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

Abstract

The application provides a media information recommendation method and device based on a recommendation model, and related embodiments can be applied to various scenes such as cloud technology, cloud security, and intelligent transportation. The method comprises the following steps: obtaining original features and domain features of media information corresponding to a target object; performing feature mapping on the original features by using a center sub-model to obtain center features of the original features in a target semantic space; performing feature mapping on the domain features in a corresponding media information domain by using each peripheral sub-model to obtain peripheral features of the domain features in the target semantic space; and determining a target media information domain from at least two media information domains by using a determination sub-model. Thus, the target media information domain can be used as a recommendation reference to perform content recommendation of a media information domain in a cold start state for the target object, and the accuracy and timeliness of media information recommendation in the cold start state are enhanced, and the quality of media information recommendation is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information processing, in particular to a media information recommendation method and device based on a recommendation model, an electronic device, a computer program product and a storage medium, so that the present application can be applied in the fields of automatic driving, Internet of Vehicles, intelligent transportation and the like. BACKGROUND

[0002] Artificial intelligence (AI) is a comprehensive technology of computer science, which makes machines have the functions of perception, reasoning and decision-making by studying the design principles and implementation methods of various intelligent machines. Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, such as natural language processing technology and machine learning / deep learning. It is believed that with the development of technology, artificial intelligence technology will be applied in more and more fields and play an increasingly important role.

[0003] In the related art, when recommending media information to a user, the content is usually recommended based on the content history interaction record of the object to be recommended, or the user object with similar historical behavior in the same system as the object to be recommended is found, and the interaction content of the user object is used as the recommended content to recommend to the user to be recommended. However, for the cold start situation, there are no or only a few user-content historical interaction records in the system, so it is difficult to accurately recommend media information. SUMMARY

[0004] Therefore, the embodiments of the present application provide a media information recommendation method and device based on a recommendation model, an electronic device, a software program and a storage medium, which can enhance the accuracy and timeliness of media information recommendation in a cold start state, effectively improve the quality of media information recommendation, and improve the user experience.

[0005] The embodiments of the present application provide a media information recommendation method based on a recommendation model, the recommendation model comprising a center sub-model, a determination sub-model and at least two peripheral sub-models, each peripheral sub-model corresponding to a media information domain, the method comprising:

[0006] obtaining original features of media information representing the preferences of a target object, and obtaining domain features of media information corresponding to the target object in each media information domain;

[0007] mapping the original features through the center sub-model to obtain center features of the original features in a target semantic space;

[0008] respectively through each of the peripheral sub-models, the domain features in the corresponding media information domain are mapped to obtain peripheral features of the domain features in the target semantic space;

[0009] through the determination sub-model, based on the center feature and the peripheral features corresponding to each of the media information domains, a target media information domain is determined from at least two of the media information domains;

[0010] the target media information domain is taken as a recommendation reference, and content recommendation of a media information domain in a cold start state of the target object is performed.

[0011] Embodiments of the present application also provide a media information recommendation device based on a recommendation model, the recommendation model comprising 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 domain, and the device comprising:

[0012] an information transmission module configured to acquire original features of media information representing preferences of a target object, and acquire domain features of media information corresponding to the target object in each of the media information domains;

[0013] an information processing module configured to map the original features to obtain center features of the original features in a target semantic space through the center sub-model;

[0014] the information processing module is configured to map the domain features in the corresponding media information domain to obtain peripheral features of the domain features in the target semantic space through each of the peripheral sub-models;

[0015] the information processing module is configured to determine a target media information domain from at least two of the media information domains through the determination sub-model based on the center feature and the peripheral features corresponding to each of the media information domains;

[0016] the information processing module is configured to take the target media information domain as a recommendation reference, and perform content recommendation of a media information domain in a cold start state of the target object.

[0017] in the above scheme, the information processing module is configured to match the center feature with each of the peripheral features to obtain matching degrees of the center feature and each of the peripheral features through the determination sub-model;

[0018] the information processing module is configured to determine a media information domain corresponding to the peripheral feature with the largest matching degree as the target media information domain.

[0019] in the above scheme, the information processing module is configured to acquire a media information category preferred by the target object in the target media information domain.

[0020] The information processing module is configured to recommend the same type of media information to the target object in the media information domain in the cold start state based on the media information type.

[0021] In the foregoing solution, the information processing module is configured to acquire historical query operation information of the target object in an upstream media information domain, and the upstream media information domain includes a plurality of media information domains.

[0022] The information processing module is configured to determine, based on the historical query operation information, media information triggered by the target object in response to the historical query operation information.

[0023] The information processing module is configured to extract features of the historical query operation information to obtain corresponding query features, and extract features of the media information triggered by the target object to obtain corresponding media features.

[0024] The information processing module is configured to concatenate the query features and the media features to obtain original features of the media information preferred by the target object.

[0025] In the foregoing solution, the information processing module is configured to, when the number of media information triggered by the target object is greater than one, classify the plurality of media information in different media information domains to obtain classification results.

[0026] The information processing module is configured to determine, according to the classification results, the number of media information included in each media information domain.

[0027] The information processing module is configured to take media information included in a media information domain whose number reaches a number threshold as target media information.

[0028] The information processing module is configured to extract features of the target media information in the media information triggered by the target object to obtain corresponding media features.

[0029] In the foregoing solution, the information processing module is configured to, when the media information included in the media information domain is an article, acquire a title of the article, a type to which the article belongs, and a named entity in the article.

[0030] The information processing module is configured to extract features of the title, the type, and the named entity respectively to obtain corresponding title features, type features, and entity features.

[0031] The information processing module is configured to concatenate the title features, the type features, and the entity features to obtain a domain feature of the article.

[0032] The information processing module is configured to, when the media information included in the media information domain is a video, acquire a title of the media information, a category to which the media information belongs, and description information of the media information.

[0033] The information processing module is configured to respectively perform feature extraction on the title, the category, and the description information to obtain corresponding title features, category features, and description features.

[0034] The information processing module is configured to splice the title features, the category features, and the description features to obtain a domain feature of the media information.

[0035] The information processing module is configured to, when the media information included in the media information domain is an application, acquire 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.

[0036] The information processing module is configured to respectively perform feature extraction on the title, the category, and the description information to obtain corresponding title features, category features, and description features.

[0037] The information processing module is configured to splice the title features, the category features, and the description features to obtain a domain feature of the media information.

[0038] The information processing module is configured to determine a constraint condition corresponding to the target semantic space.

[0039] The information processing module is configured to perform feature mapping on the original feature according to the constraint condition through the center sub-model to obtain a center feature vector.

[0040] The information processing module is configured to perform weighted processing on the center feature vector to obtain a center feature of the original feature in the target semantic space.

[0041] The information processing module is configured to trigger a corresponding peripheral sub-model according to a type of the media information domain.

[0042] The information processing module is configured to determine a constraint condition corresponding to the target semantic space.

[0043] The information processing module is configured to perform feature mapping on the domain feature according to the constraint condition through the peripheral sub-model to obtain a peripheral feature vector.

[0044] In the scheme, 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.

