating to other computing devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 including: bluetooth, wireless compatibility authentication (WiFi), and Universal Serial Bus (USB), etc.;

in some embodiments, the artificial intelligence-based information recommendation apparatus provided in the embodiments of the present application may be implemented in a software manner, for example, the apparatus may be the information recommendation service in the server described above, or may be the information recommendation plug-in the terminal described above. Of course, without limitation, the artificial intelligence based information recommendation apparatus provided in the embodiments of the present application may be provided in various software embodiments, including various forms of applications, software modules, scripts or code.

FIG. 2 shows an artificial intelligence based information recommendation apparatus 555 stored in memory 550, which may be software in the form of programs and plug-ins, such as an information recommendation plug-in, and includes a series of modules, including an acquisition module 5551, a feature interaction module 5552, a prediction module 5553, and a recommendation module 5554; the obtaining module 5551, the feature interaction module 5552, the prediction module 5553, and the recommendation module 5554 are configured to implement the information recommendation function provided in the embodiment of the present application.

As described above, the artificial intelligence based information recommendation method provided by the embodiment of the present application can be implemented by various types of electronic devices. Referring to fig. 3, fig. 3 is a schematic flowchart of an artificial intelligence based information recommendation method provided in an embodiment of the present application, and is described with reference to the steps shown in fig. 3.

In the following steps, the information to be recommended may be text, image, graphics, video, etc. The recommendation index is used for guiding the recommendation system to perform recommendation, for example, whether the target user clicks the information to be recommended, whether the target user is interested in the information to be recommended, whether the target user performs conversion based on the information to be recommended, whether the target user evaluates the information to be recommended, and the like.

In step 101, a plurality of recommendation features of a task to be recommended are obtained, wherein the plurality of recommendation features include at least one item feature of information to be recommended and at least one user feature of a target user.

For example, feature extraction processing is performed on the information to be recommended by an offline means or an online means to obtain article features of the information to be recommended, for example, the article features are used for representing features of the information to be recommended, for example, the article features include categories, prices, and the like of the information to be recommended.

For example, a user portrait of the target user is subjected to feature extraction processing through an offline means or an online means, so as to obtain user features of the target user, for example, the user features are used for characterizing features of the target user, for example, the user features include age, preference, gender, and the like of the target user.

For example, after the recommendation features of the task to be recommended (the item features of the information to be recommended and the user features of the target user) are extracted in advance by an offline means, and after the target user opens the client running on the terminal, the terminal automatically acquires an information recommendation request (including the identifier of the target user) for the target user, and acquires the recommendation features of the task to be recommended extracted in advance based on the information recommendation request for the target user, so that feature interaction operation is performed subsequently based on the recommendation features of the task to be recommended.

In step 102, feature projection processing is performed on the plurality of recommended features based on the projection space corresponding to each recommended feature, so as to obtain projection features of the projection space.

For example, after the recommended features of the task to be recommended are obtained, feature projection processing is performed on the plurality of recommended features through a feature interaction network for the projection space corresponding to each recommended feature to obtain the projection features (i.e., projection features) of each projection space, for example, feature projection is performed on all the recommended features for the projection space i corresponding to the recommended features i to obtain the projection features of all the recommended features in the projection space i, i.e., the projection features of the projection space i.

In some embodiments, performing feature projection processing on a plurality of recommended features based on a projection space corresponding to each recommended feature to obtain projection features of the projection space includes: performing nonlinear feature extraction processing on the plurality of recommended features based on the weight of the projection space corresponding to each recommended feature to obtain extracted features of the projection space; and multiplying the extracted features of the projection space by the projection matrix of the projection space to obtain the projection features of the projection space.

As shown in fig. 6, the nonlinear feature extraction may be implemented by: and based on the weight of the projection space corresponding to each recommendation feature, carrying out scaling processing on the recommendation features to obtain a plurality of scaled recommendation features, and carrying out splicing processing on the plurality of scaled recommendation features to obtain the extraction features of the projection space.

As shown in FIG. 6, the formula for the feature projection is

Wherein s is _j Representing recommended features j, w _ij Representing the weight, U, of the corresponding recommended feature j in the projection space i _i A projection matrix representing a projection space i, S represents a set of recommended features, K represents the number of recommended features, z _i Representing projection characteristics of the projection space i, the characteristic projection in the embodiment of the present application is not limited to

Other projected deformation formulas are also possible.

In step 103, feature interaction processing is performed based on each recommended feature and the projection feature of the corresponding projection space, so as to obtain a feature domain corresponding to each recommended feature.

For example, in obtaining a projection feature z of a projection space i _i Then, to the projection characteristic z _i And a recommendation feature s _i Performing feature interaction to obtain a feature domain s 'corresponding to the recommended feature i' _i (i.e., feature domain aware feature representation) to obtain a representation of each recommended featureFeature domain s' ₁ ，’ ₂ ，…，s’ _K Therefore, projection processing of a plurality of recommendation features is carried out in different projection spaces, feature interaction is carried out on the projection features in each projection space, and therefore diversity of feature domains is improved, and accuracy of recommendation indexes is improved in the following.

