Media information recommendation method and device, equipment, storage medium and program product

Abstract

The invention provides a media information recommendation method and device, equipment, a storage medium and a program product. The method comprises the steps that media information features corresponding to multiple pieces of media information to be recommended and object features of objects to be recommended are acquired; performing feature intersection on the media information features corresponding to the to-be-recommended media information and the object features to obtain intersection features corresponding to the to-be-recommended media information; for the cross feature corresponding to each piece of to-be-recommended media information, performing importance degree evaluation on each feature element in the cross feature to obtain the feature importance degree of each feature element, and determining a first recommendation parameter value of the to-be-recommended media information in combination with the feature importance degree of each feature element and the element value of each feature element; and based on the first recommendation parameter value, selecting target media information from the multiple pieces of media information to be recommended, and recommending the target media information to the object to be recommended. According to the invention, the recommendation accuracy of the media information can be effectively improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device, equipment, storage medium and program product for recommending media information. Background Art

[0002] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. Artificial intelligence technology is an interdisciplinary subject that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies.

[0003] In the related art, for the recommendation of media information, the object features of the object to be recommended are usually constructed based on the basic information of the object to be recommended, such as age and gender, and the constructed object features of the object to be recommended are directly fused with the media information features to recommend the media information. In this way, due to the different importance of different feature elements, ignoring the importance of feature elements will lead to low recommendation accuracy. Summary of the invention

[0004] The embodiments of the present application provide a method, device, electronic device, computer-readable storage medium, and computer program product for recommending media information, which can effectively improve the accuracy of recommending media information.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] The present application provides a method for recommending media information, including:

[0007] Acquire media information features corresponding to the plurality of media information to be recommended, and object features of the object to be recommended;

[0008] Performing feature crossover on the media information features corresponding to each of the media information to be recommended and the object features to obtain crossover features corresponding to each of the media information to be recommended, wherein the crossover features include at least two feature elements;

[0009] For each cross-feature corresponding to each of the media information to be recommended, an importance evaluation is performed on each feature element in the cross-feature to obtain a feature importance of each feature element, and a first recommendation parameter value of the media information to be recommended is determined in combination with the feature importance of each feature element and an element value of each feature element;

[0010] Based on the first recommendation parameter value, target media information is selected from the plurality of media information to be recommended, and the target media information is recommended to the object to be recommended.

[0011] The present application provides a media information recommendation device, including:

[0012] An acquisition module, used to acquire media information features respectively corresponding to a plurality of media information to be recommended, and object features of an object to be recommended;

[0013] A feature crossover module, used to perform feature crossover on the media information features corresponding to each of the media information to be recommended and the object features, respectively, to obtain crossover features corresponding to each of the media information to be recommended, wherein the crossover features include at least two feature elements;

[0014] An evaluation module, configured to evaluate the importance of each characteristic element in the cross-characteristic corresponding to each of the media information to be recommended, obtain the characteristic importance of each characteristic element, and determine the first recommended parameter value of the media information to be recommended in combination with the characteristic importance of each characteristic element and the element value of each characteristic element;

[0015] The recommendation module is configured to select target media information from the plurality of media information to be recommended based on the first recommendation parameter value, and recommend the target media information to the object to be recommended.

[0016] In the above scheme, the recommendation device for the above media information also includes: a prediction module, which is used to predict the recommended parameter value of each media information to be recommended based on the object characteristics and the characteristics of each media information, and obtain the second recommended parameter value of each media information to be recommended; the above recommendation module is also used to determine the recommended value of the media information to be recommended for each media information to be recommended, in combination with the corresponding first recommended parameter value and the second recommended parameter value; the media information to be recommended whose recommendation value is greater than the recommendation threshold is determined as the target media information.

[0017] In the above scheme, the above prediction module is also used to perform the following processing for each of the media information features: feature splicing the object feature and the media information feature to obtain the splicing feature corresponding to the media information feature; obtaining the recommended parameter value prediction network, calling the recommended parameter value prediction network, and predicting the recommended parameter value of the media information feature based on the splicing feature to obtain the second recommended parameter value of the media information feature.

[0018] In the above scheme, the above prediction module is also used to obtain object feature samples and media information feature samples, and perform feature splicing on the object feature samples and the media information feature samples to obtain spliced ​​feature samples; obtain an initial recommended parameter value prediction network and a label recommended parameter value, and call the initial recommended parameter value prediction network to predict the recommended parameter value of the media information feature sample based on the spliced ​​feature sample to obtain the predicted recommended parameter value of the media information feature sample; and train the initial recommended parameter value prediction network in combination with the label recommended parameter value and the predicted recommended parameter value to obtain the recommended parameter value prediction network.

[0019] In the above scheme, the above recommendation module is also used to perform the following processing for each of the media information to be recommended: summing the first recommendation parameter value and the second recommendation parameter value corresponding to the media information to be recommended to obtain a summed parameter value, and determining the opposite of the summed parameter value as the target parameter value; normalizing the target parameter value to obtain the recommended value of the media information to be recommended.

[0020] In the above scheme, the above-mentioned feature intersection includes independent feature intersection and mixed feature intersection. The above-mentioned feature intersection module is also used to perform the following processing respectively for the media information features corresponding to each of the media information to be recommended: perform independent feature intersection on the media information features and the object features respectively to obtain a first cross feature value and a second cross feature value; perform mixed feature intersection on the media information features and the object features to obtain a third cross feature value; perform feature value splicing on the first cross feature value, the second cross feature value and the third cross feature value to obtain the cross feature corresponding to the media information to be recommended.

