CN112231580B

CN112231580B - Information recommendation method and device based on artificial intelligence, electronic equipment and storage medium

Info

Publication number: CN112231580B
Application number: CN202011245105.5A
Authority: CN
Inventors: 赵教生
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-11-10
Filing date: 2020-11-10
Publication date: 2024-04-02
Anticipated expiration: 2040-11-10
Also published as: CN112231580A

Abstract

The application provides an information recommendation method and device based on artificial intelligence, electronic equipment and a computer readable storage medium; big data technology relating to artificial intelligence; the method comprises the following steps: acquiring a content tag of information displayed in an information display page; determining a correspondence between content tags of the information samples and content tags of the recommended information sample set based on the correspondence between the information samples and the recommended information sample set and the operation data of the recommended information sample set; determining a content label for recommendation based on the corresponding relation and the content label of the information; inquiring information matched with the content label to be recommended from an information base to be used as recommendation information; and performing a recommendation operation in the information presentation page based on the recommendation information. Through the information recommendation method and device, information recommendation efficiency can be improved.

Description

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

Technical Field

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

Background

Artificial intelligence (Artificial Intelligence, AI) is a comprehensive technology of computer science, and by researching the design principles and implementation methods of various intelligent machines, the machines have the functions of sensing, reasoning and decision. Artificial intelligence technology is a comprehensive subject, and relates to a wide range of fields, such as natural language processing technology, machine learning/deep learning and other directions, and with the development of technology, the artificial intelligence technology will be applied in more fields and has an increasingly important value.

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

Although, the recommendation system in the related art may recommend information to the user that may be of interest to the user. However, the recommendation system in the related art recommends to the user through a large amount of history recommendation information, resulting in improving the efficiency of recommendation to the user.

Disclosure of Invention

The embodiment of the application provides an information recommendation method, device, electronic equipment and computer readable storage medium based on artificial intelligence, which can be used for recommending information by combining content tags, so that the efficiency of information recommendation is 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:

acquiring a content tag of information displayed in an information display page;

determining a correspondence between content tags of an information sample and content tags of a recommended information sample set based on the correspondence between the information sample and the recommended information sample set and operation data of the recommended information sample set;

determining a content label for recommendation based on the corresponding relation and the content label of the information;

inquiring information matched with the content label to be recommended from an information base to serve as recommendation information;

and executing recommendation operation in the information display page based on the recommendation information.

In the above technical solution, the determining the content tag for the to-be-recommended content tag based on the correspondence and the content tag of the information includes:

training the neural network model for information recommendation based on the corresponding relation between the content label of the information sample and the content label of the recommended information sample set to obtain a trained neural network model;

And carrying out label-based prediction processing on the content labels of the information through the trained neural network model to obtain the content labels to be recommended.

In the above technical solution, the performing, by using the trained neural network model, label-based prediction processing on the content label of the information to obtain a content label to be recommended includes:

executing the following processing through the trained neural network model:

coding the content label of the information to obtain a coding vector corresponding to the content label of the information;

decoding the coded vector to generate a label vector corresponding to the content label of the information;

and combining the labels corresponding to the label vectors according to the sequence of the generated label vectors to obtain the content labels for recommendation.

In the above technical solution, the decoding the encoded vector to generate a tag vector corresponding to a content tag of the information includes:

decoding the coded vector to generate a 1 st tag vector corresponding to the content tag of the information;

decoding processing is carried out based on the coding vector and the generated first i-1 tag vectors, and an ith tag vector corresponding to the content tag of the information is generated;

Wherein, i is more than or equal to 2 and less than or equal to M, i is a natural number, and M is the total number of label vectors corresponding to the content labels of the information.

In the above technical solution, the trained neural network model includes a plurality of cascaded decoders;

the decoding process is performed based on the encoded vector and the generated first i-1 tag vectors, and an ith tag vector corresponding to the content tag of the information is generated, which includes:

decoding, by a first decoder of the plurality of concatenated decoders, based on the encoded vector and the generated first i-1 tag vectors;

outputting the decoding result of the first decoder to a subsequent cascade decoder so as to continuously perform decoding processing and outputting the decoding result in the subsequent cascade decoder until outputting to a last decoder;

and taking the decoding result output by the last decoder as an ith tag vector corresponding to the content tag of the information.

In the above technical solution, the continuing the decoding process and outputting the decoding result in the decoder of the subsequent cascade includes:

performing, by a kth decoder of the plurality of concatenated decoders, the following:

Mapping the decoding result of the (k-1) decoder to obtain a (k) mapping vector;

carrying out fusion processing on the kth mapping vector and the coding vector to obtain a kth fusion vector;

mapping the kth fusion vector, taking the obtained mapping result as a decoding result of the kth decoder, and outputting the decoding result of the kth decoder to a (k+1) th decoder;

wherein k is more than or equal to 2 and less than or equal to T-1, k is a natural number, and T is the number of the plurality of cascaded decoders.

In the above technical scheme, the mapping processing is performed on the decoding result of the kth-1 decoder to obtain the kth mapping vector, including:

performing self-attention processing on the decoding result of the kth-1 decoder to obtain a kth self-attention vector;

and carrying out residual connection processing on the kth self-attention vector and the decoding result of the kth-1 decoder, and taking the residual connection result as a kth mapping vector.

In the above technical solution, the fusing processing is performed on the kth mapping vector and the coding vector to obtain the kth fusion vector, including:

performing multi-head attention processing on the kth mapping vector and the coding vector to obtain a kth multi-head attention vector;

And carrying out residual connection processing on the kth multi-head attention vector and the kth mapping vector, and taking a residual connection result as a kth fusion vector.

In the above technical solution, the mapping the kth fusion vector, taking the obtained mapping result as the decoding result of the kth decoder, includes:

carrying out nonlinear mapping processing on the kth fusion vector to obtain a kth mapping vector;

and carrying out residual connection processing on the kth mapping vector and the kth fusion vector, and taking a residual connection result as a decoding result of the kth decoder.

The embodiment of the application provides an information recommendation device, which comprises:

the acquisition module is used for acquiring the content tag of the information displayed in the information display page;

the determining module is used for determining the corresponding relation between the content label of the information sample and the content label of the recommended information sample set based on the corresponding relation between the information sample and the recommended information sample set and the operation data of the recommended information sample set;

the processing module is used for determining a content label to be recommended based on the corresponding relation and the content label of the information;

The matching module is used for inquiring information matched with the content tag to be recommended from the information base to be used as recommendation information; and executing recommendation operation in the information display page based on the recommendation information.

The embodiment of the application provides an electronic device for information recommendation, which comprises:

a memory for storing executable instructions;

and the processor is used for realizing the information recommendation method based on artificial intelligence when executing the executable instructions stored in the memory.

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

The embodiment of the application has the following beneficial effects:

and determining the content label to be recommended according to the corresponding relation between the content label of the information sample and the content label of the recommended information sample set, and recommending the information according to the content label to be recommended, so that the information can be recommended through a small number of content labels, and the information recommendation efficiency is improved.