In some embodiments, the obtaining a feature domain corresponding to each recommended feature based on feature interaction processing performed on each recommended feature and the projection feature of the corresponding projection space includes: multiplying each recommended feature and the projection feature of the corresponding projection space to obtain a feature vector of each recommended feature; and mapping the feature vector of each recommended feature to obtain a feature domain corresponding to each recommended feature.

As shown in FIG. 6, the formula of feature interaction processing is s' _i ＝ReLU(s _i ·z _i ) Wherein s is _i Representing recommended features i, z _i Represents the projection feature of the projection space i, and ReLU represents a mapping process, s' _i The reference character denotes a feature domain corresponding to the recommended feature i, and the embodiments of the present application are not limited to s' _i ＝ReLU(s _i ·z _i ) Other feature interaction deformation formulas are also applicable.

Referring to fig. 4, fig. 4 is an alternative flowchart of an artificial intelligence based information recommendation method provided in an embodiment of the present application, and fig. 4 shows that step 102 of fig. 3 can also be implemented by step 1021A: the projection space corresponding to each recommended feature comprises a plurality of channels; in step 1021A, performing feature projection processing on the plurality of recommended features based on any channel of the projection space corresponding to each recommended feature to obtain a projection feature of any channel; FIG. 4 shows that step 103 of FIG. 3 can also be implemented by steps 1031A-1032A: in step 1031A, feature interaction processing is performed on each recommended feature and the projection feature corresponding to any one of the channels to obtain a feature domain corresponding to any one of the channels; in step 1032A, feature domains corresponding to the multiple channels of the projection space are fused to obtain a feature domain corresponding to each recommended feature.

See FIG. 7 for projection spacei, the formula of the feature projection is

Wherein s is _j Representing recommended features j, W _ijh Representing the weight, U, of the corresponding recommended feature j in the channel h of the projection space i _i A projection matrix representing a channel h of a projection space i, S represents a set of recommended features, K represents the number of recommended features, z _ih The projection characteristics of the channel h representing the projection space i, the characteristic projection in the embodiment of the present application is not limited to

Other projected deformation formulas are also possible.

Referring to FIG. 7, the formula of feature interaction processing is s' _ih ＝ReLU(s _i ·z _ih ) Wherein s is _i Representing recommended features i, z _ih Represents the projection characteristics of channel h of projection space i, and ReLU represents a mapping process, s' _ih The feature domain corresponding to the channel h representing the recommended feature i is not limited to s' _ih ＝ReLU(s _i ·z _ih ) Other feature interaction deformation formulas are also applicable.

For example, the formula for fusion is

Wherein, H represents the number of channels,

representing any combination function, such as element-by-element summation or element-by-element averaging, that is, summing up feature domains corresponding to a plurality of channels of the projection space, and taking the result of the summing up as a feature domain corresponding to each recommended feature; or averaging the feature domains corresponding to the multiple channels of the projection space, and taking the result of the averaging as the feature domain corresponding to each recommended feature.

Referring to fig. 5, fig. 5 is an optional flowchart of the artificial intelligence based information recommendation method according to the embodiment of the present application, and fig. 5 shows that step 102 of fig. 3 can also be implemented by step 1021B: the feature interaction network for feature interaction comprises a plurality of cascaded feature interaction layers; in step 1021B, the following processing is performed by any feature interaction layer in the feature interaction network: performing feature projection on a plurality of recommended features input to any feature interaction layer based on a projection space corresponding to each recommended feature to obtain projection features of any feature interaction layer; fig. 5 shows that step 103 of fig. 3 can also be implemented by steps 1031B-1032B: performing, by any feature interaction layer in the feature interaction network: performing feature interaction processing based on a plurality of recommended features input to any feature interaction layer and projection features corresponding to any feature interaction layer to obtain a feature domain of any feature interaction layer, and outputting a plurality of recommended features included in the feature domain; in step 1032B, the feature domain of the last feature interaction layer is determined as the feature domain corresponding to each recommended feature.

The plurality of recommended features input into the first feature interaction layer are a plurality of recommended features included in the task to be recommended, the plurality of recommended features input into the subsequent feature interaction layer are a plurality of recommended features output by a last feature interaction layer of the subsequent feature interaction layer, and the subsequent feature interaction layer is a feature interaction layer except the first feature interaction layer in the plurality of cascaded feature interaction layers.

Referring to fig. 8, the feature interaction network includes L cascaded feature interaction layers, performs, for an L-th feature interaction layer, feature projection on a plurality of recommended features input to the L-th feature interaction layer based on a projection space corresponding to each recommended feature to obtain a projection feature of the L-th feature interaction layer, performs feature interaction processing based on the plurality of recommended features input to the L-th feature interaction layer and the projection feature of the L-th feature interaction layer to obtain a feature domain of the L-th feature interaction layer, and outputs a plurality of recommended features included in the feature domain (i.e., the feature domain of the L-th feature interaction layer is used as the recommended feature input to the L + 1-th feature interaction layer by the L-th feature interaction layer), and determines the feature domain of the last feature interaction layer as the feature domain corresponding to each recommended feature.

For example, the feature interaction network is processed as

Wherein, FIL-Layer represents the processing procedure of the l characteristic interaction Layer,

represents recommended features input to the ith feature interaction layer,

the feature domain of the ith feature interaction layer is represented.