[0021] In the above scheme, the feature cross module is further used to sum the element values ​​of each feature element in the media information feature to obtain the first cross feature value; and sum the element values ​​of each feature element in the object feature to obtain the second cross feature value.

[0022] In the above scheme, the above-mentioned feature cross-module is also used to obtain the standard feature dimension, and according to the standard feature dimension, dimensionally standardize the media information feature and the object feature respectively, to obtain the target media information feature and the target object feature of the standard feature dimension; and determine the inner product of the target media information feature and the target object feature as the third cross-feature.

[0023] In the above scheme, the above-mentioned feature intersection module is also used to obtain the feature dimension of the media information feature, and when the feature dimension of the media information feature is not equal to the standard feature dimension, the feature dimension of the media information feature is adjusted to obtain the target media information feature of the standard feature dimension; obtain the feature dimension of the object feature, and when the feature dimension of the object feature is not equal to the standard feature dimension, the feature dimension of the object feature is adjusted to obtain the target object feature of the standard feature dimension.

[0024] In the above scheme, the above evaluation module is also used to compress the cross-features to obtain compressed cross-features, wherein the number of feature elements in the compressed cross-features is less than the number of feature elements in the cross-features; stretch the compressed cross-features to obtain importance features, wherein the number of feature elements in the importance features is equal to the number of feature elements in the cross-features, and the feature elements in the importance features correspond one-to-one to the feature elements in the cross-features; for each feature element in the importance feature, the element value of the feature element in the importance feature is determined as the feature importance of the corresponding feature element in the cross-feature.

[0025] In the above scheme, the above evaluation module is also used to obtain a compressed feature for reducing the dimension of the cross feature, and multiply the cross feature and the compressed feature to obtain a reference compressed feature, the number of feature elements in the reference compressed feature is less than the number of feature elements in the cross feature; and perform feature transformation on the reference compressed feature to obtain the compressed cross feature.

[0026] In the above scheme, the above evaluation module is also used to obtain a stretching feature for upgrading the dimension of the compressed cross feature, and multiply the stretching feature and the compressed cross feature to obtain a reference importance feature, the number of feature elements in the reference importance feature is equal to the number of feature elements in the cross feature; perform feature transformation on the reference importance feature to obtain the importance feature.

[0027] An embodiment of the present application provides an electronic device, including:

[0028] A memory for storing computer executable instructions or computer programs;

[0029] The processor is used to implement the media information recommendation method provided in the embodiment of the present application when executing the computer executable instructions or computer programs stored in the memory.

[0030] An embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions for causing a processor to execute instructions to implement a method for recommending media information provided in an embodiment of the present application.

[0031] The embodiment of the present application provides a computer program product, which includes a computer program or a computer executable instruction, and the computer program or the computer executable instruction is stored in a computer-readable storage medium. The processor of the electronic device reads the computer executable instruction from the computer-readable storage medium, and the processor executes the computer executable instruction, so that the electronic device executes the media information recommendation method described in the embodiment of the present application.

[0032] The embodiments of the present application have the following beneficial effects:

[0033] By acquiring media information features corresponding to a plurality of media information to be recommended and object features of an object to be recommended, the media information features corresponding to each piece of media information to be recommended are cross-featured with the object features to obtain cross-features corresponding to each piece of media information to be recommended, and for each cross-feature, importance evaluation is performed on each feature element in the cross-feature to obtain feature importance of the feature element, and a first recommendation parameter value of the media information to be recommended is determined in combination with the feature importance of the feature element and the element value of each feature element, and based on the first recommendation parameter value, target media information is selected from a plurality of media information to be recommended, and the target media information is recommended to the object to be recommended. In this way, by performing feature crossover between the media information features corresponding to each media information to be recommended and the object features respectively, the cross features corresponding to each media information to be recommended are obtained, thereby integrating the object features into each media information to be recommended, and obtaining the cross features corresponding to each media information to be recommended, so that the cross features can comprehensively reflect the interactive characteristics of the media information to be recommended and the object features, and by evaluating the importance of each feature element in the cross features, the target media information recommended to the object to be recommended is determined based on the feature importance of the feature elements obtained, and because the feature importance comprehensively reflects the importance of the feature elements in the cross features, the determined first recommended parameter value can fully consider the influence of each feature element in the cross features on the first recommended parameter value, and effectively reduce the influence of invalid feature elements in the cross features on the first recommended parameter value, so that the determined first recommended parameter value is more accurate, and the selected target media information is more accurate, which effectively improves the recommendation accuracy of the media information. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the architecture of the media information recommendation system provided in the embodiment of the present application;

[0035] Figure 2 is a schematic diagram of the structure of an electronic device for recommending media information provided by an embodiment of the present application;

[0036] Figure 3 It is a flowchart of a method for recommending media information provided in an embodiment of the present application;

[0037] Figure 4 It is a schematic diagram of the principle of the method for recommending media information provided in an embodiment of the present application;

[0038] Figures 5 to 10 It is a flowchart of a method for recommending media information provided in an embodiment of the present application;

[0039] Fig.11 It is a schematic diagram of an interface of a method for recommending media information provided in an embodiment of the present application;

[0040] Fig.12 It is a schematic diagram of the principle of the method for recommending media information provided in an embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0042] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0043] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described here can be implemented in an order other than that illustrated or described here.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0045] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0046] 1) Machine Learning (ML): It is a multi-disciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance.

