CN116150464A

CN116150464A - Media information recommendation method, device, apparatus, program and storage medium

Info

Publication number: CN116150464A
Application number: CN202111347189.8A
Authority: CN
Inventors: 陈昊
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2021-11-15
Filing date: 2021-11-15
Publication date: 2023-05-23

Abstract

The invention provides a media information recommendation method, a device and electronic equipment based on a recommendation model, and the related embodiments can be applied to various scenes such as cloud technology, cloud security, intelligent traffic and the like, and the method comprises the following steps: acquiring domain features of media information corresponding to original features and target objects; performing feature mapping on the original features through the center sub-model to obtain center features of the original features under a target semantic space; performing feature mapping on the domain features under the corresponding media information domain through each peripheral sub-model to obtain peripheral features of the domain features under the target semantic space; determining, by the determining sub-model, a target media information domain from at least two of the media information domains; therefore, the content recommendation of the media information domain in the cold start state of the target object can be realized by taking the target media information domain as a recommendation reference, the accuracy and timeliness of the media information recommendation in the cold start state are enhanced, and the quality of the media information recommendation is effectively improved.

Description

Media information recommendation method, device, apparatus, program and storage medium

Technical Field

The present invention relates to information processing technology, and in particular, to a media information recommendation method, apparatus, electronic device, computer program product, and storage medium based on a recommendation model, so that the applicable fields of the present invention include, but are not limited to, fields of automatic driving, internet of vehicles, intelligent transportation, and the like.

Background

Artificial intelligence (AI, artificial Intelligence) is a comprehensive technology of computer science, and by researching the design principle and implementation method of various intelligent machines, the machines have the functions of sensing, reasoning and deciding. Artificial intelligence technology is a comprehensive discipline, and is widely related to fields, such as natural language processing technology, machine learning/deep learning and other directions, and it is believed that with the development of technology, the artificial intelligence technology will be applied in more fields and become more and more valuable.

In the related art, when recommending media information to a user, content recommendation is generally performed based on a content history interaction record of an object to be recommended, or a user object having similar history behavior with the object to be recommended in the same system is searched, and the interaction content of the user object is used as recommendation content to recommend to the user to be recommended, however, in the case of cold start, there is no or only few user-content history interaction records in the system, and accurate media information recommendation is difficult to realize.

Disclosure of Invention

In view of the above, embodiments of the present invention provide a media information recommendation method, apparatus, electronic device, software program, and storage medium based on a recommendation model, which can enhance accuracy and timeliness of media information recommendation in a cold start state, effectively improve quality of media information recommendation, and improve user experience.

The embodiment of the invention provides a media information recommendation method based on a recommendation model, wherein the recommendation model comprises a center sub-model, a determination sub-model and at least two peripheral sub-models, each peripheral sub-model corresponds to one media information domain, and the method comprises the following steps:

acquiring original characteristics of media information for representing preference of a target object, and acquiring domain characteristics of the media information corresponding to the target object under each media information domain;

performing feature mapping on the original features through the center sub-model to obtain center features of the original features under a target semantic space;

performing feature mapping on the domain features under the corresponding media information domain through each peripheral sub-model to obtain peripheral features of the domain features under the target semantic space;

Determining a target media information domain from at least two media information domains based on the central feature and peripheral features corresponding to each media information domain through the determination sub-model;

and recommending the content of the media information domain in the cold start state to the target object by taking the target media information domain as a recommendation reference.

The embodiment of the invention also provides a media information recommending device based on a recommending model, wherein the recommending model comprises a center sub-model, a determining sub-model and at least two peripheral sub-models, each peripheral sub-model corresponds to one media information field, and the device comprises:

the information transmission module is used for acquiring original characteristics of media information for representing the preference of a target object and acquiring domain characteristics of the media information corresponding to the target object under each media information domain;

the information processing module is used for acquiring original characteristics of media information for representing the preference of a target object and acquiring domain characteristics of the media information corresponding to the target object under each media information domain;

the information processing module is used for carrying out feature mapping on the original features through the center sub-model to obtain center features of the original features under a target semantic space;

The information processing module is used for carrying out feature mapping on the domain features under the corresponding media information domain through each peripheral sub-model to obtain peripheral features of the domain features under the target semantic space;

the information processing module is used for determining a target media information domain from at least two media information domains based on the central characteristics and peripheral characteristics corresponding to each media information domain through the determination sub-model;

and the information processing module is used for recommending the content of the media information domain in the cold start state to the target object by taking the target media information domain as a recommendation reference.

In the above scheme, the information processing module is configured to match the central feature with each peripheral feature through the determination sub-model, so as to obtain a matching degree between the central feature and each peripheral feature;

and the information processing module is used for determining the media information domain corresponding to the peripheral feature with the largest matching degree as the target media information domain.

In the above solution, the information processing module is configured to obtain a media information category preferred by the target object in the target media information domain;

The information processing module is used for recommending the media information of the same category to the target object based on the media information category and the media information field in the cold start state.

In the above scheme, the information processing module is configured to obtain historical query operation information of the target object in an upstream media information domain; wherein the upstream media information field comprises a plurality of the media information fields;

the information processing module is used for determining media information triggered by the target object in a plurality of media information used for responding to the historical query operation information based on the historical query operation information;

the information processing module is used for extracting features of the historical query operation information to obtain corresponding query features, and extracting features of the media information triggered by the target object to obtain corresponding media features;

and the information processing module is used for splicing the query feature and the media feature to obtain the original feature of the media information for representing the preference of the target object.

In the above scheme, the information processing module is configured to, when the number of media information triggered by the target object is multiple, classify multiple media information into different media information domains, and obtain a classification result;

The information processing module is used for determining the quantity of the media information included in each media information domain according to the classification result;

the information processing module is used for taking the media information included in the media information fields with the quantity reaching the quantity threshold as target media information;

the information processing module is used for extracting the characteristics of the target media information in the media information triggered by the target object to obtain corresponding media characteristics.

In the above solution, the information processing module is configured to obtain, when the media information included in the media information field is an article, a title of the article, a category to which the article belongs, and a named entity in the article;

the information processing module is used for extracting the characteristics of the title, the category and the named entity respectively to obtain corresponding title characteristics, category characteristics and entity characteristics;

and the information processing module is used for splicing the title features, the category features and the entity features to obtain domain features of the articles.

In the above scheme, the information processing module is configured to obtain, when the media information included in the media information field is a video, a title of the media information, a category to which the media information belongs, and description information of the media information;

The information processing module is used for extracting the characteristics of the title, the category and the description information respectively to obtain corresponding title characteristics, category characteristics and description characteristics;

and the information processing module is used for splicing the title features, the category features and the description features to obtain domain features of the media information.