In some embodiments, performing feature interaction processing based on a plurality of recommended features input to any one of the feature interaction layers and the projection feature corresponding to any one of the feature interaction layers to obtain a feature domain of any one of the feature interaction layers, includes: performing feature interaction processing based on a plurality of recommended features input to any feature interaction layer and projection features corresponding to any feature interaction layer to obtain hidden features of any feature interaction layer; and adding the hidden features of any feature interaction layer and the recommended features input into any feature interaction layer to obtain a feature domain of any feature interaction layer.

For example, in order to include any level of feature interaction information in the final output result, residual connection is used in each layer of feature interaction, that is, the processing procedure of the feature interaction network is as follows

represents recommended features input to the ith feature interaction layer,

representing the ith feature interactionA characteristic field of the layer.

In step 104, index prediction processing is performed based on the feature domains corresponding to the plurality of recommended features, so as to obtain the recommendation index of the target user corresponding to the information to be recommended.

For example, after the feature domains corresponding to the multiple recommendation features are obtained, index prediction processing is performed based on the feature domains corresponding to the multiple recommendation features to obtain recommendation indexes of target users corresponding to information to be recommended, for example, whether the target users click the information to be recommended, whether the target users are interested in the information to be recommended, whether the target users convert based on the information to be recommended, whether the target users evaluate the information to be recommended, and the like, so that tasks to be recommended are executed based on the recommendation indexes of the target users corresponding to the information to be recommended, and accuracy of information recommendation is improved.

In some embodiments, the metric prediction process is implemented by a first prediction model and a second prediction model; the index prediction processing is carried out based on the feature domains corresponding to the plurality of recommended features, so that the recommended index of the target user corresponding to the information to be recommended is obtained, and the index prediction processing comprises the following steps: index prediction processing is carried out on the feature domains corresponding to the recommended features through the first prediction model, and the recommended indexes of the first prediction model are obtained; index prediction processing is carried out on the plurality of recommended features of the task to be recommended through a second prediction model, and recommendation indexes of the second prediction model are obtained; and obtaining the recommendation index of the target user corresponding to the information to be recommended based on the recommendation index of the first prediction model and the recommendation index of the second prediction model.

For example, the process of performing the index prediction processing on the feature domains corresponding to the plurality of recommended features by the first prediction model is as follows: splicing the feature domains corresponding to the recommended features through a first prediction model to obtain spliced feature domains; weighting the spliced feature domain based on the weight of the first prediction model to obtain a weighted feature domain; and adding the bias of the first prediction model and the weighted feature domain to obtain the recommendation index of the first prediction model.

For example, the first prediction model comprises a single forward layer, with a single forward layerThe layer performs index prediction into

Where W represents the weight of the forward layer, b represents the bias of the forward layer, y ₁ The first prediction model in the embodiment of the present application is not limited to the forward layer, and other prediction models are also applicable.

In some embodiments, the second predictive model comprises a plurality of cascaded hidden layers; index prediction processing is carried out on a plurality of recommendation characteristics of the task to be recommended through a second prediction model to obtain recommendation indexes of the second prediction model, and the index prediction processing method comprises the following steps: mapping a plurality of recommendation features of a task to be recommended through a first hidden layer of a plurality of cascaded hidden layers; outputting the mapping result of the first hidden layer to the hidden layers of the subsequent cascade connection, so as to continue mapping processing and outputting the mapping result in the hidden layers of the subsequent cascade connection until the mapping result is output to the last hidden layer; and taking the mapping result output by the last hidden layer as a recommendation index of the second prediction model.

For example, the second prediction model is a Multilayer Perceptron (MLP), and the index prediction by the Multilayer Perceptron is performed as

Wherein, MLP represents the processing procedure of a multi-layer perceptron (comprising a plurality of cascaded hidden layers),

a plurality of recommendation features, y, representing tasks to be recommended ₂ The second prediction model in the embodiment of the present application is not limited to the multi-layer perceptron, and other prediction models are also applicable.

For example, the process of performing mapping processing and mapping result output in the subsequent cascaded hidden layer is as follows: performing the following processing by a jth hidden layer of the plurality of cascaded hidden layers: weighting the mapping result of the j-1 hidden layer based on the weight of the j hidden layer to obtain a weighted mapping result; adding the bias of the jth hidden layer and the weighted mapping result to obtain a mapping result of the jth hidden layer, and outputting the mapping result of the jth hidden layer to a (j + 1) th hidden layer; wherein j is an increasing natural number and the value range of j is more than or equal to 2 and less than or equal to N-1, and N is the number of the plurality of cascaded hidden layers.

In some embodiments, obtaining the recommendation index of the target user corresponding to the information to be recommended based on the recommendation index of the first prediction model and the recommendation index of the second prediction model includes: adding the recommendation index of the first prediction model and the recommendation index of the second prediction model to obtain a summed recommendation index; and normalizing the added recommendation index to obtain the recommendation index of the target user corresponding to the information to be recommended.

For example, the recommendation index of the target user corresponding to the information to be recommended is

Wherein, y ₁ Recommendation index, y, representing the first prediction model prediction ₂ The recommendation index representing the prediction of the second prediction model is represented by σ, and σ represents a normalization function.

In step 105, the task to be recommended is executed based on the recommendation index of the target user corresponding to the information to be recommended.