[0047] 2) Artificial Intelligence (AI): It is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. Artificial intelligence technology is an interdisciplinary subject that covers a wide range of fields, including both hardware-level technology and software-level technology. Basic artificial intelligence technologies generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies.

[0048] 3) Convolutional Neural Networks (CNN): It is a type of feed forward neural network (FNN) that includes convolution calculations and has a deep structure. It is one of the representative algorithms of deep learning. Convolutional neural networks have representation learning capabilities and can perform shift-invariant classification on input images according to their hierarchical structure.

[0049] 4) In response to: used to indicate the conditions or states on which the executed operations depend. When the dependent conditions or states are met, one or more operations executed may be in real time or have a set delay. Unless otherwise specified, there is no restriction on the order in which the multiple operations executed are executed.

[0050] 5) Convolutional layer: Each convolutional layer in a convolutional neural network consists of several convolutional units, and the parameters of each convolutional unit are optimized through the back propagation algorithm. The purpose of the convolution operation is to extract different features of the input. The first convolutional layer may only extract some low-level features such as edges, lines, and corners. More layers of the network can iteratively extract more complex features from low-level features.

[0051] 6) Pooling layer: After the convolution layer performs feature extraction, the output feature map is passed to the pooling layer for feature selection and information filtering. The pooling layer contains a pre-set pooling function, which replaces the result of a single point in the feature map with the feature map statistics of its adjacent area. The pooling layer selects the pooling area in the same way as the convolution kernel scans the feature map, which is controlled by the pooling size, step size, and padding.

[0052] 7) Fully-Connected Layer: The fully connected layer in a convolutional neural network is equivalent to the hidden layer in a traditional feedforward neural network. The fully connected layer is located at the end of the hidden layer of the convolutional neural network and only transmits signals to other fully connected layers. The feature map loses its spatial topological structure in the fully connected layer, is expanded into a vector and passes through the activation function.

[0053] 8) Factor Machine (FM): The algorithm is a machine learning algorithm based on matrix decomposition. In order to solve the problem of feature combination in large-scale sparse data, the FM algorithm is one of the recommendation algorithms with better results in the recommendation field and is widely used in various recommendation fields.

[0054] During the implementation of the embodiments of the present application, the applicant discovered that the related technology has the following problems:

[0055] In the related art, for the recommendation of media information, the object features of the object to be recommended are usually constructed based on the basic information of the object to be recommended, such as age and gender, and the constructed object features of the object to be recommended are directly fused with the media information features to recommend the media information. In this way, due to the different importance of different feature elements, ignoring the importance of feature elements will lead to low recommendation accuracy.

[0056] The embodiments of the present application provide a method, device, electronic device, computer-readable storage medium, and computer program product for recommending media information, which can effectively improve the accuracy of media information recommendation. The following describes an exemplary application of the media information recommendation system provided by the embodiments of the present application.

[0057] See also Figure 1 , Figure 1 1 is a schematic diagram of the architecture of a media information recommendation system 100 provided in an embodiment of the present application. A terminal (terminal 400 is shown as an example) is connected to a server 200 via a network 300. The network 300 may be a wide area network or a local area network, or a combination of the two.

[0058] The terminal 400 is used for the user to use the client 410 to display target media information on a graphical interface 410-1 (graphic interface 410-1 is shown as an example). The terminal 400 and the server 200 are connected to each other via a wired or wireless network.

[0059] In some embodiments, the server 200 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. The terminal 400 may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart TV, a smart watch, a car terminal, etc., but is not limited thereto. The electronic device provided in the embodiment of the present application may be implemented as a terminal or as a server. The terminal and the server may be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiment of the present application.

[0060] In some embodiments, the server 200 obtains media information features corresponding to multiple media information to be recommended and object features of the object to be recommended, cross-features the media information features corresponding to the media information to be recommended with the object features to obtain cross-features, selects target media information based on the cross-features, and recommends the target media information to the object to be recommended.

[0061] In other embodiments, the terminal 400 obtains media information features corresponding to multiple media information to be recommended and object features of the object to be recommended, cross-features the media information features corresponding to the media information to be recommended with the object features to obtain cross-features, selects target media information based on the cross-features, and recommends the target media information to the object to be recommended.

[0062] In other embodiments, the embodiments of the present application can be implemented with the aid of cloud technology. Cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software, and networks within a wide area network or a local area network to achieve data computing, storage, processing, and sharing.

[0063] Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, and application technology based on the cloud computing business model. It can form a resource pool that can be used on demand and is flexible and convenient. Cloud computing technology will become an important support. The background services of the technical network system require a large amount of computing and storage resources.

[0064] See also Figure 2 , Figure 2 is a schematic diagram of the structure of an electronic device 500 for recommending media information provided by an embodiment of the present application, wherein: Figure 2 The electronic device 500 shown may be Figure 1The server 200 or the terminal 400 in Figure 2 The electronic device 500 shown includes: at least one processor 410, a memory 450, and at least one network interface 420. The various components in the electronic device 500 are coupled together via a bus system 440. It is understood that the bus system 440 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 440 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, the bus system 440 is not described in detail. Figure 2 Various buses are labeled as bus system 440 .

[0065] Processor 410 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., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0066] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard disk drives, optical disk drives, etc. The memory 450 may optionally include one or more storage devices that are physically remote from the processor 410.

[0067] The memory 450 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 450 described in the embodiments of the present application is intended to include any suitable type of memory.

[0068] In some embodiments, memory 450 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplarily described below.