[0045] The information processing module is configured to obtain corresponding original features and domain features and actual matching probabilities between each original feature and domain feature in the training sample set according to the training target object.

[0046] The information processing module is configured to input the original feature into the center sub-model and input the domain feature into the peripheral sub-model, and obtain a predicted matching probability between the original feature and the domain feature of the same training target object in the training sample set output by the determination sub-model.

[0047] The information processing module is configured to compare the predicted matching probability with the actual matching probability, and if they are inconsistent, adjust model parameters of different sub-models in the recommendation model until the predicted matching probability between the original feature and the domain feature is the same as the actual matching probability.

[0048] The embodiment of the present application further provides an electronic device, which comprises:

[0049] A memory is configured to store executable instructions.

[0050] A processor is configured to run the executable instructions stored in the memory, and implement the media information recommendation method based on the recommendation model.

[0051] The embodiment of the present application further provides a computer program product, which comprises a computer program or instructions, and the computer program or instructions are executed by a processor to implement the media information recommendation method based on the recommendation model.

[0052] The embodiment of the present application further provides a computer readable storage medium, which stores executable instructions, and the executable instructions are executed by a processor to implement the media information recommendation method based on the recommendation model.

[0053] The embodiment of the present application has the following beneficial effects:

[0054] The application obtains original features of media information for characterizing preferences of a target object, and obtains domain features of media information corresponding to the target object under each media information domain; the central sub-model is used to perform feature mapping on the original features to obtain central features of the original features in a target semantic space; each peripheral sub-model is used to perform feature mapping on the domain features under the corresponding media information domain to obtain peripheral features of the domain features in the target semantic space; and the determination sub-model is used to determine a target media information domain from at least two media information domains based on the central features and the peripheral features corresponding to each media information domain. Therefore, the target media information domain can be used as a recommendation reference to perform content recommendation of a media information domain in a cold start state for the target object, the accuracy and timeliness of media information recommendation in the cold start state are enhanced, the quality of media information recommendation is effectively improved, and the use experience of a user is improved. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 A use scenario diagram of a media information recommendation method based on a recommendation model provided by an embodiment of the application;

[0056] Figure 2 A component structure diagram of a media information recommendation device based on a recommendation model provided by an embodiment of the application;

[0057] Figure 3 An optional flow diagram of a media information recommendation method based on a recommendation model provided by an embodiment of the application;

[0058] Figure 4 A structure diagram of a recommendation model in an embodiment of the application;

[0059] Figure 5 An original feature acquisition diagram in an embodiment of the application;

[0060] Figure 6 An acquisition process diagram of peripheral features and central features in an embodiment of the application;

[0061] Figure 7 An optional model structure diagram for determining a target media information domain in an embodiment of the application;

[0062] Figure 8 An optional flow diagram of a training method of a recommendation model provided by an embodiment of the application. DETAILED DESCRIPTION

[0063] In order to make the objects, technical solutions and advantages of the present application clearer, the following further describes the present application with reference to the accompanying drawings, the described embodiments should not be regarded as limitations to the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0064] In the following description, "some embodiments" are related to a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0065] The related data collection and processing in the embodiments of the present application should strictly comply with the requirements of relevant laws and regulations, obtain the informed consent or separate consent of the personal information subject, and within the scope of authorization of laws and regulations and the personal information subject, carry out subsequent data use and processing.

[0066] Before further detailing the embodiments of the present application, the terms and phrases involved in the embodiments of the present application are explained, and the terms and phrases involved in the embodiments of the present application are applicable to the following explanations.

[0067] 1) responsive to, used to represent the condition or state on which the operation is dependent, when the dependent condition or state is met, one or more operations performed can be real-time or have a set delay; in the absence of special instructions, there is no restriction on the execution order of multiple operations performed.

[0068] 2) based on, used to represent the condition or state on which the operation is dependent, when the dependent condition or state is met, one or more operations performed can be real-time or have a set delay; in the absence of special instructions, there is no restriction on the execution order of multiple operations performed.

[0069] 3) softmax: a very common and important function in machine learning, especially widely used in multi-classification scenarios, which maps some inputs to real numbers between 0 and 1, and normalizes to ensure the sum is 1.

[0070] 4) Neural Network (NN): Artificial Neural Network (ANN), simply neural network or neural network, in the field of machine learning and cognitive science, is a mathematical model or computational model that simulates the structure and function of biological neural networks (animal central nervous system, especially brain), used to estimate or approximate functions.

[0071] 5) Multi-task Learning: Multi-task Learning, in the field of machine learning, through the joint learning and optimization of multiple related tasks at the same time, the model accuracy can be better than that of a single task, multiple tasks help each other through shared representation layers, this training method is called multi-task learning, also called joint learning.

[0072] 6) Media information timeliness: the content of media information has a certain effect within a period of time, and the effect is measured by the user's interest in the media information. Within a period of time, the user is interested in a certain type of media information, so timeliness plays an important role in the end-side online user retention, click and CTR. Pushing content to users within the effective period of content timeliness can have a positive effect, otherwise it will cause the user's aversion.

[0073] Among them, the embodiment of the application can be realized in combination with cloud technology. Cloud technology refers to a kind of hosting technology that unifies a series of resources such as hardware, software and network in a wide area network or local area network to realize data calculation, storage, processing and sharing. It can also be understood as a general term for network technology, information technology, integration technology, management platform technology and application technology based on cloud computing business model application. The background service of the technical network system needs a large amount of computing and storage resources, such as video websites, picture websites and more portal websites, so cloud technology needs to be supported by cloud computing.

[0074] Through cloud technology, the media information recommendation method based on the recommendation model provided in the application can take the target media information domain as a recommendation reference and record the target media information domain in the corresponding cloud server. When the target object browses media information in different terminals, the corresponding media information can be recommended to the target object through the target media information domain stored in the cloud server as a recommendation reference, so that the target object obtains more accurate media information recommendation results.

[0075] It should be noted that cloud computing is a computing mode which distributes computing tasks on a resource pool composed of a large number of computing machines, so that various application systems can obtain computing power, storage space and information services according to needs. The network providing resources is called "cloud". The resources in the "cloud" are infinitely expandable to users and can be obtained at any time, used on demand, expanded at any time and paid according to use. As a basic capability provider of cloud computing, a cloud computing resource pool platform, referred to as a cloud platform, is established, generally referred to as infrastructure as a service (IaaS), and various types of virtual resources are deployed in the resource pool for external customers to select and use. The cloud computing resource pool mainly includes: computing devices (which can be virtualized machines containing operating systems), storage devices and network devices.

[0076] 7) Media information, various forms of information available on the Internet, such as advertising information, video files, media information to be recommended, news information, etc. presented in the client or intelligent device.

[0077] Before introducing the media information recommendation method based on the recommendation model provided in the present application, the defects of media information recommendation in the related art are briefly described. In the related art, the media information recommendation can be performed in the following ways:

[0078] 1) Media information recommendation based on collaborative filtering recommendation technology, which is specifically represented as: finding other users with similar historical behaviors from the system through the historical interaction behavior records of users in the system, taking the contents interacted by these users as the most relevant contents; or finding other contents with similar interaction records of the contents interacted, taking these contents as the most relevant contents, and then displaying these most relevant contents to the target object. The defect of this way is that a large number of user-content historical interaction records are needed to generate high-quality representations for users and contents for effective recommendation. When the user or content lacks historical interaction records, the cold start problem will occur; for the recommendation of new users or new contents, because the historical interaction records are almost zero, accurate recommendation in the cold start environment cannot be realized.