Drawings

Fig. 1 is an application scenario schematic diagram 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 according to an embodiment of the present application;

FIGS. 3A-3C are schematic flow diagrams of an artificial intelligence based information recommendation method provided in an embodiment of the present application;

fig. 4 is a schematic structural diagram of a neural network model provided in an embodiment of the present application;

fig. 5 is a schematic flow chart of a decoding process provided in an embodiment of the present application;

FIG. 6 is a schematic illustration of a connection formed by a random walk provided by the related art;

FIG. 7 is a schematic diagram of a sequence map provided by the related art;

FIG. 8 is a flow diagram of a hierarchical classification provided by the related art;

FIG. 9 is a schematic diagram of a related art structure of video recommendation based on user operations;

FIG. 10 is a schematic diagram of a master feeds scenario provided by an embodiment of the present application;

fig. 11 is a schematic diagram of a video floating layer scene provided in an embodiment of the present application;

FIG. 12 is a schematic diagram of a transducer model according to an embodiment of the present disclosure;

fig. 13 is a comparative schematic diagram of effect improvement provided in the embodiment of the present application.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the present application more apparent, the present application will be described in further detail with reference to the accompanying drawings, and the described embodiments should not be construed as limiting the present application, and all other embodiments obtained by those skilled in the art without making any inventive effort are within the scope of the present application.

In the following description, the terms "first", "second", and the like are merely used to distinguish similar objects and do not represent a particular ordering of the objects, it being understood that the "first", "second", and the like may be interchanged with one another, if permitted, to enable embodiments of the application described herein to be practiced otherwise than as 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 present application.

Before further describing embodiments of the present application in detail, the terms and expressions that are referred to in the embodiments of the present application are described, and are suitable for the following explanation.

1) Convolutional neural network (CNN, convolutional Neural Networks): one type of feedforward neural network (FNN, feedforward Neural Networks) that includes convolution calculations and has a Deep structure is one of representative algorithms of Deep Learning. Convolutional neural networks have the capability of token learning (Representation Learning) to enable a translation invariant classification (Shift Invariant Classification) of input images in their hierarchical structure.

2) Target user: a user currently using the recommendation system (e.g., a real user or a virtual user modeled by a computer program), for example, real user a currently uses the news recommendation system to brush news, real user a is the target object. Wherein the sample user is a user who uses the recommendation system other than the target user.

3) Feed (feeds) stream: continuously updating and presenting the information stream of the user content. feed is the grouping together of several sources of messages actively subscribed to by a user to form a content aggregator that helps the user to continuously acquire the latest feed content, where feeds are typically news websites and blogs. There are various presentation forms of the feed stream, mainly including a timeline (timeline) and a rank (rank), where the timeline is a presentation manner of the feed stream, and the content is presented to the user according to the chronological order of the content update of the feed stream, such as microblog and friend circle; rank is the weight of calculating the content according to some factors, so as to determine the sequence of content presentation, for example, the current microblog homepage information flow algorithm discards the original timeline and adopts the latest intelligent ordering. The main feeds recommend pages for the top page in the information flow product.

4) Video floating layer: in the information flow consumption scene, after clicking a certain video on the main feeds, the user enters a video playing page, and the page can continuously slide downwards and push other videos.

5) Content tag: a tag that characterizes the content of the information, e.g., the content tag of a video is "xiaoming", then the content that characterizes the video is a related xiaoming video; if the content label of a commodity is meat, the commodity is characterized as belonging to the meat.

6) Multilayer perceptron (MLP, multilayer Perceptron): a feed-forward artificial neural network model maps multiple data sets of an input onto a single data set of an output.

The embodiment of the application provides an information recommending method and device based on artificial intelligence, electronic equipment and a computer readable storage medium, and can improve information recommending efficiency.

The information recommendation method based on artificial intelligence provided by the embodiment of the application can be independently realized by a terminal/server; the information recommendation method based on artificial intelligence described below may also be implemented cooperatively by the terminal and the server, for example, the terminal alone bears the information recommendation method based on artificial intelligence described below, or the terminal transmits an information recommendation request to the server, and the server executes the information recommendation method based on artificial intelligence according to the received information recommendation request and transmits recommendation information in response to the information recommendation request to the terminal, so as to display the recommendation information at the terminal.

The electronic device for information recommendation provided by the embodiment of the application may be various types of terminal devices or servers, wherein the servers may be independent physical servers, may be a server cluster or a distributed system formed by a plurality of physical servers, and may also be cloud servers for providing cloud computing services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, basic cloud computing services such as big data and artificial intelligent platforms; the terminal may be, but is not limited to, a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, a car computer, etc. The terminal and the server may be directly or indirectly connected through wired or wireless communication, which is not limited herein.

Taking a server as an example, for example, a server cluster deployed in a cloud may be used, an artificial intelligence cloud Service (aias a Service, AIaaS) is opened to users, an AIaaS platform splits several common AI services and provides independent or packaged services in the cloud, and the Service mode is similar to an AI theme mall, and all users can access one or more artificial intelligence services provided by using the AIaaS platform through 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 a program of information recommendation provided in the embodiments of the present application. A user invokes an information recommendation service in a cloud service through a terminal (a client such as a news client, a shopping client and the like is operated), so that a server deployed at a cloud end invokes a packaged information recommendation program, a content label for recommendation is determined through a corresponding relation between a content label of an information sample and a content label of a recommendation information sample set, information recommendation is performed according to the content label for recommendation, for example, for a news application, the content label for recommendation is determined based on a content label of news displayed on a news display page, and a recommended news stream (recommended news sequence) is determined based on the content label for recommendation, so that a target user can continuously browse news meeting the interest of the target user in order to quickly respond to a news recommendation request; for shopping application, based on the content label of the commodity displayed on the commodity display page, determining a content label for recommendation, and based on the content label for recommendation, determining a recommended commodity flow (recommended commodity sequence), so as to quickly respond to a commodity recommendation request, recommend the commodity meeting the interest of the target user to the target user, and promote the shopping desire of the target user.

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, a terminal 200 is connected to a server 100 through a network 300, where the network 300 may be a wide area network or a local area network, or a combination of the two.

The terminal 200 (running with clients, e.g., news clients, shopping clients, etc.) may be used to obtain information recommendation requests, e.g., when the target user opens a news application and clicks on a certain news displayed on the news presentation page, the terminal automatically obtains the news recommendation request.

In some embodiments, an information recommendation plug-in can be implanted in a client running in the terminal to locally implement an artificial intelligence based information recommendation method on the client. For example, after the terminal 200 obtains the information recommendation request, the information recommendation plug-in is called to implement an information recommendation method based on artificial intelligence, a content label for recommendation is determined according to a corresponding relation between a content label of an information sample and a content label of a recommendation information sample set, and information recommendation is performed according to the content label for recommendation, so as to respond to the information recommendation request, for example, for a news application, a target user clicks a certain news displayed on a news display page, then the news recommendation request is automatically obtained, and based on the content label of the news, the content label for recommendation is determined, and based on the content label for recommendation, a recommendation news stream is determined, so that the target user can browse news meeting the interest of the target user continuously, and the news recommendation request is responded quickly.

In some embodiments, after the terminal 200 obtains the information recommendation request, the information recommendation interface of the server 100 (may be provided in a cloud service form, that is, an information recommendation service) is invoked, and the server 100 determines a content label for recommendation according to a corresponding relationship between a content label of an information sample and a content label of a recommendation information sample set, and performs information recommendation according to the content label for recommendation, so as to respond to the information recommendation request, for example, for a shopping application, when a target user clicks a certain commodity displayed on a commodity display page, automatically obtain the commodity recommendation request, determine the content label for recommendation based on the content label of the commodity, and determine a recommended commodity flow based on the content label for recommendation, so as to quickly respond to the commodity recommendation request, thereby recommending a commodity meeting the interest of the target user to the target user, and improving the shopping desire of the target user.