[0069] Operating system 451, including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;

[0070] The network communication module 452 is used to reach other electronic devices via one or more (wired or wireless) network interfaces 420. Exemplary network interfaces 420 include: Bluetooth, Wireless Fidelity (WiFi), and Universal Serial Bus (USB).

[0071] In some embodiments, the device for recommending media information in the embodiments of the present application may be implemented in software. Figure 2 The media information recommendation device 455 stored in the memory 450 is shown, which can be software in the form of a program and a plug-in, etc., and includes the following software modules: an acquisition module 4551, a feature cross module 4552, an evaluation module 4553, and a recommendation module 4554. These modules are logical, and therefore can be arbitrarily combined or further split according to the functions implemented. The functions of each module will be described below.

[0072] In other embodiments, the media information recommendation device provided in the embodiments of the present application can be implemented in hardware. As an example, the media information recommendation device provided in the embodiments of the present application can be a processor in the form of a hardware decoding processor, which is programmed to execute the media information recommendation method provided in the embodiments of the present application. For example, the processor in the form of a hardware decoding processor can adopt 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.

[0073] In some embodiments, the terminal or server can implement the media information recommendation method provided in the embodiments of the present application by running a computer program or a computer executable instruction. For example, the computer program can be a native program (e.g., a dedicated information recommendation program) or a software module in the operating system, for example, an information recommendation module that can be embedded in any program (such as an instant messaging client, an album program, an electronic map client, a navigation client); for example, it can be a native application (APP, Application), that is, a program that needs to be installed in the operating system to run. In short, the above-mentioned computer program can be any form of application, module or plug-in.

[0074] The method for recommending media information provided by the embodiment of the present application will be described in conjunction with the exemplary application and implementation of the server or terminal provided by the embodiment of the present application.

[0075] See also Figure 3 , Figure 3 is a flowchart of a method for recommending media information provided by an embodiment of the present application, which will be combined with Figure 3 Steps 101 to 105 are shown for illustration. The media information recommendation method provided in the embodiment of the present application can be implemented by a server or a terminal alone, or by a server and a terminal in collaboration. The following description will be made by taking the server alone as an example.

[0076] In step 101, media information features respectively corresponding to a plurality of media information to be recommended and object features of an object to be recommended are obtained.

[0077] In some embodiments, the media information to be recommended includes recommendation information of various media types such as video information, audio information, and text information.

[0078] In some embodiments, the number of objects to be recommended obtained in the above step 101 is at least one. The media information recommendation method provided in the embodiment of the present application will be described below from the perspective of one object to be recommended. It can be understood that when the number of objects to be recommended is multiple, the above steps 101 to 105 of the media information recommendation method provided in the embodiment of the present application can be executed in parallel.

[0079] In some embodiments, the object to be recommended may be a client of a user to whom media information needs to be recommended. The user to be recommended may be a user among sample users or a new user. Any user to whom an item needs to be recommended may be considered a user to be recommended.

[0080] In some embodiments, the above step 101 can be implemented in the following ways: obtain multiple media information to be recommended, and for each piece of media information to be recommended, perform feature extraction on the media information to be recommended to obtain media information features corresponding to the media information to be recommended; obtain object information of the object to be recommended, perform feature extraction on the object information, and obtain object features of the object to be recommended.

[0081] In some embodiments, the above-mentioned object features are used to indicate features corresponding to the interactive behaviors of the to-be-recommended object at various historical moments.

[0082] As an example, see Figure 4 , Figure 4It is a schematic diagram of the principle of the media information recommendation method provided in an embodiment of the present application, obtaining the media information F1 to be recommended and the media information F2 to be recommended, performing feature extraction on the media information F1 to be recommended, and obtaining the media information feature F11 corresponding to the media information F1 to be recommended, and performing feature extraction on the media information F2 to be recommended, and obtaining the media information feature F21 corresponding to the media information F2 to be recommended.

[0083] As an example, see Figure 4 , obtain the object information F3 of the object to be recommended, extract features of the object information F3, and obtain object features F31 of the object to be recommended F3.

[0084] As an example, obtain media information feature A (video information feature) corresponding to media information A to be recommended (video A to be recommended), media information feature B (video information feature) corresponding to media information B to be recommended (video B to be recommended), media information feature C (video information feature) corresponding to media information C to be recommended (video C to be recommended), and media information feature D (video information feature) corresponding to media information D (video D to be recommended), wherein media information feature A is used to indicate the author of video A to be recommended, the video identifier of video A to be recommended, etc., media information feature B is used to indicate the author of video B to be recommended, the video identifier of video B to be recommended, etc., media information feature C is used to indicate the author of video C to be recommended, the video identifier of video C to be recommended, etc., and media information feature D is used to indicate the author of video D to be recommended, the video identifier of video D to be recommended, etc.

[0085] As an example, the object characteristics of the object to be recommended are used to indicate the age, gender, whether the current time to be recommended is a weekend, etc. of the object to be recommended.

[0086] In this way, by accurately obtaining the media information features corresponding to the multiple media information to be recommended and the object features of the object to be recommended, it is convenient to subsequently accurately determine the accurate target media information from the multiple media information to be recommended, so as to recommend the target media information to the object to be recommended, and provide accurate data support for the subsequent recommendation of media information, thereby effectively improving the recommendation accuracy of media information.

[0087] In step 102, the media information features corresponding to each piece of media information to be recommended are cross-referenced with the object features to obtain cross-referenced features corresponding to each piece of media information to be recommended.

[0088] In some embodiments, the cross-features, the media information to be recommended and the media information features correspond to each other one by one, one cross-feature corresponds to one media information to be recommended, and one media information to be recommended corresponds to one media information feature.