For example, after the recommendation index of the target user corresponding to the information to be recommended is obtained, the task to be recommended is executed based on the recommendation index of the target user corresponding to the information to be recommended, for example, when the recommendation index of the target user corresponding to the information to be recommended represents that the target user clicks the information to be recommended, the task to be recommended is executed, so that user behavior data are quickly obtained, and the information recommendation effect is improved.

In the following, an exemplary application of the embodiments of the present application in a practical application scenario will be described.

The method and the device can be applied to various product applications including e-commerce shopping, video (or music) recommendation, news information stream recommendation, life service scenes and the like, and can deduce an optimal article set according to the known interest and specific environment information of the user, so that mutual benefits and win-win among a platform party, a content party and the user party are realized.

For example, in movie recommendation, the embodiment of the application can train the CTR prediction model by using scene-related features and user-related features as model input, and can be online for recommending articles that a user may be interested in, so that the accuracy of a recommendation system is remarkably improved, and the use satisfaction of the user on the product is further improved. Similarly, in an online advertising scene, the trained CTR estimation model can be used for accurately recommending commercial advertisements which users may click on, and direct commercial income improvement can be brought to a company.

According to the embodiment of the application, extraction limitations of different feature domains are fully considered, the method has better extraction capability in the aspect of feature interaction, redundant model training is not needed, accuracy of CTR estimation can be improved on the premise of ensuring model training efficiency, and better user experience and more remarkable commercial income are provided.

The click rate estimation has a wide application prospect in an individualized recommendation system, and the probability of clicking a recommended item list by a user in a specific context situation can be estimated, so that the user can efficiently obtain interested information under the condition of information overload. The accurate click rate estimation model is a key link of the personalized recommendation system capable of being efficiently pushed, and the scoring accuracy directly influences the use experience of users and brings remarkable commercial return to company products.

In a specific recommendation scene, because feature spaces of different user interests are often sparse and high-dimensional, the existing click rate estimation has a larger promotion space in two aspects: 1) The accuracy of the model can be improved by manual feature combination, and a click rate estimation model has a larger promotion space when high-order feature representation is extracted; 2) The existing depth model utilizes vector product to constrain the interaction strength between feature domains, or simply projects a single feature space to multiple shared spaces, but cannot acquire deep interaction information between different feature domains. These problems may cause time-consuming and tedious training process, and poor estimation performance, which limits the application effect of CTR estimation technology to a certain extent.

In the related art, the click rate estimation method based on machine learning includes the following two steps: 1. a CTR prediction model based on a Field-aware Factorization (FFM) Machine; 2. and (3) a CTR pre-estimation model based on a depth model. The applicant has found that the above two solutions have their respective drawbacks:

1) Due to the fact that k-dimensional hidden vectors of n features on f domains need to be learned in the training process of the CTR prediction model based on FFM, the calculation complexity of the CTR prediction model rises to kn ² (ii) a On the other hand, the FFM model and the FFM variant are limited by the inherent structure, and only low-order interaction among features can be extracted, so that the estimation capability of the model is limited to a certain extent;

2) The CTR prediction model based on the depth model uses the interaction strength between scalar product constraint characteristic domains, but only a single shared space is simply expanded to a plurality of shared projection spaces, and complete information of different projection spaces cannot be captured completely, so that the prediction performance is reduced.

In order to solve the above problem, an embodiment of the present application provides an artificial intelligence-based information recommendation method, which satisfies a further requirement of a recommendation system on CTR prediction accuracy by performing nonlinear extraction on multi-domain features through a reinforced model. The method learns the semantic diversity of each feature domain (one feature corresponds to one feature domain) through a feature interaction network, wherein the network has domain-aware attributes and can be combined to most of depth models, the high-level ambiguity of feature interaction is further improved by combining a hierarchical structure, and the accuracy of click rate estimation can be effectively improved by combining the feature interaction network and the hierarchical structure.

As shown in fig. 9, the embodiment of the present application provides a feature interaction network based on feature domain awareness, aiming at the problem that a CTR model in a recommendation system has limitations when processing interaction characterizations of different feature domains, and a nonlinear extraction capability can be improved, and including the following steps:

step 1, constructing a data set.

Firstly, converting each film scoring sample into a format required by training, and configuring the film scoring samples into sparse features, continuous features and labels, wherein the labels are historical scores of a user on a film and can be further converted according to an optimization target; inputting the feature interaction network sensed by the feature domain, learning key feature interaction representation by using a feature domain sensing interaction module (namely a feature interaction layer), and further strengthening the estimation accuracy by combining a hierarchical structure; finally, the model predicts the click prediction probability corresponding to each input sample.

The information recommendation method based on artificial intelligence comprises the following steps:

for example, a data set is a collection of historical scoring records of a movie by a user, with data volumes of different sizes, e.g., 1M (containing 1 ten thousand scores), 10M (containing 10 ten thousand scores), and 20M (containing 20 ten thousand scores).

The present embodiment uses a 1M data set that includes 6040 individual users scoring about 100 ten thousand times of 3900 movies. Each sample comprises movie article characteristics, user information characteristics and historical scores of movies by the user, wherein the user score is 5stars and is increased by half a star (0.5 stars-5 stars), and 7 types of characteristics related to movie articles and user information are provided in total.