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

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

Memory 550 includes volatile memory or nonvolatile memory, and may also include both volatile and nonvolatile memory. The non-volatile memory may be a read only memory (ROM, read Onl y Memory) and the volatile memory may be a random access memory (RAM, random Access M emory). The memory 550 described in embodiments herein is intended to comprise any suitable type of memory. Memory 550 may optionally include one or more storage devices physically located remote from processor 510.

In some embodiments, memory 550 is capable of storing 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 handling 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 handling hardware-based tasks;

Network communication module 552 is used to reach other computing devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 include: bluetooth, wireless compatibility authentication (WiFi), and universal serial bus (USB, universal Serial Bus), etc.;

in some embodiments, the information recommending apparatus based on artificial intelligence provided in the embodiments of the present application may be implemented in a software manner, for example, may be an information recommending plug-in the terminal described above, and may be an information recommending service in the server described above. Of course, without being limited thereto, the artificial intelligence based information recommendation apparatus provided in the embodiments of the present application may be provided in various forms including application programs, software modules, scripts or codes.

FIG. 2 illustrates 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, such as information recommendation plug-ins, and includes a series of modules including an acquisition module 5551, a determination module 5552, a processing module 5553, a matching module 5554, and a traversal module 5555; the acquiring module 5551, the determining module 5552, the processing module 5553, the matching module 5554 and the traversing module 5555 are configured to implement the information recommending function provided in the embodiments of the present application.

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

In the following steps, the information displayed by the information display page and the recommended information can be text, graphics context, video and other data, for example, the recommended information can be an answer in text form for a search engine; for news applications, the recommendation information may be a message in the form of a graphic text; for video applications, the recommendation information may be data in the form of video.

In step 101, a content tag of information displayed in an information presentation page is acquired.

As the information displayed in the information display page is obtained, for example, after the target user opens a certain information recommendation application at the terminal, various information is presented in the information display page, the target user clicks a certain information in the information display page, a specific page of the information is displayed in the information display page, the terminal automatically obtains an information recommendation request and sends the information recommendation request to the server, and the server obtains a content tag of the information displayed in the information display page according to the information recommendation request, so that the content tag for the information to be recommended is obtained according to the content tag of the information.

In step 102, a correspondence between content tags of the information samples and content tags of the recommended information sample set is determined based on the correspondence between the information samples and the recommended information sample set and the operation data of the recommended information sample set.

Wherein, there is no obvious sequence of steps 101 and 102. In order to be able to determine the content to be recommended subsequently, it is necessary to determine the correspondence between the content tag of the information sample and the content tag of the recommended information sample set first, i.e. the correspondence between the content tag of the information sample and the content tag of the recommended information sample set may be determined based on the correspondence between the information sample and the recommended information sample set and the operation data of the recommended information sample set.

Wherein the operation data characterizes data of user operation recommendation information, such as a viewing duration of user viewing history recommendation information; the user requests, forwards, pays attention to and the like data of the historical recommendation information. The recommended information sample set is a sample information set that is recommended for an information sample, for example, if the information sample is video 1, the recommended information sample set is video 2, video 3, and video 4.

In some embodiments, before determining the correspondence between the content tags of the information samples and the content tags of the recommended information sample set, the method further comprises: traversing the historical operation sequence of at least one sample user to obtain the corresponding relation between the recommended information sample set and the information samples received by each sample user and the operation data of the recommended information samples in the recommended information sample set.

The historical operation sequence characterizes a set of operation data generated by interaction between a sample user and the information display page. Traversing a historical operation sequence of a sample user to obtain a corresponding relation between a recommended information sample set and an information sample received by the sample user and operation data of recommended information samples in the recommended information sample set. Wherein the set of recommended information samples received by the sample user characterizes the set of recommended information samples presented on the information presentation page.

With the above example in mind, the following processing is performed for any one of the recommended information samples in the recommended information sample set: based on operation data of recommended information samples of each sample user, carrying out aggregation average processing on the recommended information samples based on the operation data, and taking an obtained aggregation average result as an operation characteristic of the recommended information samples; screening content labels included in each recommended information sample based on the operation characteristics of each recommended information sample; and constructing a corresponding relation between the content label of the information sample and the content label remained after the screening process, and taking the corresponding relation as the corresponding relation between the content label of the information sample and the content label of the recommended information sample set.

For example, the operational data of the recommended information sample for the sample user characterizes the sample user's viewing duration of the recommended information sample. For example, if the viewing time of the sample user 1 for the recommended information sample 1 is 5 minutes, the viewing time of the sample user 2 for the recommended information sample 1 is 10 minutes, and the viewing time of the sample user 3 for the recommended information sample 1 is 3 minutes, the aggregate average of the recommended information sample 1 based on the viewing time of the sample user 1 for the recommended information sample 1, the viewing time of the sample user 2 for the recommended information sample 1, and the viewing time of the sample user 3 for the recommended information sample 1 is performed, so that the aggregate average result is 6, and the operation feature of the recommended information sample 1 is 6. And screening the content labels included in each recommended information sample based on the operation characteristics of each recommended information sample, and if the content labels remaining after the screening are the content label 1 and the content label 2 and the content label of the information sample is the content label 3, constructing the corresponding relation between the content label 3 and the content label 1 and the content label 2, so that the characteristics of the watching duration are integrated in the content labels.

For example, the operational data of the recommended information sample for the sample user may also characterize the number of praise times of the recommended information sample by the sample user, and so on. However, because the number of praise times of the sample user on the recommended information sample is relatively small, the preference degree of the sample user on various recommended information samples cannot be covered completely.

In some embodiments, filtering content tags included in each recommended information sample based on the operational characteristics of each recommended information sample includes: performing de-duplication processing on content labels respectively included in each recommended information sample to obtain a plurality of content labels for screening processing; the following processing is performed for any one of the plurality of content tags for performing the filtering processing: when the recommended information sample comprises a content tag, carrying out aggregation average processing on the content tag based on the operation characteristics of the recommended information sample, and taking an obtained aggregation average result of the content tag as the operation characteristics of the content tag; and sorting the content labels in a descending order based on the operation characteristics of the content labels, and taking a plurality of content labels sorted in the descending order as the content labels remained after the screening processing.

In the above example, the content tags included in each recommended information sample are determined, and then the determined content tags are subjected to a deduplication process to obtain a plurality of non-repeated content tags for performing a subsequent filtering process, and the following process is performed for any content tag (for example, content tag 1) of the plurality of content tags for performing the filtering process: when the recommended information sample includes the content tag 1, based on the operation characteristic of the recommended information sample, performing aggregation average on the content tags based on the operation characteristic, taking the obtained aggregation average result of the content tags as the operation characteristic of the content tags, for example, the recommended information sample 1 includes the content tag 1, the recommended information sample 2 includes the content tag 1, the recommended information sample 3 includes the content tag 1, the operation characteristic of the recommended information sample 1 is 3, the operation characteristic of the recommended information sample 2 is 6, and the operation characteristic of the recommended information sample 3 is 9, then the operation characteristic of the content tag 1 is 6, finally, based on the operation characteristic of each content tag, performing descending order on each content tag, taking a plurality of content tags ranked in the descending order as content tags remaining after the screening processing, for example, taking L content tags ranked in front as content tags remaining after the screening processing, wherein L is a natural number.

In some embodiments, taking the obtained aggregate average result of the content tag as the operation feature of the content tag includes: performing feature extraction processing on the portrait information of the target user to obtain the weight of each content label; and weighting the obtained aggregation average result of the content tags based on the weight of each content tag, and taking the weighted result as the operation characteristic of the content tag.