[0089] As an example, see Figure 4, the media information features (media information feature F11 and media information feature F21) corresponding to each media information to be recommended (media information F1 to be recommended and media information F2 to be recommended) are respectively cross-featured with the object feature F31 to obtain the cross-features corresponding to the media information to be recommended, wherein the cross-features corresponding to the media information to be recommended F1 are [S1, X2], and the cross-features corresponding to the media information to be recommended F2 are [S2, X3].

[0090] As an example, see Figure 4 , the media information features (media information feature F11 and media information feature F21) corresponding to each media information to be recommended (media information F1 to be recommended and media information F2 to be recommended) are respectively cross-featured with the object feature F31 to obtain the cross-features corresponding to the media information to be recommended, wherein the cross-features corresponding to the media information to be recommended F1 are [S1, X1, X2], and the cross-features corresponding to the media information to be recommended F2 are [S2, X1, X3].

[0091] As an example, the media information to be recommended includes media information A to be recommended, media information B to be recommended, media information C to be recommended, media information D to be recommended, and media information E to be recommended. The media information features corresponding to the media information A to be recommended are cross-featured with the object features to obtain the cross-features corresponding to the media information A to be recommended. The media information features corresponding to the media information B to be recommended are cross-featured with the object features to obtain the cross-features corresponding to the media information B to be recommended. The media information features corresponding to the media information C to be recommended are cross-featured with the object features to obtain the cross-features corresponding to the media information C to be recommended. The media information features corresponding to the media information D to be recommended are cross-featured with the object features to obtain the cross-features corresponding to the media information D to be recommended.

[0092] In some embodiments, the intersection feature includes at least two feature elements, and the feature intersection includes an independent feature intersection and a mixed feature intersection.

[0093] In some embodiments, see Figure 5 , Figure 5 is a flowchart of a method for recommending media information provided in an embodiment of the present application. Figure 3 Step 102 shown can be performed for each media information feature corresponding to the media information to be recommended. Figure 5 Steps 1021A to 1023A are implemented as shown.

[0094] In step 1021A, independent feature crossover is performed on the media information features and the object features to obtain a first crossover feature value and a second crossover feature value.

[0095] In some embodiments, the above-mentioned independent feature intersection refers to the process of intersecting each feature element within the feature.

[0096] In some embodiments, the above-mentioned independent feature crossover of the media information features and the object features to obtain the first crossover feature value and the second crossover feature value can be achieved as follows: summing the element values ​​of each feature element in the media information features to obtain the first crossover feature value; summing the element values ​​of each feature element in the object features to obtain the second crossover feature value.

[0097] As an example, see Figure 4 , sum the element values ​​of each feature element in the media information feature F1 to obtain the first cross feature value S1 corresponding to the media information feature F1. Sum the element values ​​of each feature element in the media information feature F2 to obtain the first cross feature value S2 corresponding to the media information feature F2. Sum the element values ​​of each feature element in the object feature F3 to obtain the second cross feature value S3.

[0098] As an example, the expression of the first cross eigenvalue may be:

[0099]

[0100] Among them, f i Used to indicate the first cross eigenvalue, x i It is used to indicate the element value of each feature element in the media information feature, and d is used to indicate the number of feature elements in the media information feature.

[0101] As an example, the expression of the second cross eigenvalue may be:

[0102]

[0103] Among them, r i Used to indicate the second cross eigenvalue, w i It is used to indicate the element value of each feature element in the media information feature, and q is used to indicate the number of feature elements in the media information feature.

[0104] In step 1022A, a mixed feature crossover is performed on the media information feature and the object feature to obtain a third crossover feature value.

[0105] As an example, see Figure 4 , perform mixed feature crossover on the media information feature F11 and the object feature F31 to obtain a third cross feature value X2. Perform mixed feature crossover on the media information feature F21 and the object feature F31 to obtain a third cross feature value X3.

[0106] In some embodiments, the above step 1022 can be implemented as follows: obtain the standard feature dimension, and standardize the media information feature and the object feature according to the standard feature dimension, so as to obtain the target media information feature and the target object feature of the standard feature dimension; determine the inner product of the target media information feature and the target object feature as the third cross feature.

[0107] In some embodiments, the feature dimension of the target media information feature is equal to the standard feature dimension, and the feature dimension of the target object feature is equal to the standard feature dimension.

[0108] In some embodiments, the above-mentioned dimension standardization of media information features and object features, respectively, to obtain target media information features and target object features of standard feature dimensions can be achieved in the following manner: obtaining the feature dimension of the media information feature, and when the feature dimension of the media information feature is not equal to the standard feature dimension, adjusting the feature dimension of the media information feature to obtain the target media information feature of the standard feature dimension; obtaining the feature dimension of the object feature, and when the feature dimension of the object feature is not equal to the standard feature dimension, adjusting the feature dimension of the object feature to obtain the target object feature of the standard feature dimension.

[0109] In some embodiments, when the feature dimension of the media information feature is equal to the standard feature dimension, the media information feature is determined as a target media information feature of the standard feature dimension.

[0110] In some embodiments, when the feature dimension of the object feature is equal to the standard feature dimension, the object feature is determined as a target object feature of the standard feature dimension.

[0111] As an example, the expression of the third cross feature above can be:

[0112] X1=X2X3 (3)

[0113] Among them, X1 is used to indicate the third cross feature, X2 is used to indicate the target media information feature, and X3 is used to indicate the target object feature.

[0114] In step 1023A, the first cross feature value, the second cross feature value, and the third cross feature value are concatenated to obtain a cross feature corresponding to the media information to be recommended.