[0079] 2) Media information recommendation based on description information, which is specifically manifested as: the content description information is characterized through feature engineering, and similar content is recommended to the user according to the similarity between the features. Since the modeling object of this kind of technology is the feature of the content, and the feature is usually the carrier of describing the commonality of things, even without the historical interaction information of the user and the content, the content can still be well represented. The defect of this way is that: the description information usually comes from the age, gender, region, etc. of the user, and the user groups can be divided according to these features; for user individualization representation, these features are not enough to capture the actual user's interest. This kind of method only takes the interaction history of the user and the content as a label to model the features of the user, ignoring the relationship between the user and the content. At the same time, the timeliness of these description information is usually limited to the initial stage of the user entering the system. With the user's processing of the media information, the user's interest is also changing, and the media information recommendation based on the description information cannot adjust the recommended media information in time according to the change of the user's interest.

[0080] In order to solve the defects in the above media information recommendation, the present application provides a media information recommendation method based on a recommendation model, referring to Figure 1 , Figure 1 The use scenario diagram of the media information processing method based on the recommendation model provided by the embodiment of the present application is shown in Figure 1 , referring to Figure 1 , the terminal (including terminal 10-1 and terminal 10-2) is provided with a client capable of displaying corresponding different media information, such as a video playing client or plug-in. The user can obtain different media information (such as different short video information or news information) through the corresponding client and display it; the terminal is connected to the server 200 through the network 300, and the network 300 can be a wide area network or a local area network, or a combination of the two, and the data transmission is realized by using a wireless link. Among them, the server 200 can also be a node in the block chain, or a node in the cloud network to realize the storage target media information domain as a recommendation reference.

[0081] As an example, the server 200 is configured to deploy a corresponding recommendation model to implement the media information recommendation method based on the recommendation model provided by the present application. When the media information recommendation method based on the recommendation model is executed, the original features of the media information used to represent the preferences of the target object are obtained, and the domain features of the media information corresponding to the target object under each media information domain are obtained. The center sub-model is used to perform feature mapping on the original features to obtain the center features of the original features in the target semantic space. Each peripheral sub-model is used to perform feature mapping on the domain features in the corresponding media information domain to obtain the peripheral features of the domain features in the target semantic space. The determination sub-model is used to determine the target media information domain from at least two media information domains based on the center features and the peripheral features corresponding to each media information domain. The target media information domain is used as a recommendation reference to perform content recommendation of the media information domain in the cold start state of the target object, and the terminal (terminal 10-1 and / or terminal 10-2) is used to display and output the to-be-recommended media information matched with the target object. Taking short video information as an example, the recommendation model provided by the present application can be applied to short video playing. In short video playing, different short video information from different data sources is usually processed, and finally different media information corresponding to the user interface UI (User Interface) is presented. The accuracy and timeliness of the recommended media information directly affect the user experience. Taking short video information as an example, the background database of video playing receives a large amount of video data from different sources every day. Through the media information recommendation method based on the recommendation model provided by the present application, in the cold start environment, the target media information domain can be used as a recommendation reference to determine the short video matched with the target object as the short video recommended to the target object. Further, the short video obtained by the media information recommendation method can be called by other application programs (for example, the recommendation result of the short video recommendation process is migrated to the long video recommendation process or the news recommendation process). Of course, the recommendation model matched with the corresponding target object can also be migrated to different video recommendation processes (for example, the web video recommendation process, the applet video recommendation process, or the video recommendation process of the long video client).

[0082] Among them, the media information recommendation method based on the recommendation model provided by the embodiment of the application is realized based on artificial intelligence. Artificial intelligence (AI) is to use a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, obtain knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning and decision-making.

[0083] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology and machine learning / deep learning and other directions.

[0084] In the embodiment of the application, the artificial intelligence software technology mainly involved includes the above-mentioned speech processing technology and machine learning and other directions. For example, it can involve speech recognition technology (Automatic Speech Recognition, ASR) in speech technology, which includes speech signal preprocessing, speech signal frequency analysis, speech signal feature extraction, speech signal feature matching / recognition, speech training and the like.

[0085] For example, it can involve machine learning (ML), which is a multidisciplinary field that involves probability theory, statistics, approximation theory, convex analysis, algorithmic complexity theory, etc. It is a specialized study of how computers simulate or implement human learning behavior to acquire new knowledge or skills, reorganize existing knowledge structure, and continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent, and its applications are widespread in various fields of artificial intelligence. Machine learning usually includes technologies such as deep learning, which includes artificial neural networks such as convolutional neural networks (CNN), recurrent neural networks (RNN), deep neural networks (DNN), etc.

[0086] It can be understood that the media information recommendation method based on the recommendation model and the voice processing provided by the present application can be applied to an intelligent device. The intelligent device can be any device with information display function, such as a smart terminal, a smart home device (such as a smart speaker, a smart washing machine, etc.), a smart wearable device (such as a smart watch), a vehicle-mounted intelligent central control system (displaying media information to users by executing different task apps), or an AI intelligent medical device (displaying treatment cases by displaying media information), etc.

[0087] The structure of the media information recommendation device based on the recommendation model of the embodiment of the present application will be described in detail below. The media information recommendation device based on the recommendation model can be implemented in various forms, such as a special terminal with a media information recommendation processing function based on the recommendation model, or a server provided with a media information recommendation device processing function based on the recommendation model, such as the server 200 in the foregoing Figure 1 . Figure 2 The schematic diagram of the composition structure of the media information recommendation device based on the recommendation model provided by the embodiment of the present application is shown in the figure, and it can be understood that Figure 2 only an exemplary structure of the media information recommendation device based on the recommendation model is shown, not all structures, and part of the structure or all the structure shown can be implemented as needed. Figure 2

[0088] ​The media information recommendation device based on the recommendation model provided by the embodiments of the present application 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 media information recommendation device based on the recommendation model are coupled together through a bus system 205. It can be understood that the bus system 205 is used to realize the connection communication between the components. The bus system 205 includes not only a data bus, but also a power bus, a control bus and a status signal bus. However, for the purpose of clear illustration, all the buses are marked as the bus system 205 in the Figure 2

[0089] The user interface 203 can include a display, a keyboard, a mouse, a trackball, a click wheel, a key, a button, a touchpad or a touch screen, etc.

[0090] It can be understood that the memory 202 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. The memory 202 in the embodiments of the present application can store data to support the operation of the terminal (such as 10-1). Examples of the data include any computer programs for operating on the terminal (such as 10-1), such as an operating system and an application program. The operating system contains various system programs, for example, a framework layer, a core library layer, a driver layer, etc., for realizing various basic services and processing hardware-based tasks. The application program can include various application programs.