In the above scheme, the information processing module is configured to obtain, when the media information included in the media information field is an application, a title of the media information, a category to which the media information belongs, and description information of the media information through a download record of the media information;

In the above scheme, the information processing module is configured to determine a constraint condition corresponding to the target semantic space;

the information processing module is used for carrying out feature mapping on the original features according to the constraint conditions through the center sub-model to obtain a center feature vector;

The information processing module is used for carrying out weighting processing on the central feature vector to obtain the central feature of the original feature under the target semantic space.

In the above scheme, the information processing module is configured to trigger a corresponding peripheral sub-model according to the type of the media information domain;

the information processing module is used for determining constraint conditions corresponding to the target semantic space;

and the information processing module is used for carrying out feature mapping on the domain features according to the constraint conditions through the peripheral submodel to obtain peripheral feature vectors.

In the above solution, the information processing module is configured to determine a training target object in a training sample set of the recommendation model based on an intersection object of the original feature and the domain feature;

the information processing module is used for acquiring corresponding original features and domain features according to the training target object and the actual matching probability between each original feature and domain feature in the training sample set;

the information processing module is used for inputting the original features into the central sub-model, inputting the domain features into the peripheral sub-model and obtaining the prediction matching probability between the original features and the domain features of the same training target object in the training sample set output by the determination sub-model;

The information processing module is used for comparing the predicted matching probability with the actual matching probability, and if the predicted matching probability is inconsistent with the actual matching probability, the model parameters of different sub-models in the recommended model are adjusted until the predicted matching probability between the original feature and the domain feature is the same as the actual matching probability.

The embodiment of the invention also provides electronic equipment, which comprises:

a memory for storing executable instructions;

and the processor is used for realizing the media information recommendation method based on the recommendation model when the executable instructions stored in the memory are operated.

The embodiment of the invention also provides a computer program product, which comprises a computer program or instructions, and is characterized in that the computer program or instructions realize the media information recommendation method based on the recommendation model when being executed by a processor.

The embodiment of the invention also provides a computer readable storage medium which stores executable instructions which are executed by a processor to realize the media information recommendation method based on the recommendation model.

The embodiment of the invention has the following beneficial effects:

the method comprises the steps of obtaining original characteristics of media information for representing preferences of a target object, and obtaining domain characteristics of the media information corresponding to the target object under each media information domain; performing feature mapping on the original features through the center sub-model to obtain center features of the original features under a target semantic space; performing feature mapping on the domain features under the corresponding media information domain through each peripheral sub-model to obtain peripheral features of the domain features under the target semantic space; determining a target media information domain from at least two media information domains based on the central feature and peripheral features corresponding to each media information domain through the determination sub-model; therefore, the content recommendation of the media information domain in the cold start state of the target object can be realized by taking the target media information domain as a recommendation reference, the accuracy and timeliness of the media information recommendation in the cold start state are enhanced, the quality of the media information recommendation is effectively improved, and the use experience of a user is improved.

Drawings

FIG. 1 is a schematic view of a usage scenario of a recommendation model-based media information recommendation method according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a composition structure of a media information recommendation device based on a recommendation model according to an embodiment of the present invention;

FIG. 3 is a schematic flow chart of an alternative media information recommendation method based on a recommendation model according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of a recommendation model according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of the original feature acquisition in an embodiment of the present invention;

FIG. 6 is a schematic diagram of a process for acquiring peripheral and central features according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of an alternative model structure for determining a target media information field according to an embodiment of the present invention;

fig. 8 is a schematic flow chart of an alternative method for training a recommendation model according to an embodiment of the present invention.

Detailed Description

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

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

Before describing embodiments of the present invention in further detail, the terms and terminology involved in the embodiments of the present invention will be described, and the terms and terminology involved in the embodiments of the present invention will be used in the following explanation.

1) In response to a condition or state that is used to represent the condition or state upon which the performed operation depends, the performed operation or operations may be in real-time or with a set delay when the condition or state upon which it depends is satisfied; without being specifically described, there is no limitation in the execution sequence of the plurality of operations performed.

2) Based on the conditions or states that are used to represent the operations that are being performed, one or more of the operations that are being performed may be in real-time or with a set delay when the conditions or states that are being relied upon are satisfied; without being specifically described, there is no limitation in the execution sequence of the plurality of operations performed.

3) softmax: very common and important functions in machine learning are widely used, especially in multi-class scenarios, mapping some inputs to real numbers between 0-1, and normalizing the guaranteed sum to 1.

4) Neural Networks (NN): an artificial neural network (Artificial Neural Network, ANN), abbreviated as neural network or neural-like network, is a mathematical or computational model that mimics the structure and function of biological neural networks (the central nervous system of animals, particularly the brain) for estimating or approximating functions in the field of machine learning and cognitive sciences.

5) Multitasking learning: in the field of machine Learning, better model accuracy than a single task can be achieved by simultaneously performing Joint Learning and optimization on a plurality of related tasks, the plurality of tasks assist each other by sharing a presentation layer, and the training method is called Multi-task Learning (Joint Learning).

6) Time-effectiveness of media information: the content of the media information has an effect over a period of time, the effect being measured by means of the user's interest level in the media information. During a period of time, the user is interested in a certain type of media information, so that the time-effectiveness plays an important role in user retention, clicking and CTR on the side line of the opposite end, and the content is pushed to the user in the time-effectiveness period range of the content to play a positive role, otherwise, the user is disliked.

The embodiment of the invention can be realized by combining Cloud technology, wherein Cloud technology (Cloud technology) refers to a hosting technology for integrating hardware, software, network and other series resources in a wide area network or a local area network to realize calculation, storage, processing and sharing of data, and can also be understood as the general term of network technology, information technology, integration technology, management platform technology, application technology and the like applied based on a Cloud computing business model. Background services of technical network systems require a large amount of computing and storage resources, such as video websites, picture websites and more portal websites, so cloud technologies need to be supported by cloud computing.

By means of the cloud technology, the media information recommending method based on the recommending model can take the target media information domain as a recommending reference, the target media information domain is recorded in the corresponding cloud server, when the target object browses media information in different terminals, corresponding media information can be recommended to the target object through the target media information domain which is stored in the cloud server and is taken as the recommending reference, and accordingly the target object can obtain a more accurate media information recommending result.

It should be noted that cloud computing is a computing mode, which distributes computing tasks on a resource pool formed by a large number of computers, so that various application systems can acquire computing power, storage space and information service as required. The network that provides the resources is referred to as the "cloud". Resources in the cloud are infinitely expandable in the sense of users, and can be acquired at any time, used as needed, expanded at any time and paid for use as needed. As a basic capability provider of cloud computing, a cloud computing resource pool platform, referred to as a cloud platform for short, is generally called infrastructure as a service (IaaS, infrastructure as a Service), and multiple types of virtual resources are deployed in the resource pool for external clients to select for use. The cloud computing resource pool mainly comprises: computing devices (which may be virtualized machines, including operating systems), storage devices, and network devices.