And carrying out feature extraction processing on the portrait information of the target user to obtain the weight of each content tag, namely the preference of the target user to each content tag, weighting the obtained aggregation average result of the content tags based on the preference of the target user to each content tag, and taking the weighted result as the operation feature of the content tag, thereby integrating the portrait information of the target user into the operation feature of the content tag to improve the pertinence of information recommendation.

In step 103, a content tag for the content to be recommended is determined based on the correspondence and the content tag of the information.

After the server obtains the correspondence between the content tag of the information sample and the content tag of the recommended information sample set, the content tag for recommendation can be obtained based on the correspondence and the content tag of the information, so that information recommendation can be performed according to the content tag for recommendation.

Referring to fig. 3B, fig. 3B is a schematic flow chart of an alternative method for recommending information based on artificial intelligence according to an embodiment of the present invention, and fig. 3B illustrates that step 103 in fig. 3A may be implemented by steps 1031A to 1032A illustrated in fig. 3B: in step 1031A, a data table including correspondence is queried with the content tag of the information as an index; in step 1032A, when the content tag of the information sample matching the content tag of the information is queried, the content tag of the recommended information sample set corresponding to the content tag of the information sample is used as the content tag to be recommended.

After the server obtains the correspondence between the content tags of the information samples and the content tags of the recommended information sample set, the correspondence between the content tags of the information samples and the content tags of the recommended information sample set is stored in the data table so as to facilitate subsequent query of the correspondence in the data table. After the server obtains the content tag of the information, the corresponding relation in the data table is inquired, and when the content tag of the information sample matched with the content tag of the information is inquired, the content tag of the recommended information sample set corresponding to the content tag of the information sample is used as the content tag to be recommended. Therefore, the content label corresponding to the content label of the information sample can be quickly searched in a table look-up mode, and the subsequent information recommendation can be carried out through the content label to be recommended.

Referring to fig. 3C, fig. 3C is a schematic flow chart of an alternative method for recommending information based on artificial intelligence according to an embodiment of the present invention, and fig. 3C illustrates that step 103 in fig. 3A may be implemented by steps 1031B to 1032B illustrated in fig. 3C: in step 1031B, training a neural network model for information recommendation based on a correspondence between content tags of the information samples and content tags of the recommended information sample set to obtain a trained neural network model; in step 1032B, the label-based prediction process is performed on the content labels of the information through the trained neural network model, so as to obtain content labels to be recommended.

After the server obtains the correspondence between the content tag of the information sample and the content tag of the recommended information sample set, training a neural network model (e.g., a transducer) for information recommendation based on the correspondence between the content tag of the information sample and the content tag of the recommended information sample set to obtain a trained neural network model, so that the trained neural network model learns the correspondence between the content tag of the information and the content tag of the recommended information. Therefore, accurate label prediction can be performed in an artificial intelligence mode, and the content label to be recommended is obtained so as to perform subsequent information recommendation through the content label to be recommended. The user operation is fitted through the content label angle, and the dimension of the content label number is smaller, so that the model is easier to train and can be converged better, and training of the model can be supported only by the content label of the new information sample appearing in the training set, namely the content label is relatively very stable, so that the cold start problem of the new information can be solved.

In some embodiments, training a neural network model for information recommendation based on a correspondence between content tags of information samples and content tags of a set of recommended information samples, resulting in a trained neural network model, comprising: performing label-based prediction processing on the information sample through a neural network model to obtain a prediction label for recommendation; constructing a loss function of the neural network model based on the predictive label and the content label of the recommended information sample set; updating parameters of the neural network model until the loss function converges, and taking the updated parameters of the neural network model when the loss function converges as parameters of the trained neural network model.

For example, after the values of the loss functions of the neural network model are determined, the values of the loss functions of the neural network model can be judged whether to exceed a preset threshold, when the values of the loss functions of the neural network model exceed the preset threshold, error signals of the neural network model are determined based on the loss functions of the neural network model, the error information is reversely propagated in the neural network model, and model parameters of each layer are updated in the propagation process.

Here, the back propagation is described, the training sample data is input to the input layer of the neural network model, passes through the hidden layer, finally reaches the output layer and outputs the result, which is the forward propagation process of the neural network model, because the output result of the neural network model has errors with the actual result, the errors between the output result and the actual value are calculated, and the errors are propagated back from the output layer to the hidden layer until the errors are propagated to the input layer, and in the back propagation process, the values of the model parameters are adjusted according to the errors; the above process is iterated until convergence.

In some embodiments, performing label-based prediction processing on the content label of the information through the trained neural network model to obtain a content label to be recommended, including: the following processing is performed through the trained neural network model: coding the content label of the information to obtain a coding vector of the content label of the corresponding information; decoding the coded vector to generate a label vector corresponding to the content label of the information; and combining the labels corresponding to the label vectors according to the sequence of the generated label vectors to obtain the content labels for recommendation.

With the above example in mind, the content tags of the information are predicted by a neural network model (e.g., a transducer) of the sequence-to-sequence to sequentially generate tag vectors corresponding to the content tags of the information, thereby obtaining content tags for recommendation from the generated tag vectors.

In some embodiments, the trained neural network model includes a plurality of cascaded encoders; encoding the content tag of the information to obtain an encoding vector of the content tag of the corresponding information, including: performing encoding processing of a first encoder on the content tag of the information through the first encoder of the plurality of cascaded encoders; outputting the encoding result of the first encoder to the encoders in the subsequent cascade so as to continuously perform encoding processing and output the encoding result in the encoders in the subsequent cascade until the encoding result is output to the last encoder; and taking the coding result output by the last coder as a coding vector of the content label of the corresponding information.

With the above example in mind, as shown in fig. 4, the trained neural network model includes N cascaded encoders, the 1 st encoder encodes the content tag of the information to obtain the 1 st encoding result, the 1 st encoding result is input to the 2 nd encoder, the 2 nd encoder encodes the 1 st encoding result to obtain the 2 nd encoding result, the encoding process and the encoding result output are sequentially performed, and the encoding result output by the last encoder is used as the encoding vector of the content tag of the corresponding information. Therefore, through cascade coding processing, the richer and more accurate characteristic information in the content label of the information is extracted, so that accurate decoding processing can be carried out later.

For example, the following processing is performed by the 1 st encoder: performing self-attention processing on the content label of the information to obtain a 1 st self-attention vector; carrying out residual connection processing on the 1 st self-attention vector and the content label of the information to obtain a 1 st residual vector; carrying out nonlinear mapping treatment on the 1 st residual vector to obtain a 1 st mapping vector; and carrying out residual connection processing on the 1 st mapping vector and the 1 st residual vector, taking the residual connection result as the coding result of the 1 st coder, and outputting the coding result of the 1 st coder to the 2 nd coder.

Continuing with the encoding process similar to that described above, the following is performed by the j-th encoder of the plurality of concatenated encoders: performing self-attention processing on the coding result of the j-1 th coder to obtain a j-th self-attention vector; carrying out residual connection processing on the j-th self-attention vector and the coding result of the j-1-th coder to obtain a j-th residual vector; nonlinear mapping processing is carried out on the jth residual vector through a feedforward network to obtain the jth mapping vector; carrying out residual connection processing on the jth mapping vector and the jth residual vector, taking the residual connection result as the coding result of the jth coder, and outputting the coding result of the jth coder to the (j+1) th coder; wherein j is more than or equal to 2 and less than or equal to N-1, j is a natural number, and N is the number of a plurality of cascaded encoders.