[0091] In some embodiments, the media information recommendation device based on the recommendation model provided by the embodiments of the present application can be realized in a combination of software and hardware. As an example, the media information recommendation device based on the recommendation model provided by the embodiments of the present application can be a processor in the form of a hardware decoding processor, which is programmed to execute the training method of the media information recommendation model based on the recommendation model provided by the embodiments of the present application. For example, the processor in the form of a hardware decoding processor can use one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs) or other electronic elements.

[0092] ​As an example of the media information recommendation device based on the recommendation model provided in this embodiment of the invention, which is implemented using a combination of hardware and software, the media information recommendation device based on the recommendation model provided in this embodiment of the invention can be directly embodied as a combination of software modules executed by processor 201. The software modules can be located in a storage medium, which is located in memory 202. Processor 201 reads the executable instructions included in the software modules in memory 202 and combines them with necessary hardware (e.g., including processor 201 and other components connected to bus 205) to complete the training method of the media information recommendation model based on the recommendation model provided in this embodiment of the invention.

[0093] As an example, processor 201 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0094] As an example of the hardware implementation of the media information recommendation device based on the recommendation model provided in this embodiment of the invention, the device provided in this embodiment of the invention can be directly executed by a processor 201 in the form of a hardware decoding processor. For example, it can be executed by one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components to implement the training method of the media information recommendation model based on the recommendation model provided in this embodiment of the invention.

[0095] In this embodiment of the invention, the memory 202 is used to store various types of data to support the operation of the media information recommendation device based on the recommendation model. Examples of such data include: any executable instructions for operation on the media information recommendation device based on the recommendation model, such as executable instructions that implement the training method of the media information recommendation model based on the recommendation model of this embodiment of the invention can be included in the executable instructions.

[0096] In other embodiments, the media information recommendation device based on the recommendation model provided in this invention can be implemented in software. Figure 2The recommendation model-based media information recommendation apparatus stored in the memory 202 can be software in the form of programs and plug-ins, and includes a series of modules. As an example of the programs stored in the memory 202, the recommendation model-based media information recommendation apparatus can include the following software modules:

[0097] The information transmission module 2081 and the information processing module 2082. When the software modules in the recommendation model-based media information recommendation apparatus 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 embodiments of the application is implemented, where the functions of the various software modules in the recommendation model-based media information recommendation apparatus include: an information transmission module, configured to acquire behavior parameter information of a target object in response to a recommendation model-based media information recommendation request;

[0098] The information transmission module 2081 is configured to acquire original features of media information used to represent preferences of the target object, and acquire domain features of media information corresponding to the target object under each of the media information domains.

[0099] The information processing module 2082 is configured to perform feature mapping on the original features by using the center sub-model, to obtain center features of the original features in a target semantic space.

[0100] The information processing module 2082 is configured to perform feature mapping on the domain features in the corresponding media information domain by using each of the peripheral sub-models, to obtain peripheral features of the domain features in the target semantic space.

[0101] The information processing module 2082 is configured to determine a target media information domain from at least two of the media information domains based on the center features and the peripheral features corresponding to each of the media information domains by using the determination sub-model.

[0102] The information processing module 2082 is configured to perform content recommendation of a media information domain in a cold start state for the target object with the target media information domain as a recommendation reference.

[0103] According to Figure 2 The electronic device shown in the figure, in one aspect of the present application, the present application further provides a computer program product or computer program, which includes computer instructions stored in a computer readable storage medium. The processor of the electronic device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the electronic device executes different embodiments and combinations of embodiments provided in various optional implementations of the above-mentioned recommendation model-based media information recommendation method.

[0104] In combination Figure 2 The media information recommendation device based on the recommendation model is shown to illustrate the media information recommendation method based on the recommendation model provided by the embodiments of the present application, and the media information recommendation method based on the recommendation model provided by the embodiments of the present application is described with reference to Figure 3 , Figure 3 An optional flowchart of the media information recommendation method based on the recommendation model provided by the embodiments of the present application can be understood as Figure 3 The steps shown can be executed by various electronic devices running the media information recommendation device based on the recommendation model, for example, can be a special terminal with the media information recommendation device based on the recommendation model, a server, and the following will be described with reference to Figure 3 The steps shown.

[0105] Step 301: The media information recommendation device based on the recommendation model obtains original features of media information representing the preferences of a target object, and obtains domain features of media information corresponding to the target object under each media information domain.

[0106] In some embodiments of the present application, one media information domain can correspond to one independent application platform, for example, the independent application platform can be a news platform, an application platform and a video platform, and taking the news platform as an example, the news platform can provide a large number of articles to users to meet different reading needs of the users, and each article clicked and read by the user corresponds to title features, category features and entity features, and the domain features of the media information in the news platform can be obtained by combining the title features, the category features and the entity features.

[0107] Among them, reference Figure 4 , Figure 4 The structure of the recommendation model used in the present application includes: 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, for different target objects, each increase of a media information domain requires an increase of a peripheral sub-model, and by comparing the peripheral features output by each peripheral sub-model with the center features output by the center sub-model, the similarity of the peripheral features and the center features can be determined to achieve the screening of different peripheral features.

[0108] In some embodiments of the present application, when obtaining the original features of the media information representing the preferences of the target object, the user behavior data of the corresponding client matched by different program components can be collected, and the original logs of the user behavior data can be effectively extracted, such as the device number (user account) of the user, the media information type information, the browsing time length of the media information, and the browsing completeness parameter of the media information. Among them, the user's historical click behavior and the browsing time length of the corresponding information are recorded by the subscription service and stored in Redis. The online recommendation system will pull the historical click behavior of the corresponding user when the user requests.

[0109] In some embodiments of the present application, the original features of the media information representing the preferences of the target object can be obtained by the following method:

[0110] Obtain the historical query operation information of the target object in the upstream media information domain; wherein the upstream media information domain includes a plurality of media information domains; based on the historical query operation information, determine the media information triggered by the target object among the plurality of media information responding to the historical query operation information; extract features from the historical query operation information to obtain corresponding query features, and extract features from the media information triggered by the target object to obtain corresponding media features; splice the query features and the media features to obtain the original features of the media information representing the preferences of the target object. Wherein the upstream media information domain corresponds to the upstream platform, and the downstream media information domain corresponds to the downstream platform. The user can log in and use the downstream platform through the login information in the upstream platform. For example, the upstream platform can be a content search applet and an information browsing applet triggered in the instant messaging client. Through the login account of the instant messaging client, the same user can log in to the news client of the downstream platform to click and browse the news.