7) Media information, various forms of information available in the internet, such as advertisement information presented in a client or intelligent device, video files, media information to be recommended, news information, and the like.

Before introducing the media information recommendation method based on the recommendation model provided by the application, firstly, the defect of media information recommendation in the related technology is briefly described, and when the related technology performs media information recommendation, the following modes can be adopted:

1) Media information recommendation is carried out based on collaborative filtering recommendation technology, and the method comprises the following specific steps: through the historical interaction behavior record of the user in the system in the past, other users with similar historical behaviors are found out from the system, and the content interacted by the users is used as the most relevant content; or other contents with similar interaction records are found out, the contents are used as the most relevant contents, and the most relevant contents are displayed to the target object. When a user or content lacks a history interaction record, a cold start problem occurs; for recommendations of new users or new content, accurate recommendations in a cold start environment cannot be achieved because the history of interactions is almost zero.

2) Media information recommendation based on the description information is specifically expressed as follows: and characterizing the description information of the content through feature engineering, and recommending the similar content to the user according to the similarity among the features. Because the modeling object of this type of technology is a feature of the content, and the feature is usually a carrier describing that things have some commonality, even if there is no historical interaction information between the user and the content, the content can still be represented well, and this way has the following drawbacks: descriptive information typically comes from the age, sex, territory, etc. of the user, from which the user population can be divided; for user personalized characterization, these features are insufficient to capture the interests of the actual user individual. The method only takes the interaction history of the user and the content as a label, models the characteristics of the user, and ignores the relationship between the user and the content. Meanwhile, timeliness of the description information is generally limited to an initial stage of entering a system by a user, and as the user processes the media information, interest of the user is continuously changed, and recommendation of the media information based on the description information cannot be timely adjusted according to the change of the interest of the user.

In order to solve the above drawbacks in media information recommendation, the present application provides a media information recommendation method based on a recommendation model, referring to fig. 1, fig. 1 is a schematic view of usage scenario of a media information processing method based on a recommendation model provided in an embodiment of the present invention, referring to fig. 1, a terminal (including a terminal 10-1 and a terminal 10-2) is provided with a client capable of displaying software corresponding to different media information, such as a client or a plug-in for video playing, and a user may obtain and display different media information (such as different short video information or news information) through the corresponding client; the terminal is connected to the server 200 through the network 300, and the network 300 may be a wide area network or a local area network, or a combination of the two, and uses a wireless link to implement data transmission. The server 200 may also be a node in a blockchain or a node in a cloud network to implement storing the target media information domain as a recommendation reference.

As an example, the server 200 is configured to lay a corresponding recommendation model to implement the recommendation model-based media information recommendation method provided by the present invention, and when the recommendation model-based media information recommendation method is executed, obtain original features of media information for characterizing preferences of a target object, and obtain domain features of media information corresponding to the target object under each media information domain; performing feature mapping on the original features through the center sub-model to obtain center features of the original features under a target semantic space; performing feature mapping on the domain features under the corresponding media information domain through each peripheral sub-model to obtain peripheral features of the domain features under the target semantic space; determining a target media information domain from at least two media information domains based on the central feature and peripheral features corresponding to each media information domain through the determination sub-model; and recommending the content of the media information domain in the cold start state to the target object by taking the target media information domain as a recommendation reference, and displaying and outputting media information to be recommended, which is matched with the target object, through a terminal (terminal 10-1 and/or terminal 10-2). Taking short video information as an example, the recommendation model provided by the invention can be applied to short video playing, different short video information with different data sources can be processed in the short video playing, and finally different media information corresponding to the different media information is presented on the user interface UI (User Interface), so that the accuracy and timeliness of the recommended media information directly influence the user experience. Taking media information as short video as an example, a background database of video playing receives a large amount of video data from different sources every day, in the method for recommending media information based on a recommendation model provided by the application, in a cold start environment, the short video matched with a target object can be determined to be the short video recommended to the target object by taking the target media information field as a recommendation reference, further, the obtained short video recommended to the media information with the target object can be called by other application programs (for example, the recommendation result of the short video recommendation process is migrated to a long video recommendation process or a news recommendation process), and of course, the recommendation model matched with the corresponding target object can be migrated to a different video recommendation process (for example, a web page video recommendation process, an applet video recommendation process or a video recommendation process of a long video client).

The media information recommendation method based on the recommendation model provided by the embodiment of the application is realized based on artificial intelligence, wherein the artificial intelligence (Artificial Intelligence, AI) is a theory, a method, a technology and an application system which simulate, extend and extend human intelligence by using a digital computer or a machine controlled by the digital computer, sense environment, acquire knowledge and acquire an optimal result by using the knowledge. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision.

The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.

In the embodiments of the present application, the mainly related artificial intelligence software technology includes the above-mentioned speech processing technology, machine learning, and other directions. For example, speech recognition techniques (Automatic Speech Recognition, ASR) in Speech technology (Speech Technology) may be involved, including Speech signal preprocessing (Speech signal preprocessing), speech signal frequency domain analysis (Speech signal frequency analyzing), speech signal feature extraction (Speech signal feature extraction), speech signal feature matching/recognition (Speech signal feature matching/recognition), training of Speech (Speech training), and the like.

For example, machine Learning (ML) may be involved, which is a multi-domain interdisciplinary, involving multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It is specially studied how a computer simulates or implements learning behavior of a human to acquire new knowledge or skills, and reorganizes existing knowledge structures to continuously improve own performance. Machine learning is the core of artificial intelligence, a fundamental approach to letting computers have intelligence, which is applied throughout various areas of artificial intelligence. Machine Learning typically includes Deep Learning (Deep Learning) techniques, including artificial neural networks (artificial neural network), such as convolutional neural networks (Convolutional Neural Network, CNN), recurrent neural networks (Recurrent Neural Network, RNN), deep neural networks (Deep neural network, DNN), and the like.

It can be appreciated that the media information recommendation method and the voice processing based on the recommendation model provided by the application can be applied to an intelligent device (Intelligent device), and the intelligent device can be any device with an information display function, for example, an intelligent terminal, an intelligent home device (such as an intelligent sound box and an intelligent washing machine), an intelligent wearable device (such as an intelligent watch), a vehicle-mounted intelligent central control system (for displaying media information to a user through a small program for executing different tasks), or an AI intelligent medical device (for displaying treatment cases through displaying media information), and the like.

The following describes in detail the structure of the recommendation model-based media information recommendation device according to the embodiment of the present invention, and the recommendation model-based media information recommendation device may be implemented in various forms, such as a dedicated terminal with a recommendation model-based media information recommendation processing function, or a server provided with a recommendation model-based media information recommendation device processing function, such as the server 200 in fig. 1. Fig. 2 is a schematic diagram of a composition structure of a media information recommendation device based on a recommendation model according to an embodiment of the present invention, and it can be understood that fig. 2 only shows an exemplary structure of the media information recommendation device based on a recommendation model, but not all the structure, and part or all of the structure shown in fig. 2 can be implemented as required.