In some embodiments, decoding the encoded vector to generate a tag vector corresponding to a content tag of the information includes: decoding the coded vector to generate a 1 st tag vector corresponding to the content tag of the information; decoding processing is carried out based on the coding vector and the generated first i-1 label vectors, and an ith label vector corresponding to the content label of the information is generated; wherein, i is more than or equal to 2 and less than or equal to M, i is a natural number, and M is the total number of label vectors corresponding to the content labels of the information.

As shown in fig. 5, after the server obtains the encoded vector, the decoder in the trained neural network model decodes the encoded vector to generate the 1 st tag vector, and inputs the 1 st tag vector to the decoder, the decoder decodes the encoded vector and the 1 st tag vector to generate the 2 nd tag vector, and inputs the 2 nd tag vector to the decoder, and the decoder decodes the encoded vector, the 1 st tag vector and the 2 nd tag vector to generate the 3 rd tag vector, and continues the above processing until the end symbol is reached, so as to stop the above decoding process, wherein the M tag vectors do not include the end symbol. And sequentially generating tag vectors in a sequence-to-sequence mode, so that content tags are accurately generated, and accurate information recommendation is performed.

In some embodiments, the trained neural network model includes a plurality of cascaded decoders; decoding processing is carried out based on the coded vector and the generated first i-1 label vectors, and an ith label vector corresponding to the content label of the information is generated, and the decoding processing comprises the following steps: decoding processing is performed by a first decoder of the plurality of cascaded decoders based on the encoded vector and the generated first i-1 tag vectors; outputting the decoding result of the first decoder to the subsequent cascaded decoders to continue the decoding processing and the decoding result output in the subsequent cascaded decoders until the decoding result is output to the last decoder; the decoding result output by the last decoder is used as an ith tag vector corresponding to the content tag of the information.

As shown in fig. 5, the trained neural network model includes T cascaded encoders, and performs decoding processing based on the encoded vector and the generated first i-1 tag vector through the 1 st decoder to obtain a 1 st decoding result, inputs the 1 st decoding result to the 2 nd decoder, performs decoding processing on the 1 st decoding result through the 2 nd decoder to obtain a 2 nd decoding result, sequentially performs the decoding processing and the decoding result output, and uses the decoding result output by the last decoder as the i-th tag vector. Therefore, through cascade decoding processing, the richer and more accurate characteristic information in the content labels of the information is sequentially extracted, and accurate decoding processing is performed.

For example, the following processing is performed by the kth decoder of the plurality of cascaded decoders: mapping the decoding result of the (k-1) decoder to obtain a (k) mapping vector; carrying out fusion processing on the kth mapping vector and the coding vector to obtain a kth fusion vector; mapping the kth fusion vector, taking the obtained mapping result as a decoding result of the kth decoder, and outputting the decoding result of the kth decoder to the kth+1th decoder; wherein, k is more than or equal to 2 and less than or equal to T-1, k is a natural number, and T is the number of a plurality of cascaded decoders. The mapping process and the fusion process described above are specifically explained as follows:

for example, the decoding result of the kth-1 decoder is mapped by: performing self-attention processing on the decoding result of the kth-1 decoder to obtain a kth self-attention vector; and carrying out residual connection processing on the kth self-attention vector and the decoding result of the kth-1 decoder, and taking the residual connection result as the kth mapping vector.

For example, the fusion processing of the kth mapping vector and the encoding vector is realized by: performing multi-head attention processing on the kth mapping vector and the coding vector to obtain a kth multi-head attention vector; and carrying out residual connection processing on the kth multi-head attention vector and the kth mapping vector, and taking the residual connection result as the kth fusion vector.

For example, the mapping process is performed on the kth fusion vector by: nonlinear mapping processing is carried out on the kth fusion vector through a feedforward network, so as to obtain a kth mapping vector; and carrying out residual connection processing on the kth mapping vector and the kth fusion vector, and taking the residual connection result as a decoding result of the kth decoder.

In step 104, information matching with the content tag to be recommended is queried from the information base as recommendation information.

After the server obtains the content tags for recommendation, accurate recommendation information can be obtained through matching among the content tags, namely information recommendation is carried out through the low-dimension tags.

In some embodiments, querying information matching with the content tags to be recommended from the information base as recommendation information includes: traversing the information to be recommended in the information base based on the content label to be recommended to determine the information to be recommended including the content label to be recommended; screening the information to be recommended, which comprises the content label to be recommended, and taking the information to be recommended obtained by screening as recommendation information.

For example, after the server obtains the content tag for recommendation, the content tag included in the information to be recommended in the information base is traversed to determine the information to be recommended including the content tag to be recommended. After the information to be recommended is determined, the information to be recommended including the content tag to be recommended can be screened based on the portrait information of the target user or the information related to the recommendation factors such as the interaction information of friends, so that the screened information to be recommended is used as the recommendation information, for example, the latest preference of the target user is determined from the interaction information of friends, so that the preference of the target user for the information to be recommended is determined based on the latest preference of the target user, and the recommendation information is determined from the information to be recommended, so that accurate information recommendation is realized.

In step 105, a recommendation operation is performed in the information presentation page based on the recommendation information.

After the server obtains the recommendation information, the server may send the recommendation information to the terminal (running with a client, such as a video client, a news client, etc.), so as to display the recommendation information on an information display page of the terminal, thereby recommending information meeting the interests of the target user to the target user.

In the following, an exemplary application of the embodiments of the present application in an actual video recommendation application scenario will be described.

The embodiment of the application may be applied to various recommended application scenarios, as shown in fig. 1, the terminal 200 is connected to the server 100 deployed at the cloud end through the network 300, a video application is installed on the terminal 200, after a video recommendation request is obtained, an information recommendation interface of the server 100 is called, the server 100 determines a content label for recommendation according to a corresponding relationship between a content label of a video sample and a content label of a recommended video sample set, and performs video recommendation according to the content label for recommendation, so as to quickly respond to the video recommendation request, thereby recommending videos meeting the interests of a target user to the target user, and the target user can continuously browse the videos meeting the interests of the target user.

In the related art, video recommendation based on Graph Embedding (Graph Embedding): as shown in fig. 6, the operation of the user on the video is constructed by a graph, wherein the nodes of the graph are video identities (I D, identity Document), for example, the node 601 represents the ID of a certain video, then, as shown in fig. 7, an operation sequence (i.e., a sequence of videos) is constructed from the graph by means of random walk (random walk), finally, as shown in fig. 8, the video ID is embedded (embedding) to generate an embedded vector of the learned video ID, and the embedded vector of the learned video ID can be used for recommending recall.

In the related art, video recommendation based on a transducer: as shown in fig. 9, the operations of the users are divided by a sequence of consumption operations (sequence of consumption operations of the users in a certain period), provided that the video consumed by the users in a certain period is sequentially S ₁ ～S _N Will S ₁ ～S _N-1 As input of model, S ₂ ～S _N As an output of the model.

In order to fully learn the interaction information from the input sequence to the output sequence, S in the user session can be processed through an Algorithm for recommendation (BST, behavior Sequence Transformer) model ₁ ～S _N-1 The video is used as a user operation sequence, S < th) _N The video target videos are input into the BST model together for full fusion and interaction of information.

In summary, in the video recommendation method in the related art, the operation sequence of the user is used as a training sample, and finally the model is fitted to the operation sequence of the user on the video. Such recommendations have the following problems: 1) The video ID dimension is too high, and the model is difficult to converge; 2) The new video cannot be supported, and only the video which appears in the training sample can be supported.