In the embodiment of the application, all collected samples are randomly divided into a training set, a verification set and a test set according to a ratio of 8.

And 2, preprocessing data.

The embodiment of the application carries out preprocessing conversion on numerical characteristics and user scores:

1) On movie scoring, positive and negative samples are divided by the original numerical value with a threshold value of 3, wherein the samples with the score of more than or equal to 3 are positive samples, and the samples with the score of less than 3 are negative samples.

2) The numerical characteristics are standardized, and the rule is as follows: if feature x >2, and normalized feature y = log2 (x), other normalization operations, such as mean variance normalization, may also be used in embodiments of the present application.

And 3, constructing a feature interaction network based on feature domain perception.

As shown in fig. 10, in consideration that each feature domain is located in a different projection space, the embodiment of the present application proposes a feature domain aware interaction module, and constructs a corresponding feature interaction network.

As shown in fig. 10, each dashed block represents a projection space associated with a feature domain. For feature s _i (i.e. the original features x of the sample _i ) Corresponding projection space i, all input features(s) ₁ To s _K ) Will first be based on the weight w _i1 ，w _i2 ，...，w _iK Zooming, splicing the zoomed features and passing through a projection matrix U _i Projection, vector z obtained after projection _i And then with s _i Element by element multiplication to obtain s _i Is characterized by being s' _i . Projection matrix U in the embodiment of the present application _i And a weight w _ij Is specific to the feature s _i Thus for feature s _i Corresponding feature field capable of projecting all features onto s _i In the corresponding projection space, and then with s _i Feature interaction is performed to capture feature domain information (i.e., new representation s' _i ). And s _i The process of performing feature interaction can be obtained from the following formulas (1) to (2):

where S represents a set of features and K represents the number of features.

s’ _i ＝ReLU(s _i ·z _i ) (2)

Wherein, s' _i Is the feature s in its corresponding projection matrix _i Feature interactions with other features are performed to form new combined features. New set of feature vectors S '= { S' ₁ ，s’ ₂ ，...，s’ _K Will be used for further feature interaction. The above steps are embodied in the data set training, the total feature number of the film article and the user information is 7, so the feature field K is 7And interacting the feature set with other features in a corresponding projection space to further obtain a new feature domain vector set, namely a new feature set.

According to the embodiment of the application, a plurality of channels are expanded for each feature domain, namely, each feature domain comprises a plurality of groups of projection matrixes and weights, so that the expressiveness of the model is improved. For example, each feature domain has H channels, and the feature interaction representation of the multiple channels is as shown in formula (3) -formula (4):

s’ _ih ＝ReLU(s _i ·z _ih ) (4)

wherein each channel h contains a separate projection matrix U _ih And a weight w _ijh . Finally, feature s _i The update of (c) is calculated from all channels, as shown in equation (5):

wherein,

representing any combination function, such as element-by-element summation or element-by-element averaging. The number of channels H can be determined according to the number of training features, and is usually selected in the interval [2,6 ]]H in the embodiment of the present application is 2.

On the other hand, a simple one-layer domain perception feature interaction layer can only capture second-order feature interaction, and the high-order feature interaction is very important for clicking the estimation task. Therefore, in the embodiment of the present application, multiple domain-aware feature interaction layers are stacked to capture higher-order feature interaction information, as shown in formula (6):

wherein, FIL-Layer represents a domain perception characteristic interaction Layer,

the input of the L-th layer is shown, L is the total number of feature interaction layers, and L in the embodiment of the application is configured to be 3.

Representing the initial feature vector, i.e. s _i 。

In order to include feature interaction information of any level in a final output result, the embodiment of the present application uses residual connection in each layer of feature interaction, as shown in formula (7):

after the L-order feature interaction information is obtained, the embodiment of the present application uses a single forward layer to calculate the output probability, as shown in formula (8):

where W represents the weight of the forward layer and b represents the bias of the forward layer.

In order to capture the implicit high-order feature interaction information, the embodiment of the present application also uses an independent multi-layer Perceptron (MLP) to calculate the probability based on the initial feature vector, as shown in formula (9):

will be above y ₁ And y ₂ Adding and normalizing to obtain final output value of the model, wherein the range is [0,1 ]]And representing the probability that the model predicts whether the sample clicks, as shown in formula (10):

and 4, constructing a loss function of the model.

The embodiment of the application uses a cross entropy loss function softmax loss as a loss function of the model, y represents a real label whether a line is clicked or not,

representing the prediction probability of the model for the sample at the time of training. The training target is a prediction task of optimizing whether the model clicks the sample, and is shown in formula (11):

wherein, y ⁽ⁱ⁾ A real tag indicating whether the ith sample clicked or not,

and the prediction probability of the model to the ith sample in training is represented.

And 5, training a feature interaction network based on feature domain perception.

The embodiment of the application adopts a first-order optimization algorithm (Adam optimizer) to solve the parameters of the model, adopts an Xavier mode (an effective neural network initialization method which avoids the attenuation of variance of an activation value by keeping the output value of each layer in Gaussian distribution) to initialize the parameters of the model, and iteratively updates the parameters of the neural network based on training data. In the solving process, training samples (including film article characteristics, user characteristics and corresponding labels) in the data set are transmitted into a model for training, and model optimization is completed through error back propagation.