The media information recommending device based on the recommending model provided by the embodiment of the invention comprises: at least one processor 201, a memory 202, a user interface 203, and at least one network interface 204. The various components in the recommendation model-based media information recommendation device are coupled together by a bus system 205. It is understood that the bus system 205 is used to enable connected communications between these components. The bus system 205 includes a power bus, a control bus, and a status signal bus in addition to the data bus. But for clarity of illustration the various buses are labeled as bus system 205 in fig. 2.

The user interface 203 may include, among other things, a display, keyboard, mouse, trackball, click wheel, keys, buttons, touch pad, or touch screen, etc.

It will be appreciated that the memory 202 may be either volatile memory or nonvolatile memory, and may include both volatile and nonvolatile memory. The memory 202 in embodiments of the present invention is capable of storing data to support operation of the terminal (e.g., 10-1). Examples of such data include: any computer program, such as an operating system and application programs, for operation on the terminal (e.g., 10-1). The operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, and the like, for implementing various basic services and processing hardware-based tasks. The application may comprise various applications.

In some embodiments, the recommendation model-based media information recommendation device provided by the embodiments of the present invention may be implemented by combining software and hardware, and as an example, the recommendation model-based media information recommendation device provided by the embodiments of the present invention may be a processor in the form of a hardware decoding processor, which is programmed to perform the training method of the recommendation model-based media information recommendation model provided by the embodiments of the present invention. For example, a processor in the form of a hardware decoding processor may employ one or more application specific integrated circuits (ASICs, application Specific Integrated Circuit), DSPs, programmable logic devices (PLDs, programmable Logic Device), complex programmable logic devices (CPLDs, complex Programmable Logic Device), field programmable gate arrays (FPGAs, field-Programmable Gate Array), or other electronic components.

As an example of implementation of the recommendation model-based media information recommendation device provided by the embodiment of the present invention by combining software and hardware, the recommendation model-based media information recommendation device provided by the embodiment of the present invention may be directly embodied as a combination of software modules executed by the processor 201, the software modules may be located in a storage medium, the storage medium is located in the memory 202, the processor 201 reads executable instructions included in the software modules in the memory 202, and the necessary hardware (including, for example, the processor 201 and other components connected to the bus 205) is combined to complete the training method of the recommendation model-based media information recommendation model provided by the embodiment of the present invention.

By way of example, the processor 201 may be an integrated circuit chip having signal processing capabilities such as a general purpose processor, such as a microprocessor or any conventional processor, a digital signal processor (DSP, digital Signal Processor), or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like.

As an example of hardware implementation of the recommendation model-based media information recommendation device provided by the embodiment of the present invention, the device provided by the embodiment of the present invention may be directly implemented by the processor 201 in the form of a hardware decoding processor, for example, one or more application specific integrated circuits (ASIC, application Specific Integrated Circuit), DSPs, programmable logic devices (PLD, programmable Logic Device), complex programmable logic devices (CPLD, complex Programmable Logic Device), field programmable gate arrays (FPGA, field-Programmable Gate Array), or other electronic components may be used to implement the training method for implementing the recommendation model-based media information recommendation model provided by the embodiment of the present invention.

The memory 202 in embodiments of the present invention is used to store various types of data to support the operation of recommendation model-based media information recommendation devices. Examples of such data include: any executable instructions, such as executable instructions, for operation on a recommendation model-based media information recommendation device, a program implementing a training method from a recommendation model-based media information recommendation model according to embodiments of the present invention may be included in the executable instructions.

In other embodiments, the recommendation model-based media information recommendation device provided in the embodiments of the present invention may be implemented in a software manner, and fig. 2 shows the recommendation model-based media information recommendation device stored in the memory 202, which may be software in the form of a program, a plug-in, or the like, and includes a series of modules, and as an example of the program stored in the memory 202, may include a recommendation model-based media information recommendation device, where the recommendation model-based media information recommendation device includes the following software modules:

an information transmission module 2081 and an information processing module 2082. When software modules in the recommendation model-based media information recommendation device are read into the RAM by the processor 201 and executed, the training method of the recommendation model-based media information recommendation model provided by the embodiment of the invention is implemented, wherein the functions of each software module in the recommendation model-based media information recommendation device include: the information transmission module is used for responding to a media information recommendation request based on a recommendation model and acquiring behavior parameter information of a target object;

the information transmission module 2081 is configured to obtain original features of media information for characterizing preferences of a target object, and obtain domain features of media information corresponding to the target object under each media information domain.

The information processing module 2082 is configured to obtain original features of media information for characterizing preferences of a target object, and obtain domain features of media information corresponding to the target object under each media information domain.

The information processing module 2082 is configured to perform feature mapping on the original feature through the center sub-model, so as to obtain a center feature of the original feature in the target semantic space.

The information processing module 2082 is configured to perform feature mapping on the domain features in the corresponding media information domain through each peripheral sub-model, so as to obtain peripheral features of the domain features in the target semantic space.

The information processing module 2082 is configured to determine, by using the determining sub-model, a target media information domain from at least two media information domains based on the central feature and peripheral features corresponding to each of the media information domains.

The information processing module 2082 is configured to recommend content of the media information domain in the cold start state to the target object by using the target media information domain as a recommendation reference.

According to the electronic device shown in fig. 2, in one aspect of the present application, there is also provided a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The computer instructions are read from the computer readable storage medium by a processor of the electronic device, which executes the computer instructions, causing the electronic device to perform the different embodiments and combinations of embodiments provided in various alternative implementations of the recommendation model-based media information recommendation method described above.

Referring to fig. 3, fig. 3 is an optional flowchart of a recommendation model-based media information recommendation method provided by the embodiment of the present invention, where it is understood that the steps shown in fig. 3 may be performed by various electronic devices running the recommendation model-based media information recommendation device, for example, a dedicated terminal or a server with the recommendation model-based media information recommendation device, and the steps shown in fig. 3 are described below.

Step 301: the media information recommending device based on the recommending model obtains the original characteristics of the media information used for representing the preference of the target object, and obtains the domain characteristics of the media information corresponding to the target object under each media information domain.

In some embodiments of the present invention, a media information domain may correspond to an independent application platform, for example, the independent application platform may be a news platform, an application platform and a video platform, where the news platform may provide a large number of articles to a user to meet different reading requirements of the user, and each article clicked and read by the user corresponds to a title feature, a category feature and an entity feature, and by combining the title feature, the category feature and the entity feature, the domain feature of the media information in the media information domain of the news platform may be obtained.