In order to solve the above problem, the embodiment of the present application uses two models (for example, graph mapping and transformation) for fitting the user operation sequence as starting points, and proposes to fit the user operation sequence from the video content perspective, that is, to fit the user operation sequence from the video content tag perspective. The transfer relation between content labels of videos is mined by analyzing the average playing time length from the videos of the main feeds page to the floating layer videos, and the transfer relation is fitted through a trans former model, wherein the trained trans former model can be used for relevant label association prediction in the video floating layer, and further used for recommending scenes such as recall and interest exploration in the main feeds page.

Therefore, the following effects can be achieved by the embodiments of the present application: 1) Fitting user operation from the view of the content labels of the video, wherein the number of the content labels of the video is greatly reduced relative to the number of the video IDs (the number of the content labels of the video is in the range of ten thousands to hundred thousands, and the number of the video IDs is in the range of millions to hundred billions), so that the model is easier to train and can be converged better; 2) Each video can be described by a content tag of the video, for a new video, training of the new video can be supported only by the fact that the tag of the new video appears in the training set, and the content tag set of the video is very stable, so that the problem of cold start of the new video can be solved.

The embodiment of the application provides a video tag association recommendation mode applied to recommendation recall, which is applied to a recommendation recall stage, wherein training data from tag set to tag set is constructed through user consumption operation, and the association relation from tag set to tag set is learned for recommendation recall in a main feeds scene and a video floating layer scene. As shown in fig. 10, the video in the main feeds scene is presented, for example, as a recommended recall after the model gets the set of associated tags based on the previous video consumption operation of the user or based on the user's portrait tag as input. As shown in fig. 11, a video in a video floating layer scene is illustrated, after a user clicks on a video 1001 in a main feed, a page jumps to a video floating layer 1101, in which a tag of the main feed video can be input as a model, and an associated tag set is obtained and then used as a recommended recall, where a video 1102 corresponds to the video 1001.

According to the method and the device, a time length relation from video to video (namely, the video in the main feed page is converted into the video in the video suspension page and the time length of the video in the video suspension page) is mined from an operation sequence of a user, then the time length relation from the video to the video is converted into a time length relation from a tag set to a tag set according to the tag of the video (namely, the time length relation from the content tag of the video in the main feed page to the content tag of the video in the video suspension), and finally the conversion relation from the tag set to the tag set is fitted through a transducer model, and the whole algorithm flow is as follows:

Step 1, screening user consumption operation of a corresponding video floating layer from a user log, constructing a corresponding relation (pair) of videos according to the relation from the videos in a main feeds page to the videos of the video floating layer page, and carrying out average aggregation on the historical watching duration of all users on the floating layer video page from the pair dimension.

Step 2, connecting and aggregating videos on the pages according to the main feeds by the aggregated pair, so as to construct a sample similar to the following:

video_0|video_1:duration_1，video_2:duration_2，…，video_n:duration_n

where video_0 represents video appearing on the main feeds page, video_1, video_2, and vi deo_n are videos that are consumed on the incoming video floating layer page after the user clicks on video_0 on the main feeds page, and duration_1, duration_2, …, and duration_n are respectively average playing durations of the videos on the corresponding floating layer video (i.e., an aggregate average of historical viewing durations of the videos by all users).

Step 3, accumulating the average playing time length of the video on each content label of the video according to the content labels of the video, carrying out aggregation average according to the dimension of the content labels, sorting in descending order according to the average time length, and selecting the first N content labels as output to obtain the following samples:

input_tag_1，input_tag_2，…，input_tag_M|output_tag_1，output_tag_2，…，output_tag_N

The input_tag_1, input_tag_2, … and input_tag_m represent content tags of video in a main feeds page, and the output_tag_1, output_tag_2, … and output_tag_n represent N content tags selected from content tags corresponding to a video floating window.

Step 4, as shown in fig. 12, training samples by using a transducer model, where in the embodiment of the present application, the samples are trained by using a transducer, and the input samples are extracted from the content tag sets of the main feeds page, that is, input_tag_1, input_tag_2, …, and input_tag_m, and the output samples are extracted from the video floating layer page, and the first N tag sets, that is, output_tag_1, output_tag_2, …, and output_tag_n, are ordered according to the tag duration. Wherein the transducer model comprises a two-layer encoder, a self-attention (self-attention) module uses 4 heads (heads) and < s > represents a starter, for example, two content tags of "internal drama" and "xiao Song" are input to the transducer model, and the transducer model outputs "xiaobai" and "xiao Ju" in sequence. Wherein softmax represents an activation function.

And 5, outputting predicted content labels according to the input content labels through the trained transducer model, carrying out video recall, and carrying out video recommendation on the recalled video.

As shown in fig. 13, the application of the trained converter model in the embodiment of the present application to the viewpoint Video floating layer increases the Video play number (VV, video View) of the time period of the converter model in the embodiment of the present application by 1.5% and increases the duration by 0.5% relative to the time period of the converter model not trained in the embodiment of the present application.

The artificial intelligence based information recommendation method provided by the embodiments of the present application has been described so far in connection with exemplary applications and implementations of the server provided by the embodiments of the present application. In practical application, each functional module in the information recommendation device may be cooperatively implemented by hardware resources of an electronic device (such as a terminal device, a server or a server cluster), such as computing resources of a processor, communication resources (such as for supporting communications in various modes such as optical cables and cellular) and a memory. Fig. 2 shows an information recommendation device 555 stored in a memory 550, which may be software in the form of a program, a plug-in or the like, for example, software C/c++, a software module designed in a programming language such as Java, an implementation of an application software designed in a programming language such as C/c++, java, or a dedicated software module in a large software system, an application program interface, a plug-in, a cloud service, etc., and different implementations are exemplified below.

Example one, the information recommendation device is a mobile end application and module

The information recommending device 555 in the embodiment of the present application may be provided as a software module designed by using a programming language such as software C/c++, java, etc., and embedded into various mobile terminal applications (stored in a storage medium of a mobile terminal as executable instructions and executed by a processor of the mobile terminal), so as to directly use the computing resources of the mobile terminal to complete related information recommending tasks, and periodically or aperiodically transmit the processing results to a remote server through various network communication modes, or locally store the processing results in the mobile terminal.

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

The information recommending apparatus 555 in the embodiment of the present application may be provided as an application software designed by using programming languages such as C/c++, java, etc. or a dedicated software module in a large software system, and run on a server side (stored in a storage medium of the server side in a manner of executable instructions and run by a processor of the server side), where the server uses its own computing resources to complete related information recommending tasks.

The embodiment of the application can also be provided for carrying a customized and easy-to-interact network (Web) Interface or other User Interfaces (UI) on a distributed and parallel computing platform formed by a plurality of servers to form an information recommendation platform (used for recommendation lists) for individuals, groups or units, and the like.

Example three information recommendation device is a server side application program interface (API, application Program Interface) and plug-in

The information recommending device 555 in the embodiment of the present application may be provided as an API or a plug-in at the server side, so as to be called by a user to execute the information recommending method based on artificial intelligence in the embodiment of the present application, and be embedded into various application programs.

Fourth example, the information recommendation device is a mobile device client API and plug-in

The information recommending device 555 in the embodiment of the application can be provided as an API or a plug-in on the mobile device side for a user to call to execute the information recommending method based on artificial intelligence in the embodiment of the application.

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

The information recommending device 555 in the embodiment of the present application may provide an information recommending cloud service developed for a user, for a person, a group or a unit to obtain a recommending list.