In summary, the embodiment of the present application has the following beneficial effects:

1) The embodiment of the application provides a feature domain perception interaction module which can realize feature interaction in different projection spaces and further capture high-order semantic diversity of each feature domain to obtain the ambiguity of each feature domain in a high-order semantic level, realize efficient feature interaction in a specific projection space and avoid excessive prior knowledge of interaction features by manual combination;

2) According to the embodiment of the application, the provided feature domain perception interaction module is combined with the hierarchical structure to jointly extract the high-order feature representation, so that the high-order interaction of user behaviors and article information can be accurately learned in movie recommendation and other recommendation scenes, and the accuracy of the prediction of the CTR model is improved;

3) The feature domain perception interaction module provided by the embodiment of the application can be directly applied to a common depth model in a click rate estimation task in a modularization mode, so that a high-efficiency and accurate recommendation model can be trained, and the recommendation model is applied to a film recommendation scene to bring better user experience and higher commercial benefits.

The artificial intelligence based information recommendation method provided by the embodiment of the present application has been described in conjunction with the exemplary application and implementation of the server provided by the embodiment of the present application. In practical applications, each functional module in the information recommendation apparatus may be cooperatively implemented by hardware resources of an electronic device (such as a terminal device, a server, or a server cluster), such as a computing resource of a processor and the like, a communication resource (such as being used to support various modes of communication, such as optical cable and cellular communication), and a memory. Fig. 2 shows an artificial intelligence based information recommendation device 555 stored in a memory 550, which may be software in the form of programs and plug-ins, for example, software modules designed by programming languages such as C/C + +, java, or dedicated software modules, application program interfaces, plug-ins, cloud services, and other implementations in a large software system, and different implementations are exemplified below.

Example I, the information recommendation device is a mobile terminal application program and a module

The information recommendation device 555 in the embodiment of the present application may provide a software module designed using a programming language such as software C/C + +, java, and the like, and embed the software module into various mobile applications based on systems such as Android or iOS (stored in a storage medium of the mobile terminal as an executable instruction and executed by a processor of the mobile terminal), so as to directly use computing resources of the mobile terminal itself to complete related information recommendation tasks, and periodically or aperiodically transmit processing results to a remote server through various network communication methods, or locally store the processing results in the mobile terminal.

Example two, the information recommendation device is a server application and platform

The information recommendation device 555 in this embodiment of the present application may be provided as application software designed using a programming language such as C/C + +, java, or a dedicated software module in a large-scale software system, and run on the server side (stored in a storage medium of the server side in the form of executable instructions and run by a processor of the server side), where the server uses its own computing resources to complete a relevant information recommendation task.

The embodiment of the application can also provide a method for carrying a customized and easily interactive network (Web) Interface or other User Interfaces (UI) on a distributed and parallel computing platform consisting of a plurality of servers to form an information recommendation platform (for information recommendation) used by individuals, groups or units, and the like.

Third, the information recommendation device is a server side Application Program Interface (API) and a plug-in

The information recommendation device 555 in the embodiment of the present application may be provided as an API or a plug-in on a server side, so that a user may call the API or the plug-in to execute the artificial intelligence based information recommendation method in the embodiment of the present application, and embed the information recommendation method in various application programs.

Example four, the information recommendation device is a Mobile device client API and a plug-in

The information recommendation device 555 in the embodiment of the present application may be provided as an API or a plug-in on a mobile device side, so that a user may call the API or the plug-in to execute the artificial intelligence based information recommendation method in the embodiment of the present application.

Example five, the information recommendation device is a cloud open service

The information recommendation device 555 in the embodiment of the application can provide information recommendation cloud services developed for users, so that information recommendation can be performed by individuals, groups or units.

The information recommendation apparatus 555 includes a series of modules, including an obtaining module 5551, a feature interaction module 5552, a prediction module 5553, and a recommendation module 5554. The following continues to describe a scheme for implementing information recommendation by cooperation of each module in the information recommendation apparatus 555, which is provided in the embodiment of the present application.

An obtaining module 5551, configured to obtain a plurality of recommended features of a task to be recommended, where the plurality of recommended features include at least one item feature of information to be recommended and at least one user feature of a target user; the feature interaction module 5552 is configured to perform feature projection processing on the plurality of recommended features based on a projection space corresponding to each recommended feature, so as to obtain projection features of the projection space; performing feature interaction processing based on each recommended feature and the projection feature corresponding to the projection space to obtain a feature domain corresponding to each recommended feature; the prediction module 5553 is configured to perform index prediction processing based on feature domains corresponding to the multiple recommended features, so as to obtain a recommendation index of the target user corresponding to the information to be recommended; a recommending module 5554, configured to execute the task to be recommended based on the recommendation index of the target user corresponding to the information to be recommended.

In some embodiments, the feature interaction module 5552 is further configured to perform nonlinear feature extraction processing on the plurality of recommended features based on a weight of a projection space corresponding to each recommended feature, so as to obtain extracted features of the projection space; and multiplying the extracted features of the projection space with the projection matrix of the projection space to obtain the projection features of the projection space.