Referring to fig. 4, fig. 4 is a schematic structural diagram of a recommendation model according to an embodiment of the present invention, where the structure of the recommendation model used in the present application includes: the method comprises a central sub-model, a determining sub-model and at least two peripheral sub-models, wherein each peripheral sub-model corresponds to one media information field, for different target objects, one peripheral sub-model needs to be added every time one media information field is added, and the similarity of the peripheral characteristics and the central characteristics can be determined by comparing the peripheral characteristics output by each peripheral sub-model with the central characteristics output by the central sub-model, so that screening of different peripheral characteristics is realized.

In some embodiments of the present invention, when the original features of the media information characterizing the preference of the target object are obtained, various behavior data of the user matched with the corresponding client may be collected through different program components, and the original log of the user behavior data is effectively extracted, for example, the device number (user account number) of the user, the media information type information, the browsing duration of the media information, and the browsing integrity parameter of the media information are extracted. The historical click behaviors of the users and the browsing time length of the corresponding information are recorded through the subscription service and stored in the Redis, and the online recommendation system pulls the historical click behaviors of the corresponding users when the user requests come.

In some embodiments of the present invention, the acquisition of the original features of the media information used to characterize the target object preferences may be achieved by:

acquiring historical query operation information of the target object in an upstream media information domain; wherein the upstream media information field comprises a plurality of the media information fields; determining media information triggered by the target object in a plurality of media information used for responding to the historical query operation information based on the historical query operation information; extracting features of the historical query operation information to obtain corresponding query features, and extracting features of media information triggered by the target object to obtain corresponding media features; and splicing the query features and the media features to obtain original features of the media information for representing the preference of the target object. The upstream platform can be a content searching applet and an information browsing applet triggered in an instant messaging client, for example, the same user can log in a news client in the downstream platform and browse news through a login account of the instant messaging client.

Referring to fig. 5, fig. 5 is a schematic diagram of original feature acquisition in an embodiment of the present invention, where a content searching and browsing platform is used as an upstream platform, and a news platform, an application platform, and a video platform are respectively used as downstream platforms using text content, mobile software, and video content as service carriers. The downstream platform shares the account of the target object in the upstream platform, the upstream platform does not share the account of the target object of the downstream platform, the news platform, the application platform and the video platform can be respectively executed with text content, mobile software and media information services of video content types through the account of the target object in the upstream platform, the news platform, the application platform and the video platform can be different media information fields, the original characteristics of the media information preferred by the target object are obtained from the content searching and browsing platform, specifically, the user interface shown in fig. 5 comprises a display interface for using a search engine in a corresponding software process from a first person viewing angle, and the display interface display control component can control and display search results matched with search words input by a user. For example: the user inputs the query request of the search word 'sports information' in the search box shown in 501 in the instant messaging client process, and the provided search results are various search results related to 'sports information'. In the presented various search results, media information can be browsed by clicking on the link 502 of any search result, after the historical search query request and the clicked link after each request are obtained through the historical behavior data of the user, the unified operation of symbols can be firstly executed on the request and the link, namely, the request is normalized and submitted in sequence Taking word stems and unit segmentation; domain-level contraction is performed on the links to reduce feature dimensions. The important features with higher frequency are then retained by means of information retrieval weighting (TF-IDF term frequency inverse document frequency). Then the behavior features extracted from the two histories are connected in series to form the original feature f containing the potential interests of the user _u ＝f _q ||f _url 。

As shown in fig. 5, when the number of media information triggered by the target object is plural (each media information presented in fig. 5 is triggered), classification of different media information fields can be performed on the plural media information, so as to obtain a classification result; determining the quantity of the media information included in each media information domain according to the classification result; taking the media information included in the media information fields with the quantity reaching the quantity threshold as target media information; and finally, extracting the characteristics of the target media information in the media information triggered by the target object to obtain corresponding media characteristics. For example, when the number threshold is 5, the presented search results of each type related to "sports information" include: sports videos, sports class applet games and sports news, and when the number of the triggered sports videos is more than or equal to 5, the media information of the sports videos is used as target media information.

In some embodiments of the present invention, when obtaining domain features of media information corresponding to the target object in each type of media information domain, a news platform, an application platform, and a video platform are still taken as different media information domains as an example, and a process for obtaining domain features of media information is described, where when media information included in the media information domain is an article, a title of the article, a category to which the article belongs, and a named entity in the article are obtained; respectively extracting features of the title, the category and the named entity to obtain corresponding title features, category features and entity features; and splicing the title features, the category features and the entity features to obtain domain features of the articles. Specifically, for a news platform, a client may record the clicking behavior of a user on an article while the article is presented. Text (A)The characteristic representation of a chapter can be divided into three parts: and taking the titles of the articles as text sentences, executing corresponding natural language codes, using binary codes to characterize news classification to obtain category characteristics (multi-level classification can be realized), and extracting named entities in each article by using a proprietary named entity algorithm after the contents of the articles are encoded to obtain entity characteristics. Concatenating representations of three parts to obtain domain features of media information of an article

In the process, arabic numerals are not converted into Chinese characters, only conversion irrelevant to the numerals, such as conversion from traditional Chinese to simplified Chinese, is carried out, the original form of the Arabic numerals in sentences is reserved, and meanwhile, the international units connected with the numerals, such as g, kg, cm and the like, are not converted, so that the original state is reserved. For Chinese text, the Chinese text is correspondingly required to be segmented, because the words can only contain complete information in Chinese. Correspondingly, a Chinese word segmentation tool Jieba can be used for word segmentation of Chinese texts. Wherein, the event happens in two zeros and one year, and the word is changed into the event/occurrence/in/two/zero/one/year after word segmentation. Wherein, the word segmentation is that verb meaning and noun meaning are adopted; each word is a word or phrase, i.e. the minimum semantic unit with definite meaning; for the use environments of different users or different text processing models, the minimum semantic units contained in the received text processing models are different, and adjustment is required to be made in time, and the process is called word segmentation, namely word segmentation can refer to the process of dividing the minimum semantic units; on the other hand, the minimum semantic unit obtained after division is also often called word segmentation, i.e., a word obtained after the word segmentation operation is performed; sometimes, in order to distinguish the two meanings from each other, the minimum semantic unit referred to by the latter meaning is referred to as a word segmentation object (Term); the term object is used in this application; the word segmentation object corresponds to a keyword in the inverted list as an index basis. For Chinese, since words as the minimum semantic units are often composed of different numbers of words, and no natural distinguishing mark in alphabetic writing such as blank partition exists among words, it is an important step for Chinese to accurately perform word segmentation to obtain reasonable word segmentation objects.

Similarly, when domain feature processing of media information in an english environment is performed, english text information is processed by means of Word hash (Word hash), and the method is based on n-gram of letters, which is mainly used for reducing the dimension of an input vector. For example, the text information is "boy", and the start and end characters are denoted by # respectively, and the input is (#boy#). The word is converted to the form of the letter n-gram, and if n is set to 3, three sets of data (#bo, boy, oy#) are obtained, which are represented by vectors of the n-gram.