The information recommending apparatus 555 includes a series of modules, including an acquisition module 5551, a determination module 5552, a processing module 5553, a matching module 5554, and a traversing module 5555. The following continues to describe a scheme for implementing information recommendation by matching each module in the information recommendation device 555 provided in the embodiment of the present application.

An acquisition module 5551, configured to acquire a content tag of information displayed in the information display page; a determining module 5552, configured to determine a correspondence between a content tag of an information sample and a content tag of a recommended information sample set based on the correspondence between the information sample and the recommended information sample set and operation data of the recommended information sample set; a processing module 5553, configured to determine a content tag for recommendation based on the correspondence and the content tag of the information; the matching module 5554 is configured to query information matched with the content tag to be recommended from an information base, so as to serve as recommendation information; and executing recommendation operation in the information display page based on the recommendation information.

In some embodiments, the information recommendation device 555 further includes: the traversing module is used for traversing the historical operation sequences of at least one sample user to obtain the corresponding relation between the recommended information sample set and the information samples received by each sample user and the operation data of the recommended information samples in the recommended information sample set; the determining module 5552 is further configured to perform the following processing for any recommended information sample in the recommended information sample set: based on the operation data of the recommended information samples of each sample user, carrying out aggregation average processing on the recommended information samples based on the operation data, and taking the obtained aggregation average result as the operation characteristics of the recommended information samples; screening content tags included in each recommended information sample based on the operation characteristics of each recommended information sample; and constructing a corresponding relation between the content label of the information sample and the content label remained after the screening processing, and taking the corresponding relation as the corresponding relation between the content label of the information sample and the content label of the recommended information sample set.

In some embodiments, the determining module 5552 is further configured to perform a deduplication process on content tags that are included in each of the recommended information samples, to obtain a plurality of content tags that are used for performing the filtering process; the following processing is performed for any one of the plurality of content tags for performing the filtering processing: when the recommended information sample comprises the content tag, carrying out aggregation average processing on the content tag based on the operation characteristics of the recommended information sample, and taking the obtained aggregation average result of the content tag as the operation characteristics of the content tag; and based on the operation characteristics of each content label, carrying out descending order sorting on each content label, and taking a plurality of content labels sorted in the descending order sorting result as the content labels remained after the screening processing.

In some embodiments, the determining module 5552 is further configured to perform feature extraction processing on the portrait information of the target user to obtain a weight of each content tag; and weighting the obtained aggregate average result of the content tags based on the weight of each content tag, and taking the weighted result as the operation characteristic of the content tag.

In some embodiments, the processing module 5553 is further configured to query a data table including the correspondence with a content tag of the information as an index; when the content label of the information sample matched with the content label of the information is queried, the content label of the recommended information sample set corresponding to the content label of the information sample is used as the content label to be recommended.

In some embodiments, the processing module 5553 is further configured to train a neural network model for information recommendation based on a correspondence between content tags of the information samples and content tags of the recommended information sample set, to obtain a trained neural network model; and carrying out label-based prediction processing on the content labels of the information through the trained neural network model to obtain the content labels to be recommended.

In some embodiments, the processing module 5553 is further configured to perform a label-based prediction process on the information sample through the neural network model, to obtain a prediction label for recommendation; constructing a loss function of the neural network model based on the prediction tag and the content tag of the recommended information sample set; updating parameters of the neural network model until the loss function converges, and taking the updated parameters of the neural network model when the loss function converges as the parameters of the neural network model after training.

In some embodiments, the processing module 5553 is further configured to perform the following processing by the trained neural network model: coding the content label of the information to obtain a coding vector corresponding to the content label of the information; decoding the coded vector to generate a label vector corresponding to the content label of the information; and combining the labels corresponding to the label vectors according to the sequence of the generated label vectors to obtain the content labels for recommendation.

In some embodiments, the trained neural network model includes a plurality of cascaded encoders; the processing module 5553 is further configured to perform, by a first encoder of the plurality of cascaded encoders, an encoding process of the first encoder on a content tag of the information; outputting the coding result of the first coder to the coder of the subsequent cascade so as to continuously carry out coding processing and coding result output in the coder of the subsequent cascade until the coding result is output to the last coder; and taking the coding result output by the last coder as a coding vector of a content label corresponding to the information.

In some embodiments, the processing module 5553 is further configured to perform, by a j-th encoder of the plurality of concatenated encoders, the following: performing self-attention processing on the coding result of the j-1 th coder to obtain a j-th self-attention vector; carrying out residual connection processing on the j-th self-attention vector and the coding result of the j-1-th coder to obtain a j-th residual vector; performing nonlinear mapping processing on the jth residual vector to obtain a jth mapping vector; carrying out residual connection processing on the jth mapping vector and the jth residual vector, taking a residual connection result as a coding result of the jth coder, and outputting the coding result of the jth coder to a (j+1) th coder; wherein j is more than or equal to 2 and less than or equal to N-1, j is a natural number, and N is the number of the plurality of cascaded encoders.

In some embodiments, the processing module 5553 is further configured to perform a decoding process on the encoded vector, and generate a 1 st tag vector corresponding to a content tag of the information; decoding processing is carried out based on the coding vector and the generated first i-1 tag vectors, and an ith tag vector corresponding to the content tag of the information is generated; wherein, i is more than or equal to 2 and less than or equal to M, i is a natural number, and M is the total number of label vectors corresponding to the content labels of the information.

In some embodiments, the trained neural network model includes a plurality of cascaded decoders; the processing module 5553 is further configured to perform, by a first decoder of the plurality of cascaded decoders, decoding processing based on the encoded vector and the generated first i-1 tag vectors; outputting the decoding result of the first decoder to a subsequent cascade decoder so as to continuously perform decoding processing and outputting the decoding result in the subsequent cascade decoder until outputting to a last decoder; and taking the decoding result output by the last decoder as an ith tag vector corresponding to the content tag of the information.

In some embodiments, the processing module 5553 is further configured to perform, by a kth decoder of the plurality of concatenated decoders, the following: mapping the decoding result of the (k-1) decoder to obtain a (k) mapping vector; carrying out fusion processing on the kth mapping vector and the coding vector to obtain a kth fusion vector; mapping the kth fusion vector, taking the obtained mapping result as a decoding result of the kth decoder, and outputting the decoding result of the kth decoder to a (k+1) th decoder; wherein k is more than or equal to 2 and less than or equal to T-1, k is a natural number, and T is the number of the plurality of cascaded decoders.

In some embodiments, the processing module 5553 is further configured to perform self-attention processing on the decoding result of the kth-1 decoder to obtain a kth self-attention vector; and carrying out residual connection processing on the kth self-attention vector and the decoding result of the kth-1 decoder, and taking the residual connection result as a kth mapping vector.

In some embodiments, the processing module 5553 is further configured to perform multi-head attention processing on the kth mapping vector and the encoding vector to obtain a kth multi-head attention vector; and carrying out residual connection processing on the kth multi-head attention vector and the kth mapping vector, and taking a residual connection result as a kth fusion vector.

In some embodiments, the processing module 5553 is further configured to perform a nonlinear mapping process on the kth fusion vector to obtain a kth mapping vector; and carrying out residual connection processing on the kth mapping vector and the kth fusion vector, and taking a residual connection result as a decoding result of the kth decoder.

In some embodiments, the matching module 5554 is further configured to perform traversal processing on the information to be recommended in the information base based on the content tag to be recommended, so as to determine the information to be recommended including the content tag to be recommended; and screening the information to be recommended, which comprises the content label to be recommended, and taking the information to be recommended obtained by screening as recommendation information.