In some embodiments, the feature interaction module 5552 is further configured to scale the recommended features based on a weight of a projection space corresponding to each recommended feature, so as to obtain scaled recommended features; and splicing the plurality of zoomed recommended features to obtain the extracted features of the projection space.

In some embodiments, the feature interaction module 5552 is further configured to multiply each of the recommended features and the projection features corresponding to the projection space to obtain a feature vector of each of the recommended features; and mapping the feature vector of each recommended feature to obtain a feature domain corresponding to each recommended feature.

In some embodiments, the projection space corresponding to each of the recommended features includes a plurality of channels; the feature interaction module 5552 is further configured to perform feature projection processing on the plurality of recommended features based on any one of the channels of the projection space corresponding to each recommended feature, so as to obtain a projection feature of any one of the channels; performing feature interaction processing on each recommended feature and the projection feature corresponding to any channel to obtain a feature domain corresponding to any channel; and performing fusion processing on the feature domains corresponding to the channels of the projection space respectively to obtain the feature domain corresponding to each recommended feature.

In some embodiments, the feature interaction module 5552 is further configured to sum feature domains corresponding to a plurality of channels of the projection space, and use a result of the summation as a feature domain corresponding to each of the recommended features; or averaging feature domains corresponding to a plurality of channels of the projection space, and taking the result of the averaging as the feature domain corresponding to each recommended feature.

In some embodiments, a feature interaction network for feature interaction includes a plurality of cascaded feature interaction layers; the feature interaction module 5552 is further configured to perform the following processing by any one of the feature interaction layers in the feature interaction network: performing feature projection processing on the plurality of recommended features input to any one of the feature interaction layers based on the projection space corresponding to each recommended feature to obtain the projection features of any one of the feature interaction layers; the plurality of recommended features input into the first feature interaction layer are the plurality of recommended features included in the task to be recommended, the plurality of recommended features input into the subsequent feature interaction layer are the plurality of recommended features output by the last feature interaction layer of the subsequent feature interaction layer, and the subsequent feature interaction layer is a feature interaction layer except the first feature interaction layer in the plurality of cascaded feature interaction layers; performing, by any of the feature interaction layers in the feature interaction network: performing feature interaction processing based on the recommended features input to any one of the feature interaction layers and the projection features corresponding to any one of the feature interaction layers to obtain a feature domain of any one of the feature interaction layers, and outputting the recommended features included in the feature domain; and determining the feature domain of the last feature interaction layer as the feature domain corresponding to each recommended feature.

In some embodiments, the feature interaction module 5552 is further configured to perform feature interaction processing based on the recommended features input to any one of the feature interaction layers and the projection features corresponding to any one of the feature interaction layers, so as to obtain hidden features of any one of the feature interaction layers; and adding the hidden features of any one feature interaction layer and the recommended features input into any one feature interaction layer to obtain a feature domain of any one feature interaction layer.

In some embodiments, the metric prediction process is implemented by a first prediction model and a second prediction model; the prediction module 5553 is further configured to perform index prediction processing on the feature domains corresponding to the recommended features through the first prediction model to obtain recommended indexes of the first prediction model; index prediction processing is carried out on the plurality of recommended features of the task to be recommended through the second prediction model, and recommendation indexes of the second prediction model are obtained; and obtaining the recommendation index of the information to be recommended corresponding to the target user based on the recommendation index of the first prediction model and the recommendation index of the second prediction model.

In some embodiments, the prediction module 5553 is further configured to perform a splicing process on feature domains corresponding to the recommended features through the first prediction model to obtain spliced feature domains; weighting the spliced feature domain based on the weight of the first prediction model to obtain a weighted feature domain; and adding the bias of the first prediction model and the weighted feature domain to obtain the recommendation index of the first prediction model.

In some embodiments, the prediction module 5553 is further configured to sum the recommendation index of the first prediction model and the recommendation index of the second prediction model to obtain a summed recommendation index; and carrying out normalization processing on the added recommendation index to obtain the recommendation index of the target user corresponding to the information to be recommended.

In some embodiments, the second predictive model comprises a plurality of cascaded hidden layers; the prediction module 5553 is further configured to map, by using a first hidden layer of the plurality of cascaded hidden layers, a plurality of recommended features of the task to be recommended; outputting the mapping result of the first hidden layer to a hidden layer of a subsequent cascade, so as to continue mapping processing and outputting the mapping result in the hidden layer of the subsequent cascade until the mapping result is output to a last hidden layer; and taking the mapping result output by the last hidden layer as a recommendation index of the second prediction model.

In some embodiments, the prediction module 5553 is further configured to perform the following processing by a jth hidden layer of the plurality of cascaded hidden layers: weighting the mapping result of the j-1 hidden layer based on the weight of the j hidden layer to obtain a weighted mapping result; adding the bias of the jth hidden layer and the weighted mapping result to obtain a mapping result of the jth hidden layer, and outputting the mapping result of the jth hidden layer to a (j + 1) th hidden layer; wherein j is an increasing natural number and the value range of j is more than or equal to 2 and less than or equal to N-1, and N is the number of the cascaded hidden layers.

Embodiments of the present application provide a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device executes the artificial intelligence based information recommendation method according to the embodiment of the application.