[0111] Reference Figure 5 , Figure 5 The original feature acquisition diagram in the embodiments of the present application is shown in the figure, wherein the content search and browsing platform is taken as the upstream platform, and the news platform, the application platform, and the video platform are taken as the downstream platforms with text content, mobile software, and video content as the business carriers. The downstream platforms share the account of the target object in the upstream platform, and the upstream platform does not share the account of the target object in the downstream platform. Through the account of the target object in the upstream platform, the news platform, the application platform, and the video platform can be logged in to perform media information business of text content, mobile software, and video content types respectively. The news platform, the application platform, and the video platform can be different media information domains, and the original features of the media information representing the preferences of the target object are obtained in the content search and browsing platform. Specifically, Figure 5The user interface shown includes a display interface for using a search engine in a corresponding software process in the first person perspective. Through the display interface, a display control component can control the display of search results matching the search terms input by the user. For example, the user inputs a query request in the search box 501 in the instant messaging client process by inputting the search term "sports information", and the search results provided are various types of search results related to "sports information". Among the various types of search results presented, the user can browse the media information by clicking on any search result link 502. Through the user's historical behavior data, the historical search query requests and the links clicked after each request can be obtained. First, the request and the link are subjected to a symbol normalization operation, i.e., normalization, stem extraction, and unit segmentation are performed on the request in sequence; and domain-level contraction is performed on the link to reduce the feature dimension. Then, the important features with higher frequency are retained using information retrieval weighting (TF-IDF term frequency-inverse document frequency). The behavior features extracted from the two types of historical records are concatenated, i.e., the original features f u q ||f url .

[0112] As shown in Figure 5 , when the number of media information triggered by the target object is multiple (e.g., each media information presented in Figure 5 is triggered), the multiple media information can be classified according to different media information domains to obtain a classification result; the number of media information included in each of the media information domains is determined according to the classification result; the media information included in the media information domain whose number reaches a number threshold is taken as target media information; and finally, the target media information in the media information triggered by the target object is subjected to feature extraction to obtain corresponding media features. For example, when the number threshold is 5, the various types of search results related to "sports information" include sports videos, sports applet games, and sports news. When the number of sports videos triggered is greater than or equal to 5, the media information of the sports video type is taken as the target media information.

[0113] ​In some embodiments of the present application, when acquiring the domain features of the media information corresponding to the target object in each type of media information domain, the acquisition process of the domain features of the media information is described by taking the news platform, the application platform and the video platform as different media information domains. When the media information included in the media information domain is an article, the title of the article, the category to which the article belongs and the named entity in the article are acquired. The title, the category and the named entity are respectively subjected to feature extraction to obtain the corresponding title feature, category feature and entity feature. The title feature, category feature and entity feature are spliced to obtain the domain feature of the article. Specifically, for the news platform, the client can record the click behavior of the user on the article when presenting the article. The feature representation of the article can be divided into three parts: the title of the article is taken as a text sentence, a corresponding natural language coding is performed, the classification of the news is represented by using a binary coding to obtain the category feature (there can be multiple levels of classification), and the named entity in each article is extracted by using a special named entity algorithm after coding the article content to obtain the entity feature. The representations of the three parts are concatenated to obtain the domain feature of the media information of the article

[0114] In this process, Arabic numerals are not converted into Chinese characters, only the conversion of numerals is irrelevant, such as conversion from traditional Chinese to simplified Chinese, etc. The original form of Arabic numerals in the sentence is retained, and the international system of units abbreviations connected with the numerals, such as g, kg, cm, etc. are not converted and remain in the original state. For Chinese text, the corresponding Chinese text needs to be segmented because the word in Chinese can contain complete information. Correspondingly, the Chinese text can be segmented by using the Chinese segmentation tool Jieba. Among them, "this event occurred in 2001", after segmentation, becomes "this / thing / happens / in / two / zero / zero / year". Among them, the so-called segmentation has both verb meaning and noun meaning; each segmentation is a word or phrase, that is, the smallest semantic unit with a definite meaning; for different users or different text processing model use environments, the need to divide the smallest semantic units contained therein is also different, which needs to be adjusted in time, this process is called segmentation, that is, the process of dividing the smallest semantic units; on the other hand, the smallest semantic unit obtained after division is also often referred to as segmentation, that is, the word obtained after the segmentation operation; sometimes in order to distinguish the two meanings from each other, the smallest semantic unit referred to by the latter is called a term (Term); the term is used in this application; the term corresponds to the keyword in the inverted list as the basis for indexing. For Chinese, since the word as the smallest semantic unit is often composed of different numbers of characters, there is no blank separation between words and other natural distinguishing marks in phonetic scripts, therefore, for Chinese, accurate segmentation to obtain reasonable terms is an important step.

[0115] Similarly, when performing domain feature processing of media information in an English environment, English text information is processed by word hashing (Word Hashing). This method is based on letter n-gram and mainly functions to reduce the dimension of the input vector. For example, the text information is "boy", the start and end characters are represented by # respectively, and the input is (#boy#). The word is converted into the form of letter n-gram. If n is set to 3, three groups of data (#bo, boy, oy#) can be obtained, and the three groups of data are represented by n-gram vectors.

[0116] In some embodiments of the application, when the media information included in the media information field is a video, the title of the media information, the category to which the media information belongs, and the description information of the media information are obtained; the title, the category, and the description information are respectively subjected to feature extraction to obtain corresponding title features, category features, and description features; and the title features, the category features, and the description features are spliced to obtain the field features of the media information. For an application platform, the application platform client can record the historical download behavior of the user to the application, and the instant messaging client can record the user's triggering of records of various types of applets. The features of the application can include encoding of the title information and the classification identifier, and encoding of the description features of the application (for example, the initial function description given by the publisher on the application platform is subjected to language coding), and the encoding of the title information and the classification identifier are concatenated to obtain the field features of the media information of the application

[0117] In some embodiments of the application, when the media information included in the media information field is a video, the title of the media information, the category to which the media information belongs, and the description information of the media information are obtained; the title, the category, and the description information are respectively subjected to feature extraction to obtain corresponding title features, category features, and description features; and the title features, the category features, and the description features are spliced to obtain the field features of the media information. Specifically, the video playing client records the historical viewing behavior data of the user to the video, uses the video viewing behavior data to take the title of the video as a text sentence, performs corresponding natural language coding, uses binary coding to represent the classification of the video type, obtains the category features (there can be multiple levels of classification), and uses a special named entity algorithm to extract the entity features from the named entities in each video after coding the video title. The representations of the three parts are concatenated to obtain the field features of the media information of the video

[0118] Step 302: The media information recommendation device based on the recommendation model performs feature mapping on the original features based on the center sub-model to obtain the center features of the original features in the target semantic space.

[0119] In some embodiments of the application, the center features of the original features in the target semantic space can be obtained by the following method:

[0120] determining a constraint condition corresponding to the target semantic space; performing feature mapping on the original feature according to the constraint condition through the center sub-model to obtain a center feature vector; and performing weighted processing on the center feature vector to obtain a center feature of the original feature in the target semantic space. Since at least one hidden layer used when the original feature is extracted is 300 dimensions and the constraint condition is dimension reduction to 128, the constraint condition corresponding to the target semantic space can be used to perform dimension reduction processing on the original feature when performing feature mapping, so as to map the original feature to a center feature vector with 128 dimensions, thereby reducing the calculation amount of the recommendation model and reducing the load of the corresponding hardware device.

[0121] When the center feature vector is subjected to weighted processing, the target object can include multiple media information domains in the historical query operation information in the upstream media information domain in the process of obtaining the original feature. Therefore, when the original feature representing the media information preferred by the target object is obtained, the query feature and the media feature need to be spliced, and therefore the query feature and the media feature in the original feature can correspond to different weight parameters. Therefore, after the original feature is subjected to feature mapping according to the constraint condition to obtain the center feature vector subjected to dimension reduction processing, the center feature vector is also composed of two parts in series, and the center feature of the original feature in the target semantic space can be obtained by continuing to perform weighted processing on the center feature vector. For example, the original feature f u = f q ||f url , and f q and f url are weighted by using the weight parameters of the query feature and the media feature respectively, and the original feature more consistent with the potential interest of the user can be obtained.