[0115] As an example, the expression of the cross-feature corresponding to the media information to be recommended may be:

[0116] Z=[Z1,Z2,Z3] (4)

[0117] Among them, Z is used to indicate the cross-feature corresponding to the media information to be recommended, Z1 is used to indicate the first cross-feature, Z2 is used to indicate the second cross-feature, and Z3 is used to indicate the third cross-feature.

[0118] In this way, by cross-feature-crossing the media information features corresponding to each media information to be recommended with the object features respectively, the cross-features corresponding to each media information to be recommended are obtained, thereby integrating the object features into each media information to be recommended, and obtaining the cross-features corresponding to each media information to be recommended. The cross-features can comprehensively reflect the interactive characteristics of the media information to be recommended and the object features, and media information recommendation is performed based on the cross-features, so that the recommended target media information can accurately reflect the preference degree of the media information to be recommended, thereby effectively improving the recommendation accuracy of the media information.

[0119] In some embodiments, see Figure 6 , Figure 6 is a flowchart of a method for recommending media information provided in an embodiment of the present application. Figure 3 Step 102 shown can be performed for each media information feature corresponding to the media information to be recommended. Figure 6 Steps 1021B to 1024B are implemented as shown.

[0120] In step 1021B, independent feature crossover is performed on the media information features and the object features to obtain a first crossover feature value and a second crossover feature value.

[0121] In some embodiments, the above-mentioned independent feature intersection refers to the process of intersecting each feature element within the feature.

[0122] In some embodiments, the above-mentioned independent feature crossover of the media information features and the object features to obtain the first crossover feature value and the second crossover feature value can be achieved as follows: summing the element values ​​of each feature element in the media information features to obtain the first crossover feature value; summing the element values ​​of each feature element in the object features to obtain the second crossover feature value.

[0123] As an example, see Figure 4 , sum the element values ​​of each feature element in the media information feature F1 to obtain the first cross feature value S1 corresponding to the media information feature F1. Sum the element values ​​of each feature element in the media information feature F2 to obtain the first cross feature value S2 corresponding to the media information feature F2. Sum the element values ​​of each feature element in the object feature F3 to obtain the second cross feature value S3.

[0124] As an example, the expression of the first cross eigenvalue may be:

[0125]

[0126] Among them, f i Used to indicate the first cross eigenvalue, x i It is used to indicate the element value of each feature element in the media information feature, and d is used to indicate the number of feature elements in the media information feature.

[0127] As an example, the expression of the second cross eigenvalue may be:

[0128]

[0129] Among them, r i Used to indicate the second cross eigenvalue, w i It is used to indicate the element value of each feature element in the media information feature, and q is used to indicate the number of feature elements in the media information feature.

[0130] In step 1022B, a mixed feature crossover is performed on the media information feature and the object feature to obtain a third crossover feature value.

[0131] As an example, see Figure 4 , perform mixed feature crossover on the media information feature F11 and the object feature F31 to obtain a third cross feature value X2. Perform mixed feature crossover on the media information feature F21 and the object feature F31 to obtain a third cross feature value X3.

[0132] In some embodiments, the above step 1022A can be implemented as follows: obtain the standard feature dimension, and standardize the media information features and object features according to the standard feature dimension, so as to obtain the target media information features and target object features of the standard feature dimension; determine the inner product of the target media information features and the target object features as the third cross feature.

[0133] In some embodiments, the feature dimension of the target media information feature is equal to the standard feature dimension, and the feature dimension of the target object feature is equal to the standard feature dimension.

[0134] In some embodiments, the above-mentioned dimension standardization of media information features and object features, respectively, to obtain target media information features and target object features of standard feature dimensions can be achieved in the following manner: obtaining the feature dimension of the media information feature, and when the feature dimension of the media information feature is not equal to the standard feature dimension, adjusting the feature dimension of the media information feature to obtain the target media information feature of the standard feature dimension; obtaining the feature dimension of the object feature, and when the feature dimension of the object feature is not equal to the standard feature dimension, adjusting the feature dimension of the object feature to obtain the target object feature of the standard feature dimension.

[0135] In some embodiments, when the feature dimension of the media information feature is equal to the standard feature dimension, the media information feature is determined as a target media information feature of the standard feature dimension.

[0136] In some embodiments, when the feature dimension of the object feature is equal to the standard feature dimension, the object feature is determined as a target object feature of the standard feature dimension.

[0137] As an example, the expression of the third cross feature above can be:

[0138] X1=X2X3 (7)

[0139] Among them, X1 is used to indicate the third cross feature, X2 is used to indicate the target media information feature, and X3 is used to indicate the target object feature.

[0140] In step 1023B, the media information features are mixed with other media information features to cross-mix and obtain a fourth feature cross-mix value.

[0141] As an example, see Figure 4 , perform mixed feature intersection on the media information feature F21 and the object feature F31 to obtain a fourth feature intersection value X2.

[0142] In some embodiments, the above step 1023B can be implemented as follows: obtain the standard feature dimension, and according to the standard feature dimension, dimensionally standardize the media information feature and other media information features respectively to obtain the target media information feature and target other media information feature of the standard feature dimension; determine the inner product of the target media information feature and the target other media information feature as the fourth cross feature.

[0143] In some embodiments, the feature dimension of the target media information feature is equal to the standard feature dimension, and the feature dimension of the target other media information feature is equal to the standard feature dimension.