In some embodiments of the present invention, when the media information included in the media information field is a video, a title of the media information, a category to which the media information belongs, and description information of the media information are obtained; respectively extracting features of the title, the category and the description information to obtain corresponding title features, category features and description features; and splicing the title features, the category features and the description features to obtain domain features of the media information. For the application platform, the application platform client can record the historical downloading behavior of the user to the application, and the instant messaging client can record the record that the user triggers each type of applet. Features of the application may include: encoding header information and class identifier, and encoding description features of the application (e.g., language encoding is performed on the initial functional description provided by the publisher on the application platform), concatenated with the encoding of header information and class identifier to obtain domain features of the media information of the application

In some embodiments of the present invention, when the media information included in the media information field is a video, a title of the media information, a category to which the media information belongs, and description information of the media information are obtained; respectively to the title and the placeExtracting the characteristics of the category and the description information to obtain corresponding title characteristics, category characteristics and description characteristics; and splicing the title features, the category features and the description features to obtain domain features of the media information. Specifically, the video playing client uses the video watching behavior data to take the title of the video as a text sentence by recording the historical watching behavior data of the user on the video, executes corresponding natural language coding, and uses binary coding to represent the classification of the video types to obtain the video; category features (which may have multiple levels of classification), named entities in each video are extracted after encoding video titles using proprietary named entity algorithms to obtain entity features. Concatenating representations of three portions to obtain domain features of media information of the video

Step 302: and the media information recommending device based on the recommending model performs feature mapping on the original features through the center sub-model to obtain the center features of the original features under the target semantic space.

In some embodiments of the present invention, obtaining the central feature of the original feature under the target semantic space may be achieved by:

determining constraint conditions corresponding to the target semantic space; performing feature mapping on the original features according to the constraint conditions through the center sub-model to obtain a center feature vector; and carrying out weighting treatment on the central feature vector to obtain the central feature of the original feature under the target semantic space. Wherein, as at least one hidden layer used in extracting the original features is 300 dimensionality, the constraint condition is dimension reduction to 128, when feature mapping can be realized through the constraint condition corresponding to the target semantic space, dimension reduction processing is carried out on the original features, the original features are mapped into 128-dimensionality center feature vectors, so as to reduce the calculation amount of a recommendation model, reduce the load of corresponding hardware equipment,

when the center feature vector is weighted, the original feature is acquiredThe historical query operation information of the target object in the upstream media information domain can comprise a plurality of media information domains, when the original feature of the media information representing the preference of the target object is obtained, the query feature and the media feature are needed to be spliced, so that the query feature and the media feature in the original feature can correspond to different weight parameters, the original feature is subjected to feature mapping according to constraint conditions, after the central feature vector subjected to the dimension reduction processing is obtained, the central feature vector is also formed by connecting two parts in series, and the central feature under the target semantic space of the original feature can be obtained by continuously carrying out weighting processing on the central feature vector, so that the central feature f is obtained _u ＝f _q ||f _url For example, f is respectively related to the weight parameters of the query feature and the media feature _q And f _url And by weighting, the original characteristics which are more in line with the potential interests of the user can be obtained.

Step 303: and the media information recommending device based on the recommending model performs feature mapping on the domain features under the corresponding media information domain through each peripheral sub-model to obtain the peripheral features of the domain features under the target semantic space.

In some embodiments of the present invention, feature mapping is performed on the domain features under the corresponding media information domain, so as to obtain peripheral features of the domain features under the target semantic space, which may be implemented by the following ways:

triggering a corresponding peripheral sub-model according to the type of the media information field; determining constraint conditions corresponding to the target semantic space; and performing feature mapping on the domain features according to the constraint conditions through the peripheral submodel to obtain peripheral feature vectors. Referring to fig. 6, fig. 6 is a schematic diagram of a process of acquiring peripheral features and central features in an embodiment of the present invention, where high-dimensional original features are obtained by processing in steps 301 and 302 in the foregoing embodiment as input, feature mapping is performed on the original features by a central sub-model, and feature mapping is performed on domain features by a peripheral sub-model, so that feature vectors in 128 dimensions are reduced to a low-dimensional space (64 or 32 dimensions). Obtaining the original characteristics according to the same constraint conditions Features centered under the target semantic space and features peripheral to the domain under the target semantic space, e.g. if x represents the input vector of one-sided network, y represents the output vector, li, i=1, … N-1, represents the middle hidden layer, W _i A weight matrix representing the i-th layer, b _i Representing the bias term of the i-th layer, the output of the hidden layer and the output of the center submodel can be expressed as formula 1:

wherein, referring to equation 2, one can use

The function is an activation function of the output layer and the hidden layer li:

in ranking the content, referring to equation 3, cosine similarity of the peripheral features and the central features of the peripheral sub-model (e.g., a peripheral tower in a double tower structure) and the central sub-model (e.g., an axial tower in a double tower structure) may be used as a ranking basis:

step 304: and the media information recommending device based on the recommending model determines target media information domains from at least two media information domains based on the central characteristics and the peripheral characteristics corresponding to each media information domain through the determining sub-model.

Since the model structure shown in fig. 4 shows that the number of peripheral sub-models is not unique, and each peripheral sub-model corresponds to one media information domain, determining a target media information domain from at least two media information domains can be achieved by:

Matching the central feature with each peripheral feature through the determination sub-model to obtain the matching degree of the central feature and each peripheral feature; and determining the media information domain corresponding to the peripheral characteristic with the largest matching degree as a target media information domain. Referring to FIG. 7, FIG. 7 is a schematic diagram showing an alternative model structure for determining a target media information field in an embodiment of the present invention, wherein the recommended model adopts a tower network including an axial tower (corresponding to the central sub-model in the previous embodiment), 2 peripheral towers (corresponding to the peripheral sub-models in the previous embodiment), and a matching network (corresponding to the determined sub-model in the previous embodiment), wherein there is an axial tower T _u And v peripheral towers T _1— T _v ，T _i With field i-specific input X _i Its dimension length is R ^di . Through the processing of the preceding embodiments, the structure of each peripheral tower may be represented as a nonlinear mapping layer f _i (X _i ,W _i ) The input of the field can be mapped into the shared semantic space through the conversion function to obtain Y _i 。

Thus, each single tower network in the recommendation model can implement a nonlinear transformation such that the sum of the degree of matching of the central feature and the peripheral feature can reach a maximum value in the semantic space, which can be expressed by equation 4,

With continued reference to FIG. 7, a hub tower T _u Corresponding to the original characteristics of the media information characterizing the preferences of the target object, a peripheral tower T _i Corresponding to the I of the media information corresponding to the target object under each media information domain _i . The matching degree of the central features and the peripheral features is compared in pairs, so that the media information domain corresponding to the peripheral feature with the largest matching degree can be determined as the target media information domain, and the matching degree of all the central features and the peripheral features can be sequenced to be determined as the target media information domain.