[0122] In step 303, the media information recommendation device based on the recommendation model performs feature mapping on the domain feature in the corresponding media information domain through each peripheral sub-model to obtain peripheral features of the domain feature in the target semantic space.

[0123] In some embodiments of the present application, performing feature mapping on the domain feature in the corresponding media information domain to obtain peripheral features of the domain feature in the target semantic space can be achieved in the following manner:

[0124] According to the type of the media information domain, a corresponding peripheral sub-model is triggered; a constraint condition corresponding to the target semantic space is determined; and feature mapping is performed on the domain feature according to the constraint condition through the peripheral sub-model to obtain a peripheral feature vector. For example, the constraint condition is dimension reduction to 128, and the peripheral feature vector is a 128-dimensional vector. Figure 6 , Figure 6This is a schematic diagram illustrating the process of obtaining peripheral and central features in an embodiment of the present invention. High-dimensional original features are obtained through steps 301 and 302 in the preceding embodiments, and used as input. The original features are then mapped using a central sub-model, and the domain features are mapped using a peripheral sub-model, reducing the 128-dimensional feature vector to a lower-dimensional space (64 or 32 dimensions). Under the same constraints, the central features of the original features in the target semantic space and the peripheral features of the domain features in the target semantic space are obtained. For example, if x represents the input vector of one side of the network, y represents the output vector, li, i = 1, ..., N-1, represents the hidden layer in the middle, and W... i Let b represent the weight matrix of the i-th layer, and b i Let represent the bias term of the i-th layer. Then, the output of the hidden layer and the output of the central submodel can be expressed as Equation 1:

[0125]

[0126] Referring to Formula 2, the following can be used: The function serves as the activation function for both the output layer and the hidden layer li:

[0127]

[0128] When sorting content, referring to Formula 3, the cosine similarity between peripheral features and central features can be used as the sorting criterion, employing both peripheral sub-models (e.g., the outer towers in a twin-tower structure) and central sub-models (e.g., the central tower in a twin-tower structure).

[0129]

[0130] Step 304: The media information recommendation device based on the recommendation model determines the target media information domain from at least two media information domains based on the central feature and the peripheral features corresponding to each media information domain through the determination sub-model.

[0131] because Figure 4 As can be seen from the model structure shown, the number of peripheral sub-models is not unique, and each peripheral sub-model corresponds to a media information domain. Therefore, determining the target media information domain from at least two media information domains can be achieved in the following way:

[0132] By using the determined sub-model, the central feature is matched with each of the peripheral features to obtain the matching degree between the central feature and each of the peripheral features; the media information domain corresponding to the peripheral feature with the highest matching degree is determined as the target media information domain. (Reference) Figure 7 , Figure 7This is a schematic diagram of an optional model structure for determining the target media information domain in an embodiment of the present invention. The recommendation model employs a tower-shaped network, including a central tower (corresponding to the central sub-model in the previous embodiment), two peripheral towers (corresponding to the peripheral sub-models in the previous embodiment), and a matching network (corresponding to the determining sub-model in the previous embodiment), wherein there is one central tower T. u and v outer towers T1-T v T i There is a domain-specific input X. i Its dimension is R di Through the processing in the preceding embodiments, the structure of each outer tower can be represented as a nonlinear mapping layer f. i (X i, W i This transformation function maps the input of the domain to a shared semantic space, resulting in Y. i .

[0133] Therefore, each tower network in the recommendation model can achieve a non-linear transformation, so that the sum of the matching degrees of the central features and the peripheral features can reach its maximum value in this semantic space, which can be expressed by Equation 4.

[0134]

[0135] Continue to refer to Figure 7 , , Axis Tower T u The original features of media information that correspond to the preferences of the target audience, the outer tower T i Under each of the aforementioned media information domains, the I of the media information corresponding to the target object i By comparing the matching degree between the central feature and each of the peripheral features pairwise, the media information domain corresponding to the peripheral feature with the highest matching degree can be determined as the target media information domain. Alternatively, the matching degree between all central features and each peripheral feature can be sorted to determine the target media information domain. This application does not impose any specific restrictions on this.

[0136] Step 305: The media information recommendation device based on the recommendation model uses the target media information domain as a recommendation reference to recommend content of the media information domain of the target object in the cold start state.

[0137] In some embodiments of the present invention, using the target media information domain as a recommendation reference, content recommendation for the media information domain of the target object in a cold start state can be achieved in the following ways:

[0138] Obtain the media information categories preferred by the target object in the target media information domain; based on the media information categories, recommend media information of the same category to the target object in the media information domain during the cold start state.

[0139] The cold start state includes: 1) the start of the client (such as the start of the short video client for short video recommendation) when the target object is a low consumption user; 2) the start of the client when the target object is a new user. The media information domain in the cold start state can be the media information domain corresponding to the client of at least one downstream platform. The target object can include one or more of a low consumption user, a new user, and a high consumption user. Specifically, the new user is a user who registers and uses the client for the first time; the low consumption user is a user who has registered and uses the short video client, but has less user behavior information; and the high consumption user is a user who has registered and uses the short video client, and generates more user behavior information during the use of the client.

[0140] For example, in the recommendation environment of the short video client, the number threshold of user behavior information can be set to 10. When the number of user behavior information of a certain user is less than or equal to 10, the user can be a low consumption user; when the number of user behavior information of a certain user is greater than 10, the user is a high consumption user. It should be noted that for different multimedia information recommendation environments, the threshold of user behavior information can be dynamically adjusted, and the embodiments of the present application do not make specific limitations.

[0141] Through the processing of the preceding steps 301-304, the target media information domain is obtained as a recommendation reference. In the recommendation environment for new users, since there is not enough information to determine the user's preference for video recommendation, when the domain feature of the article is determined as the target media information domain by the media information recommendation method based on the recommendation model provided by the present application, the domain feature of the article can be used as a recommendation reference. In video recommendation, the user is preferentially recommended video information of the same category as the domain feature of the article.

[0142] When recommending a low consumption user, although the original recommendation strategy can be determined according to the use information of the low consumption user (the user has triggered an entertainment video), since the number of samples is small, the original recommendation strategy of the low consumption user can be ignored. When the domain feature of the article is determined as the target media information domain by the media information recommendation method based on the recommendation model provided by the present application, the domain feature of the article can be used as a recommendation reference. In video recommendation, the user is preferentially recommended video information of the same category as the domain feature of the article (such as video information of the finance category), avoiding the error recommendation strategy caused by the small number of samples, and improving the user experience.