[0144] As an example, the expression of the fourth cross characteristic may be:

[0145] X4=X5X6 (8)

[0146] Among them, X4 is used to indicate the fourth cross feature, X5 is used to indicate the target media information feature, and X6 is used to indicate the target other media information feature.

[0147] In step 1024B, the first cross feature value, the second cross feature value, the third cross feature value, and the fourth cross feature value are concatenated to obtain cross features corresponding to the media information to be recommended.

[0148] As an example, the expression of the cross-feature corresponding to the media information to be recommended may be:

[0149] Z=[Z1,Z2,Z3,Z4] (9)

[0150] Among them, Z is used to indicate the cross-feature corresponding to the media information to be recommended, Z1 is used to indicate the first cross-feature, Z2 is used to indicate the second cross-feature, Z3 is used to indicate the third cross-feature, and Z4 is used to indicate the fourth cross-feature.

[0151] In step 103, for each cross-feature corresponding to each media information to be recommended, importance evaluation is performed on each feature element in the cross-feature to obtain feature importance of each feature element.

[0152] In some embodiments, the feature importance is used to indicate the importance of the corresponding feature element in the cross-feature. The importance can be measured by the weight of the feature element in the cross-feature.

[0153] In some embodiments, the above-mentioned importance evaluation includes compression processing and stretching processing. The above-mentioned compression processing is used to compress the feature dimension of the feature, which is a feature transformation process for compressing the feature dimension. The above-mentioned stretching processing is used to stretch the feature dimension of the feature, which is a feature transformation process for stretching the feature dimension.

[0154] In some embodiments, see Figure 7 , Figure 7 is a flowchart of a method for recommending media information provided in an embodiment of the present application. Figure 3 Step 103 shown may be performed by executing Figure 7 Steps 1031 to 1033 are shown to be implemented.

[0155] In step 1031, the cross features are compressed to obtain compressed cross features.

[0156] In some embodiments, the number of feature elements in the compressed intersection feature is less than the number of feature elements in the intersection feature.

[0157] In some embodiments, the above step 1031 can be implemented as follows: obtain a compressed feature for reducing the dimension of the cross feature, and multiply the cross feature and the compressed feature to obtain a reference compressed feature, where the number of feature elements in the reference compressed feature is less than the number of feature elements in the cross feature; perform feature transformation on the reference compressed feature to obtain a compressed cross feature.

[0158] As an example, the expression of the above compressed cross feature can be:

[0159] R sq =f sq (Z) = a1(Wsq ·Z T ) (10)

[0160] Among them, R sq Used to indicate compression crossover features, Z is used to indicate crossover features, Z T Used to indicate the transpose of crossed features, W sq It is used to indicate the compression features used to reduce the dimension of the cross features, a1 is used to indicate the feature transformation function, f sq () is used to indicate the compression function.

[0161] In step 1032, the compressed cross features are stretched to obtain importance features.

[0162] In some embodiments, the number of feature elements in the importance feature is equal to the number of feature elements in the cross feature, and the feature elements in the importance feature correspond one-to-one to the feature elements in the cross feature.

[0163] In some embodiments, the above step 1032 can be implemented as follows: obtain a stretched feature for upgrading the dimension of the compressed cross feature, and multiply the stretched feature and the compressed cross feature to obtain a reference importance feature, wherein the number of feature elements in the reference importance feature is equal to the number of feature elements in the cross feature; perform feature transformation on the reference importance feature to obtain an importance feature.

[0164] As an example, the expression of the above importance feature can be:

[0165] R ex =f ex (R sq )=a2(W ex ·[R sq ] T ) (11)

[0166] Among them, R ex Used to indicate the importance of features, R sq Used to indicate the compression crossover feature, f ex () is used to indicate the stretching function, W ex Used to indicate the stretched features used to increase the dimension of the compressed cross features, [R sq ] T It is used to indicate the transpose of the compressed cross feature, and a2 is used to indicate the feature transformation function.

[0167] In step 1033, for each feature element in the importance feature, the element value of the feature element in the importance feature is determined as the feature importance of the corresponding feature element in the cross feature.

[0168] As an example, the expression of the above importance feature can also be:

[0169] R ex =[R1, R2, R3, R4] (12)

[0170] Among them, R ex Used to indicate importance features, R1 is used to indicate the first feature element in the importance feature, R2 is used to indicate the second feature element in the importance feature, R3 is used to indicate the third feature element in the importance feature, and R4 is used to indicate the fourth feature element in the importance feature.

[0171] Correspondingly, the expression of the above cross-feature can be:

[0172] Z=[Z1,Z2,Z3,Z4] (13)

[0173] Among them, Z is used to indicate the cross feature, Z1 is used to indicate the first feature element in the cross feature, Z2 is used to indicate the second feature element in the cross feature, Z3 is used to indicate the third feature element in the cross feature, and Z4 is used to indicate the fourth feature element in the cross feature.

[0174] Continuing with the above example, the element value of the first feature element R1 in the importance feature is determined as the feature importance of the first feature element Z1 in the cross-feature, the element value of the first feature element R2 in the importance feature is determined as the feature importance of the first feature element Z2 in the cross-feature, the element value of the third feature element R3 in the importance feature is determined as the feature importance of the third feature element Z3 in the cross-feature, and the element value of the fourth feature element R4 in the importance feature is determined as the feature importance of the fourth feature element Z4 in the cross-feature.

[0175] In step 104, the first recommendation parameter value of the media information to be recommended is determined in combination with the feature importance of each feature element and the element value of each feature element.