Step 305: and the media information recommending device based on the recommending model uses the target media information domain as a recommending reference to recommend the content of the media information domain in the cold start state to the target object.

In some embodiments of the present invention, with the target media information domain as a recommendation reference, content recommendation of the media information domain in the cold start state is performed on the target object, which may be implemented by the following ways:

acquiring media information categories preferred by the target object in the target media information domain; and based on the media information category, recommending the media information of the same category to the target object in the media information field in the cold start state.

Wherein, cold start state includes: 1) When the target object is a low-resolution user, starting the client (for example, starting a short video client to perform short video recommendation); 2) And when the target object is a new user, starting the client. The media information domain in the cold start state may be a media information domain corresponding to a client corresponding to at least one downstream platform in the downstream platforms. The target object may include: one or more of low-resolution users, new users and high-resolution users, specifically, the new users are users who register and use the client for the first time; the low-resolution user is a user registered to use the short video client, but less user behavior information is generated; the high-resolution user is a user registered to use the short video client, and more user behavior information is generated in the use process of the client.

For example, in the recommendation environment of the short video client, the threshold of the number of the user behavior information may be set to 10 pieces, and when the number of the user behavior information of a certain user is less than or equal to 10 pieces, the user may be a low-resolution user; when the number of user behavior information of a certain user is more than 10 pieces, the user is regarded as a high-resolution user. It should be noted that, for different multimedia information recommendation environments, the threshold value of the user behavior information may be dynamically adjusted, and the embodiment of the present invention is not limited in particular.

The target media information domain is obtained as a recommendation reference through the processing of the preceding steps 301-304, and when media information recommendation is performed, as the recommendation environment of a new user does not have enough information to determine the preference of the user for video recommendation, when the media information recommendation method based on the recommendation model provided by the application determines that the domain features of an article are used as the target media information domain, the domain features of the article can be used as the recommendation reference, and when video recommendation is performed, video information with the same category as the domain features of the article is preferentially recommended to the user.

When the low-consumption user is recommended, although the corresponding original recommendation strategy (for example, the entertainment video which is interested by the user is determined) can be determined according to the use information of the low-consumption user (the user triggers the entertainment video once), the original recommendation strategy of the low-consumption user can be ignored because of the small number of samples, when the domain features of the article are determined as the target media information domain through the media information recommendation method based on the recommendation model provided by the application, the domain features of the article can be used as recommendation references, and when the video is recommended, video information (for example, video information of the financial category is recommended) with the same category as the domain features of the article is recommended to the user preferentially, so that the false recommendation strategy caused by the small number of samples is avoided, and the use experience of the user is improved.

Referring to fig. 8, fig. 8 is an optional flowchart of a training method of a recommendation model according to an embodiment of the present invention, it may be understood that the steps shown in fig. 8 may be performed by various electronic devices running the media information recommendation device, for example, a dedicated terminal with the media information recommendation device, a server, or a server cluster, where the dedicated terminal with the media information recommendation device may be an electronic device with the media information recommendation device in the embodiment shown in fig. 2. The following is a description of the steps shown in fig. 8.

Step 801: building a training sample, the training sample comprising: the training object comprises sample original characteristics of media information for representing the preference of the training object, and sample domain characteristics of the training object corresponding to the media information under at least two media information domains.

The training samples are marked with sample matching degrees of the sample domain features and the sample original features. When constructing training sample sets, it is necessary to find the upstream user space U _h For downstream platform user space

The users in each intersection constitute the downstream platform specific user set for the model training>

Step 802: and performing feature mapping on the sample original features through the center sub-model to obtain sample center features of the sample original features under a target semantic space.

In order to reduce the number of samples in the training sample set, the training cost of the recommendation model may be reduced by controlling the number of samples in the training sample set in any of the following ways.

1) Through common feature screening, the sample number is reduced, and the method is specifically expressed as follows: the top K most common features are selected, which may be, for example, all features with a probability of occurrence in the target object greater than 0.001, which may characterize the common online behavior of the user.

2) Extracting features by a clustering algorithm, which is embodied by representing each feature by using a vector fi of length |U|, where |U| is the number of target objects in training data, f _i (j) And (3) carrying out vector normalization on the times of containing the feature i for the user j, then executing a K-means algorithm on all feature vectors, and setting the number of clusters |F| according to the specific situation of the features. After clustering, each new feature is allocated a new vector with the length of |F|

The new features are expressed as: />

Cls (i) represents the category to which the original feature i belongs after clustering.

3) Extracting features by a local sensitive hash mode, wherein the specific expression is as follows: configuring a transformation matrix A epsilon R ^d*k Where d is the number of features in the original space and k is the dimension of the low-dimensional space. Multiplying the original features by a transformation matrix, i.e. x·a=y∈r ^k And obtaining the new feature Y on the converted low-dimensional space. Next, the negative value in Y, i.e. Y, is replaced by zero ⁽ⁱ⁾ ＝max(Y ⁽ⁱ⁾ ,0). Feature X in original space ₁ ,X ₂ ∈R ^d Can be used as

Approximately, where H (Y ₁ ,Y ₂ ) Is the hamming distance of the locally sensitive hash of the two original features.

Step 803: and respectively carrying out feature mapping on the sample domain features under the corresponding media information domain through each peripheral sub-model to obtain sample peripheral features of the sample domain features under the target semantic space.

Step 804: and respectively determining the matching degree of the peripheral features of each sample and the central features of the sample through the determination sub-model.

Step 805: and acquiring the difference between the determined matching degree and the corresponding sample matching degree, and updating the model parameters of the recommendation model based on the difference.

A softmax function is performed on the recommendation scores of each user for the candidate content, by normalizing the scores in the following manner,

Wherein, gamma is a smoothing factor in the softfmax function, and needs to be adjusted through experiments; i _l The candidate sets to be ordered are shown. In the training process, il should ideally contain the field; l all possible contents. But in practice, all the interactions are usually usedContent

And N randomly sampled non-interactive content

Is to approximate I by the union of _l 。

Specifically, each time parameter updating is performed, an error between the predicted matching probability and the actual matching probability between the original feature and the domain feature can be determined, and model parameters of different sub-models (a peripheral sub-model, a central sub-model and a determining sub-model) in the recommended model are adjusted according to the error value until the predicted matching probability between the original feature and the domain feature is the same as the actual matching probability.

Specifically, the value of the loss function is determined based on the difference, whether the value of the loss function exceeds a preset threshold value is judged, when the value of the loss function exceeds the preset threshold value, an error signal of a recommendation model is determined based on the loss function, the error signal is reversely propagated in the recommendation model, and model parameters of each layer are updated in the propagation process.