[0143] In combination with Figure 2 The media information recommendation device shown illustrates the media information recommendation method based on the recommendation model provided by the embodiments of the present application, which is described with reference toFigure 8 , Figure 8 An optional flowchart of the training method of the recommendation model provided by the embodiments of the present application can be understood as follows, Figure 8 The steps shown in the figure can be executed by various electronic devices running the media information recommendation apparatus, such as a dedicated terminal with the media information recommendation apparatus, a server or a server cluster, wherein the dedicated terminal with the media information recommendation apparatus can be the electronic device with the media information recommendation apparatus shown in the foregoing Figure 2 embodiments. The following describes the steps shown in the figure. Figure 8

[0144] Step 801: Construct a training sample, which includes sample original features for representing media information preferences of a training object and sample domain features of media information corresponding to the training object in at least two media information domains.

[0145] Wherein, the training sample is labeled with sample matching degrees of each sample domain feature and the sample original feature. When constructing the training sample set, it is necessary to obtain the intersection of the upstream user space U h and the downstream platform user space The users in each intersection constitute a user set specific to the downstream platform actually used for training of the model

[0146] Step 802: Perform feature mapping on the sample original features by the center sub-model to obtain sample center features of the sample original features in a target semantic space.

[0147] In order to reduce the number of samples in the training sample set, the number of samples in the training sample set can be controlled in any of the following ways to reduce the training cost of the recommendation model.

[0148] 1) Reduce the number of samples by common feature screening, which is specifically manifested as: selecting the top K most common features, for example, all features with a probability greater than 0.001 in the target object, which can represent common online behaviors of users.

[0149] 2) Extract features by a clustering algorithm, which is specifically manifested as using a vector fi of length |U| to represent each feature, wherein |U| is the number of target objects in the training data, fi i (j) is the number of times that the user j contains the feature i, and the vector is normalized, and then the K-means algorithm is executed on all feature vectors, and the number of clusters |F| is set according to the specific circumstances of the features. After clustering, a new vector of length |F| is assigned to each new feature The new feature is represented as: ​

[0150] wherein Cls(i) represents the class to which the original feature i belongs after clustering.

[0151] 3) Extract features by local sensitive hashing, which is specifically represented as: configuring a transformation matrix A e R d*k , wherein d is the number of features in the original space, and k is the dimension of the low-dimensional space. Multiplying the original feature by the transformation matrix, i.e. X·A=Y e R k , obtains the new feature Y in the converted low-dimensional space. Then, replacing the negative values in Y with zero, i.e. Y (i) =max(Y (i) ,0). The features X1,X2 e R d in the original space can be approximated as , wherein H(Y1,Y2) is the Hamming distance of the local sensitive hashing of the two original features.

[0152] Step 803: respectively mapping the sample domain features in the corresponding media information domain to the target semantic space by the peripheral sub-models to obtain sample peripheral features of the sample domain features in the target semantic space.

[0153] Step 804: respectively determining the matching degrees of the sample peripheral features and the sample center features by the determination sub-model.

[0154] Step 805: obtaining the differences between the determined matching degrees and the corresponding sample matching degrees, and updating the model parameters of the recommendation model based on the differences.

[0155] The softmax function is performed on the recommendation score of each user to the candidate content, and the normalization processing is performed in the following manner,

[0156]

[0157] wherein γ is a smoothing factor in the sotfmax function, which needs to be adjusted through experiments; I l is the candidate set to be sorted. It should be noted that, in the training process, ideally, I should contain all possible content in the field. However, in actual operation, the union of all interactive content and N randomly sampled non-interactive content l is usually used to approximate I .

[0158] Specifically, the error between the predicted matching probability and the actual matching probability between the original feature and the domain feature can be determined each time the parameter is updated, and the model parameters of different sub-models (peripheral sub-model, center sub-model and determination sub-model) in the recommendation model are adjusted according to the error value, until the predicted matching probability and the actual matching probability between the original feature and the domain feature are the same.

[0159] Specifically, a value of the loss function is determined based on the difference, and whether the value of the loss function exceeds a preset threshold is determined, when the value of the loss function exceeds the preset threshold, an error signal of the recommendation model is determined based on the loss function, the error signal is back propagated in the recommendation model, and model parameters of each layer are updated in the process of propagation.

[0160] Here, the back propagation is described, the training sample 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, since there is an error between the output result of the neural network model and the actual result, the error between the output result and the actual value is calculated, and the error is back propagated from the output layer to the hidden layer until it propagates to the input layer, in the process of back propagation, the value of the model parameter is adjusted according to the error; the above process is iterated continuously until convergence.

[0161] In combination with Table 1, the training process of the recommendation model is described by taking a tower network as an example, wherein the tower network includes an axis tower (corresponding to the center sub-model in the previous embodiment), at least two peripheral towers (corresponding to the peripheral sub-model in the previous embodiment) and a matching network (corresponding to the determination sub-model in the previous embodiment), and the training process is shown in Table 1,

[0162]

[0163] Therefore, the trained recommendation model can be encapsulated in the corresponding APP or saved in the instant messaging client in the form of a plug-in to realize the recommendation of different media information.

[0164] It can be understood that in the embodiments of the present application, the data related to user information, user behavior data, and the click behavior of target objects are involved, when the embodiments of the present application are applied to specific products or technologies, the user permission or consent needs to be obtained, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0165] Beneficial technical effects:

[0166] The application obtains original features of media information for characterizing the preference of a target object, and obtains domain features of media information corresponding to the target object under each media information domain; the central sub-model is used to perform feature mapping on the original features to obtain central features of the original features in a target semantic space; each peripheral sub-model is used to perform feature mapping on the domain features under the corresponding media information domain to obtain peripheral features of the domain features in the target semantic space; the determination sub-model is used to determine a target media information domain from at least two media information domains based on the central features and the peripheral features corresponding to each media information domain; thus, the target media information domain can be used as a recommendation reference to perform content recommendation of the media information domain in a cold start state for the target object, 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 user experience is improved.

[0167] The above merely describes the embodiments of the application, but is not used to limit the protection scope of the application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A method for recommending media information based on a recommendation model, characterized in that, The recommendation model comprises a center sub-model, a determination sub-model and at least two peripheral sub-models, each of the peripheral sub-models corresponds to a media information domain, one of the media information domains corresponds to one application platform, and the method comprises: obtaining original features of media information for representing preferences of a target object, and obtaining domain features of media information corresponding to the target object under each of the media information domains; performing feature mapping on the original features by the center sub-model to obtain center features of the original features in a target semantic space; performing feature mapping on the domain features in the corresponding media information domain by each of the peripheral sub-models to obtain peripheral features of the domain features in the target semantic space; determining a target media information domain from at least two of the media information domains based on the center features and the peripheral features corresponding to each of the media information domains by the determination sub-model; performing content recommendation of a media information domain in a cold start state for the target object with the target media information domain as a recommendation reference.

2. The method of claim 1, wherein, The determination of the target media information domain from at least two of the media information domains based on the center features and the peripheral features corresponding to each of the media information domains by the determination sub-model comprises: matching the center features with each of the peripheral features by the determination sub-model to obtain matching degrees of the center features and each of the peripheral features; determining the media information domain corresponding to the peripheral feature with the largest matching degree as the target media information domain.