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

Embodiments of the present application provide a computer readable storage medium having stored therein executable instructions that, when executed by a processor, cause the processor to perform the artificial intelligence based information recommendation method provided by embodiments of the present application, for example, as shown in fig. 3A-3C.

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

In some embodiments, the executable instructions may be in the form of programs, software modules, scripts, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and they may be deployed in any form, including as stand-alone programs or as modules, components, subroutines, or other units suitable for use in a computing environment.

As an example, the executable instructions may, but need not, correspond to files in a file system, may be stored as part of a file that holds other programs or data, such as in one or more scripts in a hypertext markup language (html, hyper Text Markup Language) 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, alternatively, distributed across multiple sites and interconnected by a communication network.

The foregoing is merely exemplary embodiments of the present application and is not intended to limit the scope of the present application. Any modifications, equivalent substitutions, improvements, etc. that are within the spirit and scope of the present application are intended to be included within the scope of the present application.

Claims

1. An artificial intelligence based information recommendation method, comprising:

acquiring a content tag of information displayed in an information display page according to the information recommendation request;

Traversing the historical operation sequence of at least one sample user to obtain a corresponding relation between a recommended information sample set and an information sample received by each sample user and operation data of the recommended information sample in the recommended information sample set, wherein the historical operation sequence represents a set of operation data generated by interaction between the sample user and the information display page;

based on the operation data of the recommended information samples of each sample user, carrying out aggregation average processing on the recommended information samples based on the operation data, and taking the obtained aggregation average result as the operation characteristics of the recommended information samples;

performing de-duplication processing on the content labels respectively included in each recommended information sample to obtain a plurality of content labels for screening processing;

the following processing is performed for any one of the plurality of content tags for performing the filtering processing:

when the recommended information sample comprises the content tag, carrying out aggregation average processing on the content tag based on the operation characteristics of the recommended information sample to obtain an aggregation average result of the content tag;

Performing feature extraction processing on the portrait information of the target user to obtain the weight of each content tag;

weighting the obtained aggregation average result of the content tags based on the weight of each content tag, and taking the weighted result as the operation characteristic of the content tag;

based on the operation characteristics of each content label, carrying out descending order sorting on each content label, and taking a plurality of content labels sorted in the descending order sorting result as the content labels remained after the screening processing;

constructing a corresponding relation between the content label of the information sample and the content label remained after the screening process, and taking the corresponding relation as the corresponding relation between the content label of the information sample and the content label of the recommended information sample set;

2. The method of claim 1, wherein the determining a content tag for a recommendation based on the correspondence and the content tag of the information comprises:

Querying a data table comprising the corresponding relation by taking the content label of the information as an index;

when the content label of the information sample matched with the content label of the information is queried, the content label of the recommended information sample set corresponding to the content label of the information sample is used as the content label to be recommended.

3. The method of claim 1, wherein the determining a content tag for a recommendation based on the correspondence and the content tag of the information comprises:

4. The method of claim 3, wherein training the neural network model for information recommendation based on the correspondence between the content tags of the information samples and the content tags of the recommended information sample set to obtain a trained neural network model comprises:

Performing label-based prediction processing on the information sample through the neural network model to obtain a prediction label for recommendation;

constructing a loss function of the neural network model based on the prediction tag and the content tag of the recommended information sample set;

updating parameters of the neural network model until the loss function converges, and taking the updated parameters of the neural network model when the loss function converges as the parameters of the neural network model after training.

5. The method of claim 3, wherein the performing, by the trained neural network model, tag-based prediction processing on the content tags of the information to obtain content tags for recommendation, includes:

executing the following processing through the trained neural network model:

6. The method of claim 5, wherein the step of determining the position of the probe is performed,

the trained neural network model comprises a plurality of cascaded encoders;

the encoding processing is performed on the content label of the information to obtain an encoding vector corresponding to the content label of the information, and the encoding method comprises the following steps:

performing, by a first encoder of the plurality of concatenated encoders, encoding processing of the first encoder on a content tag of the information;

outputting the coding result of the first coder to the coder of the subsequent cascade so as to continuously carry out coding processing and coding result output in the coder of the subsequent cascade until the coding result is output to the last coder;

and taking the coding result output by the last coder as a coding vector of a content label corresponding to the information.

7. The method of claim 6, wherein continuing the encoding process and the encoding result output in the subsequent concatenated encoder comprises:

performing, by a j-th encoder of the plurality of concatenated encoders, the following:

performing self-attention processing on the coding result of the j-1 th coder to obtain a j-th self-attention vector;

Carrying out residual connection processing on the j-th self-attention vector and the coding result of the j-1-th coder to obtain a j-th residual vector;

performing nonlinear mapping processing on the jth residual vector to obtain a jth mapping vector;

carrying out residual connection processing on the jth mapping vector and the jth residual vector, taking a residual connection result as a coding result of the jth coder, and outputting the coding result of the jth coder to a (j+1) th coder;

wherein j is more than or equal to 2 and less than or equal to N-1, j is a natural number, and N is the number of the plurality of cascaded encoders.

8. The method of claim 5, wherein decoding the encoded vector to generate a tag vector corresponding to a content tag of the information, comprises:

9. The method according to claim 1, wherein the querying information matching the content tag to be recommended from the information base as recommendation information includes:

traversing the information to be recommended in the information base based on the content label to be recommended to determine the information to be recommended including the content label to be recommended;

and screening the information to be recommended, which comprises the content label to be recommended, and taking the information to be recommended obtained by screening as recommendation information.

10. An information recommendation device, characterized in that the device comprises:

the acquisition module is used for acquiring the content tag of the information displayed in the information display page according to the information recommendation request;

the determining module is used for performing traversal processing on a historical operation sequence of at least one sample user to obtain a corresponding relation between a recommended information sample set and an information sample received by each sample user and operation data of the recommended information sample in the recommended information sample set, wherein the historical operation sequence represents a set of operation data generated by interaction between the sample user and the information display page; based on the operation data of the recommended information samples of each sample user, carrying out aggregation average processing on the recommended information samples based on the operation data, and taking the obtained aggregation average result as the operation characteristics of the recommended information samples; performing de-duplication processing on the content labels respectively included in each recommended information sample to obtain a plurality of content labels for screening processing; the following processing is performed for any one of the plurality of content tags for performing the filtering processing: when the recommended information sample comprises the content tag, carrying out aggregation average processing on the content tag based on the operation characteristics of the recommended information sample to obtain an aggregation average result of the content tag; performing feature extraction processing on the portrait information of the target user to obtain the weight of each content tag; weighting the obtained aggregation average result of the content tags based on the weight of each content tag, and taking the weighted result as the operation characteristic of the content tag; based on the operation characteristics of each content label, carrying out descending order sorting on each content label, and taking a plurality of content labels sorted in the descending order sorting result as the content labels remained after the screening processing; constructing a corresponding relation between the content label of the information sample and the content label remained after the screening process, and taking the corresponding relation as the corresponding relation between the content label of the information sample and the content label of the recommended information sample set;

11. An electronic device, the electronic device comprising:

a memory for storing executable instructions;

a processor for implementing the artificial intelligence based information recommendation method of any one of claims 1 to 9 when executing executable instructions stored in the memory.

12. A computer readable storage medium storing executable instructions for implementing the artificial intelligence based information recommendation method of any one of claims 1 to 9 when executed by a processor.