Embodiments of the present application provide a computer-readable storage medium storing executable instructions, which when executed by a processor, cause the processor to perform an artificial intelligence based information recommendation method provided by embodiments of the present application, for example, the artificial intelligence based information recommendation method shown in fig. 3-5.

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

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

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

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

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

Claims

1. An artificial intelligence based information recommendation method, characterized in that the method comprises:

the method comprises the steps of obtaining a plurality of recommendation characteristics of a task to be recommended, wherein the recommendation characteristics comprise at least one article characteristic of information to be recommended and at least one user characteristic of a target user;

2. The method according to claim 1, wherein the performing feature projection processing on the plurality of recommended features based on the projection space corresponding to each recommended feature to obtain the projection features of the projection space comprises:

performing nonlinear feature extraction processing on the plurality of recommended features based on the weight of the projection space corresponding to each recommended feature to obtain extracted features of the projection space;

and multiplying the extracted features of the projection space with the projection matrix of the projection space to obtain the projection features of the projection space.

3. The method according to claim 2, wherein the performing a nonlinear feature extraction process on the plurality of recommended features based on the weight of the projection space corresponding to each recommended feature to obtain an extracted feature of the projection space comprises:

based on the weight of the projection space corresponding to each recommended feature, carrying out scaling processing on the recommended features to obtain the scaled recommended features;

4. The method according to claim 1, wherein the obtaining a feature domain corresponding to each recommended feature based on feature interaction processing on each recommended feature and a projection feature corresponding to the projection space comprises:

multiplying each recommended feature and the projection feature corresponding to the projection space to obtain a feature vector of each recommended feature;

5. The method according to any one of claims 1 to 3,

the projection space corresponding to each recommended feature comprises a plurality of channels;

the feature projection processing is performed on the plurality of recommended features based on the projection space corresponding to each recommended feature to obtain the projection features of the projection space, and the feature projection processing includes:

performing feature projection processing on the plurality of recommended features based on any channel of the projection space corresponding to each recommended feature to obtain the projection feature of any channel;

the obtaining a feature domain corresponding to each recommended feature based on feature interaction processing on each recommended feature and the projection feature corresponding to the projection space includes:

6. The method according to claim 5, wherein the fusing the feature domains corresponding to the plurality of channels of the projection space to obtain the feature domain corresponding to each of the recommended features includes:

adding feature domains corresponding to a plurality of channels of the projection space respectively, and taking the result of the addition as the feature domain corresponding to each recommended feature; or,

7. The method of claim 1,

the feature interaction network for feature interaction comprises a plurality of cascaded feature interaction layers;

the performing feature projection processing on the plurality of recommended features based on the projection space corresponding to each recommended feature to obtain the projection features of the projection space includes:

performing, by any one of the feature interaction layers in the feature interaction network:

the performing feature interaction processing based on each recommended feature and the projection feature corresponding to the projection space to obtain a feature domain corresponding to each recommended feature includes:

8. The method of claim 7, wherein the performing feature interaction processing based on the recommended features input to any one of the feature interaction layers and the projection features corresponding to any one of the feature interaction layers to obtain a feature domain of any one of the feature interaction layers comprises:

performing feature interaction processing based on the recommended features input to any one of the feature interaction layers and the projection features corresponding to any one of the feature interaction layers to obtain hidden features of any one of the feature interaction layers;

9. The method of claim 1,

the index prediction processing is realized by a first prediction model and a second prediction model;

the index prediction processing is performed based on the feature domains corresponding to the plurality of recommended features to obtain the recommendation index of the target user corresponding to the information to be recommended, and the index prediction processing includes:

10. The method according to claim 9, wherein the performing, by the first prediction model, index prediction processing on the feature domains corresponding to the plurality of recommended features to obtain the recommended index of the first prediction model includes:

splicing the feature domains corresponding to the recommended features through the first prediction model to obtain spliced feature domains;

11. The method of claim 9,

the second predictive model comprises a plurality of cascaded hidden layers;

the index prediction processing is performed on the plurality of recommended features of the task to be recommended through the second prediction model to obtain the recommendation index of the second prediction model, and the index prediction processing comprises the following steps:

mapping a plurality of recommendation features of the task to be recommended through a first hidden layer of the plurality of cascaded hidden layers;

12. The method according to claim 11, wherein the continuing mapping processing and mapping result output in the subsequent cascaded hidden layers comprises:

performing, by a jth hidden layer of the plurality of cascaded hidden layers:

13. An artificial intelligence-based information recommendation apparatus, characterized in that the apparatus comprises:

the system comprises an acquisition module, a recommendation module and a recommendation module, wherein the acquisition module is used for acquiring a plurality of recommendation characteristics of a task to be recommended, and the plurality of recommendation characteristics comprise at least one article characteristic of information to be recommended and at least one user characteristic of a target user;

and the recommending module is used for executing the tasks to be recommended according to the recommending indexes of the target users corresponding to the information to be recommended.

14. An electronic device, characterized in that the electronic device comprises:

a memory for storing executable instructions;

a processor for implementing the artificial intelligence based information recommendation method of any one of claims 1 to 12 when executing the executable instructions stored in the memory.

15. A computer-readable storage medium storing executable instructions for implementing the artificial intelligence based information recommendation method of any one of claims 1 to 12 when executed by a processor.