3. The method of claim 1, wherein, The content recommendation of the media information domain in the cold start state for the target object with the target media information domain as the recommendation reference comprises: obtaining a media information category preferred by the target object in the target media information domain; performing media information recommendation of the same category for the target object in the media information domain in the cold start state based on the media information category.

4. The method of claim 1, wherein, The obtaining of the original features of the media information for representing the preferences of the target object comprises: obtaining historical query operation information of the target object in an upstream media information domain; wherein the upstream media information domain comprises a plurality of the media information domains; determining media information triggered by the target object from a plurality of media information for responding to the historical query operation information based on the historical query operation information; performing feature extraction on the historical query operation information to obtain corresponding query features, and performing feature extraction on the media information triggered by the target object to obtain corresponding media features; splicing the query features and the media features to obtain the original features of the media information for representing the preferences of the target object.

5. The method of claim 4, wherein, Before the feature extraction on the historical query operation information, the method further comprises: when the number of the media information triggered by the target object is a plurality, classifying the plurality of the media information in different media information domains to obtain a classification result; determining the number of media information included in each of the media information domains according to the classification result; taking the media information included in a media information domain with a number reaching a number threshold as target media information. The media information triggered by the target object is subjected to feature extraction to obtain corresponding media features. The media information triggered by the target object is subjected to feature extraction to obtain corresponding media features.

6. The method of claim 1, wherein, The domain features of the media information corresponding to the target object in each of the media information domains are obtained. When the media information included in the media information domain is an article, the title of the article, the category to which the article belongs, and the named entity in the article are obtained. The title, the category, and the named entity are subjected to feature extraction to obtain corresponding title features, category features, and entity features. The title features, the category features, and the entity features are spliced to obtain the domain features of the article.

7. The method of claim 1, wherein, The domain features of the media information corresponding to the target object in each of the media information domains are obtained. When the media information included in the media information domain is a video, the title of the media information, the category to which the media information belongs, and the description information of the media information are obtained. The title, the category, and the description information are subjected to feature extraction to obtain corresponding title features, category features, and description features. The title features, the category features, and the description features are spliced to obtain the domain features of the media information.

8. The method of claim 1, wherein, The domain features of the media information corresponding to the target object in each of the media information domains are obtained. When the media information included in the media information domain is an application, the title of the media information, the category to which the media information belongs, and the description information of the media information are obtained through the download record of the media information. The title, the category, and the description information are subjected to feature extraction to obtain corresponding title features, category features, and description features. The title features, the category features, and the description features are spliced to obtain the domain features of the media information.

9. The method of claim 4, wherein, The original features are subjected to feature mapping through the center sub-model to obtain the center features of the original features in the target semantic space, including: A constraint condition corresponding to the target semantic space is determined. The original features are subjected to feature mapping according to the constraint condition through the center sub-model to obtain a center feature vector. The center feature vector is subjected to weighted processing to obtain the center features of the original features in the target semantic space.

10. The method of claim 1, wherein, The domain features in the corresponding media information domain are subjected to feature mapping through each of the peripheral sub-models to obtain peripheral features of the domain features in the target semantic space, including: According to the type of the media information domain, a corresponding peripheral sub-model is triggered. A constraint condition corresponding to the target semantic space is determined. The domain features are subjected to feature mapping according to the constraint condition through the peripheral sub-model to obtain a peripheral feature vector.

11. The method of claim 1, wherein, The method further includes: Training samples are constructed, including sample original features for representing media information preferred by a training object, and sample domain features of media information corresponding to the training object in at least two media information domains. The training samples are labeled with sample matching degrees of the sample domain features and the sample original features; The sample original features are mapped by the center sub-model to obtain sample center features of the sample original features in a target semantic space; The sample domain features in the corresponding media information domains are mapped by the peripheral sub-models to obtain sample peripheral features of the sample domain features in the target semantic space; The matching degrees of the sample peripheral features and the sample center features are determined by the determination sub-model; The differences between the determined matching degrees and the corresponding sample matching degrees are obtained, and the model parameters of the recommendation model are updated based on the differences. 12.A media information recommendation apparatus based on a recommendation model, characterized by 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 corresponds to a media information domain, and one media information domain corresponds to one application platform, and the device includes: An information transmission module is configured to obtain original features of media information representing a preference of a target object, and obtain domain features of media information corresponding to the target object in each media information domain; An information processing module is configured to map the original features by the center sub-model to obtain center features of the original features in a target semantic space; The information processing module is configured to map the domain features in the corresponding media information domains by the peripheral sub-models to obtain peripheral features of the domain features in the target semantic space; The information processing module is configured to determine a target media information domain from at least two media information domains based on the center features and the peripheral features corresponding to each media information domain by the determination sub-model; The information processing module is configured to recommend content of a media information domain in a cold start state of the target object by taking the target media information domain as a recommendation reference.

13. The apparatus of claim 12, wherein, The information processing module is further configured to: Match the center features with each of the peripheral features by the determination sub-model to obtain matching degrees of the center features and each of the peripheral features, and determine a media information domain corresponding to the peripheral feature with the largest matching degree as the target media information domain.

14. The apparatus of claim 12, wherein, The information processing module is further configured to: Obtain a media information category preferred by the target object in the target media information domain, and recommend media information of the same category to the target object in the media information domain in the cold start state based on the media information category.

15. The apparatus of claim 12, wherein, The information processing module is further configured to: obtaining historical query operation information of the target object in an upstream media information domain; wherein the upstream media information domain comprises a plurality of the media information domains; determining, based on the historical query operation information, media information triggered by the target object from a plurality of media information used to respond to the historical query operation information; performing feature extraction on the historical query operation information to obtain corresponding query features, and performing feature extraction on the media information triggered by the target object to obtain corresponding media features; and concatenating the query features and the media features to obtain original features of media information representing preferences of the target object.

16. The apparatus of claim 15, wherein, The information processing module is further configured to: when the number of media information triggered by the target object is a plurality, classifying the plurality of media information according to different media information domains to obtain a classification result; determining the number of media information included in each media information domain according to the classification result; and taking media information included in a media information domain whose number reaches a number threshold as target media information. The feature extraction on the media information triggered by the target object to obtain corresponding media features comprises feature extraction on the target media information in the media information triggered by the target object to obtain corresponding media features.

17. The apparatus of claim 12, wherein, The information processing module is further configured to: when the media information included in the media information domain is an article, obtaining a title of the article, a category to which the article belongs, and a named entity in the article; performing feature extraction on the title, the category, and the named entity respectively to obtain corresponding title features, category features, and entity features; and concatenating the title features, the category features, and the entity features to obtain a domain feature of the article.

18. A computer program product comprising computer programs or instructions, characterized in that, The computer program or instructions, when executed by a processor, implement the media information recommendation method based on a recommendation model according to any one of claims 1 to 11.

19. An electronic device, comprising: The electronic device comprises: a memory configured to store executable instructions; a processor configured to execute the executable instructions stored in the memory to implement the media information recommendation method based on a recommendation model according to any one of claims 1 to 11.

20. A computer-readable storage medium storing executable instructions, the instructions causing a computer to perform operations comprising: The executable instructions, when executed by a processor, implement the media information recommendation method based on a recommendation model according to any one of claims 1 to 11.

Images