The back propagation is described, a training sample is input into an input layer of a neural network model, passes through a hidden layer, finally reaches an output layer and outputs a result, which is a forward propagation process of the neural network model, and as the output result of the neural network model has an error with an actual result, the error between the output result and the actual value is calculated, and the error is propagated back from the output layer to the hidden layer until the error is propagated to the input layer, and in the back propagation process, the value of a model parameter is adjusted according to the error; the above process is iterated until convergence.

The training process of the recommended model will be described by taking a tower-type network as an example of the recommended model in conjunction with table 1, wherein the tower-type network includes an axial tower (corresponding to the central sub-model in the previous embodiment), at least 2 peripheral towers (corresponding to the peripheral sub-model in the previous embodiment), and a matching network (corresponding to the determined sub-model in the previous embodiment), the training process is as shown in table 1,

therefore, the trained recommendation model can be packaged in a corresponding APP or stored in an instant messaging client in a plug-in mode, and recommendation of different media information is achieved.

It can be appreciated that, in the embodiments of the present application, related data such as user information, user behavior data, click behavior of a target object, etc., when the embodiments of the present application are applied to specific products or technologies, user permission or consent needs to be obtained, and the collection, use and processing of related data needs to comply with related laws and regulations and standards of related countries and regions.

The beneficial technical effects are as follows:

The foregoing description of the embodiments of the invention is not intended to limit the scope of the invention, but is intended to cover any modifications, equivalents, and alternatives falling within the spirit and principles of the invention.

Claims

1. A media information recommendation method based on a recommendation model, wherein the recommendation model includes a center sub-model, a determination sub-model, and at least two peripheral sub-models, each of the peripheral sub-models corresponding to a media information field, the method comprising:

2. The method of claim 1, wherein said determining, by said determining sub-model, a target media information field from at least two of said media information fields based on said central feature and a peripheral feature corresponding to each of said media information fields, comprises:

matching the central feature with each peripheral feature through the determination sub-model to obtain the matching degree of the central feature and each peripheral feature;

and determining the media information domain corresponding to the peripheral feature with the largest matching degree as the target media information domain.

3. The method according to claim 1, wherein the recommending the content of the media information field in the cold start state to the target object with the target media information field as a recommendation reference includes:

acquiring media information categories preferred by the target object in the target media information domain;

and based on the media information category, recommending the media information of the same category to the target object in the media information field in the cold start state.

4. The method of claim 1, wherein obtaining raw features of media information characterizing target object preferences comprises:

acquiring historical query operation information of the target object in an upstream media information domain; wherein the upstream media information field comprises a plurality of the media information fields;

determining media information triggered by the target object in a plurality of media information used for responding to the historical query operation information based on the historical query operation information;

extracting features of the historical query operation information to obtain corresponding query features, and extracting features of media information triggered by the target object to obtain corresponding media features;

and splicing the query features and the media features to obtain original features of the media information for representing the preference of the target object.

5. The method of claim 4, wherein prior to the feature extraction of the historical query operation information, the method further comprises:

when the number of the media information triggered by the target object is a plurality of media information, classifying the media information into different media information domains to obtain a classification result;

Determining the quantity of the media information included in each media information domain according to the classification result;

taking the media information included in the media information fields with the quantity reaching the quantity threshold as target media information;

the feature extraction of the media information triggered by the target object to obtain corresponding media features comprises the following steps:

and extracting the characteristics of the target media information in the media information triggered by the target object to obtain corresponding media characteristics.

6. The method according to claim 1, wherein the obtaining the domain feature of the media information corresponding to the target object in each of the media information domains includes:

when the media information included in the media information field is an article, acquiring the title of the article, the category to which the article belongs and the named entity in the article;

respectively extracting features of the title, the category and the named entity to obtain corresponding title features, category features and entity features;

and splicing the title features, the category features and the entity features to obtain domain features of the articles.

7. The method according to claim 1, wherein the obtaining the domain feature of the media information corresponding to the target object in each of the media information domains includes:

When the media information included in the media information field is video, acquiring a title of the media information, a category to which the media information belongs and description information of the media information;

respectively extracting features of the title, the category and the description information to obtain corresponding title features, category features and description features;

and splicing the title features, the category features and the description features to obtain domain features of the media information.

8. The method according to claim 1, wherein the obtaining the domain feature of the media information corresponding to the target object in each of the media information domains includes:

when the media information included in the media information field is an application, acquiring a title of the media information, a category to which the media information belongs and description information of the media information through a download record of the media information;

9. The method according to claim 4, wherein the feature mapping the original feature by the center sub-model to obtain a center feature of the original feature in a target semantic space comprises:

determining constraint conditions corresponding to the target semantic space;

performing feature mapping on the original features according to the constraint conditions through the center sub-model to obtain a center feature vector;

and carrying out weighting treatment on the central feature vector to obtain the central feature of the original feature under the target semantic space.

10. The method according to claim 1, wherein the feature mapping the domain features under the corresponding media information domain through each peripheral sub-model to obtain peripheral features of the domain features under the target semantic space includes:

triggering a corresponding peripheral sub-model according to the type of the media information field;

determining constraint conditions corresponding to the target semantic space;

and performing feature mapping on the domain features according to the constraint conditions through the peripheral submodel to obtain peripheral feature vectors.

11. The method according to claim 1, wherein the method further comprises:

Building a training sample, the training sample comprising: the training object comprises sample original characteristics of media information for representing the preference of the training object and sample domain characteristics of the training object corresponding to the media information under at least two media information domains;

the training samples are marked with sample matching degrees of the sample domain features and the sample original features;

performing feature mapping on the sample original features through the center sub-model to obtain sample center features of the sample original features under a target semantic space;

performing feature mapping on the sample domain features under the corresponding media information domain through each peripheral sub-model to obtain sample peripheral features of the sample domain features under the target semantic space;

respectively determining the matching degree of the peripheral features of each sample and the central features of the sample through the determination sub-model;

and acquiring the difference between the determined matching degree and the corresponding sample matching degree, and updating the model parameters of the recommendation model based on the difference.

12. A media information recommendation device based on a recommendation model, wherein the recommendation model includes a central sub-model, a determination sub-model, and at least two peripheral sub-models, each of the peripheral sub-models corresponding to a media information field, the device comprising:

13. A computer program product comprising a computer program or instructions which, when executed by a processor, implements the recommendation model based media information recommendation method of any one of claims 1 to 11.

14. An electronic device, the electronic device comprising:

a memory for storing executable instructions;

a processor configured to implement the recommendation model-based media information recommendation method of any one of claims 1 to 11 when executing the executable instructions stored in the memory.

15. A computer readable storage medium storing executable instructions which when executed by a processor implement the recommendation model based media information recommendation method of any one of claims 1-11.