[0176] In some embodiments, the above step 104 can be implemented as follows: for each cross-feature corresponding to the media information to be recommended, weighted sum of the element values ​​of each feature element in the cross-feature is performed according to the feature importance to obtain the first recommended parameter value of the corresponding media information feature.

[0177] Following the above example, the expression of the first recommended parameter value of the above media information to be recommended can be:

[0178] Q=R1Z1+R2Z2+R3Z3+R4Z4 (14)

[0179] Among them, Q is used to indicate the first recommended parameter value of the media information to be recommended, R1 is used to indicate the first characteristic element in the importance feature, Z1 is used to indicate the first characteristic element in the cross-feature, R2 is used to indicate the first characteristic element in the importance feature, Z2 is used to indicate the first characteristic element in the cross-feature, R3 is used to indicate the third characteristic element in the importance feature, Z3 is used to indicate the third characteristic element in the cross-feature, R4 is used to indicate the fourth characteristic element in the importance feature, and Z4 is used to indicate the fourth characteristic element in the cross-feature.

[0180] In step 105, target media information is selected from a plurality of media information to be recommended based on the first recommendation parameter value, and the target media information is recommended to the object to be recommended.

[0181] In some embodiments, the first recommendation parameter value is used to indicate the recommendation level of the corresponding media information to be recommended, or the first recommendation parameter value is used to indicate the attention level of the recommended object to the corresponding media information to be recommended.

[0182] In some embodiments, the above-mentioned step 105 of selecting target media information from multiple media information to be recommended based on the first recommended parameter value can be achieved in the following way: determining the opposite of the first recommended parameter value as the first target parameter value, normalizing the first target parameter value to obtain the first recommended value of the media information to be recommended, and selecting the media information to be recommended with the largest first recommended value from the multiple media information to be recommended as the target media information.

[0183] As an example, the expression of the first recommended value may be:

[0184]

[0185] Among them, s FM It is used to indicate the first recommended parameter value, sigmoid() is used to indicate the normalization processing function, and p1 is used to indicate the first recommended value.

[0186] In some embodiments, see Figure 8 , Figure 8 is a flow chart of a method for recommending media information provided in an embodiment of the present application. Figure 3 Prior to step 105, the Figure 8 Illustrated step 106 determines a second recommended parameter value.

[0187] In some embodiments, the second recommendation parameter value is used to indicate the recommendation level of the corresponding media information to be recommended, or the second recommendation parameter value is used to indicate the attention level of the recommended object to the corresponding media information to be recommended.

[0188] In step 106, based on the object features and the features of each piece of media information, a recommendation parameter value is predicted for each piece of media information to be recommended, to obtain a second recommendation parameter value for each piece of media information to be recommended.

[0189] In some embodiments, see Fig. 9 , Fig. 9 is a flowchart of a method for recommending media information provided in an embodiment of the present application. Figure 8 Step 106 shown can be performed by performing the following steps for each media information feature: Fig. 9 Steps 1061 to 1063 are implemented as shown.

[0190] In step 1061, the object feature and the media information feature are spliced ​​to obtain a spliced ​​feature corresponding to the media information feature.

[0191] As an example, see Figure 4 , the object feature F31, the media information feature F11 and the media information feature F21 are feature spliced ​​to obtain a spliced ​​feature F8 corresponding to the media information feature.

[0192] In step 1062, a recommended parameter value prediction network is obtained.

[0193] In some embodiments, the above step 1061 can be implemented in the following manner: obtaining object feature samples and media information feature samples, and performing feature splicing on the object feature samples and the media information feature samples to obtain spliced ​​feature samples; obtaining an initial recommended parameter value prediction network and a label recommended parameter value, and calling the initial recommended parameter value prediction network, predicting the recommended parameter value for the media information feature sample based on the spliced ​​feature sample, to obtain the predicted recommended parameter value for the media information feature sample; training the initial recommended parameter value prediction network in combination with the label recommended parameter value and the predicted recommended parameter value to obtain the recommended parameter value prediction network.

[0194] In some embodiments, the above-mentioned combination of label recommended parameter values ​​and predicted recommended parameter values ​​to train the initial recommended parameter value prediction network to obtain the recommended parameter value prediction network can be achieved in the following way: the difference between the label recommended parameter value and the predicted recommended parameter value is determined as the first loss value of the initial recommended parameter value prediction network, and based on the first loss value, the initial recommended parameter value prediction network is trained to obtain the recommended parameter value prediction network.

[0195] As an example, the expression of the first loss value may be:

[0196] L=T1-T2 (16)

[0197] Among them, L is used to indicate the first loss value, T1 is used to indicate the label recommendation parameter value, and T2 is used to indicate the predicted recommendation parameter value.

[0198] In some other embodiments, the above-mentioned step 1061 can also be implemented in the following manner: obtaining object feature samples and media information feature samples, and performing feature splicing on the object feature samples and the media information feature samples to obtain spliced ​​feature samples; obtaining an initial recommended parameter value prediction network, and calling the initial recommended parameter value prediction network, based on the spliced ​​feature samples, predicting recommended parameter values ​​for the media information feature samples to obtain predicted recommended parameter values ​​for the media information feature samples, based on the predicted recommended parameter values, determining a target predicted recommended value for the media information feature samples, based on the target predicted recommended value, determining a second loss value of the initial recommended parameter value prediction network, and based on the second loss value, training the initial recommended parameter value prediction network to obtain a recommended parameter value prediction network.

[0199] In some embodiments, the above-mentioned determination of the target predicted recommendation value of the media information feature sample based on the predicted recommendation parameter value can be achieved in the following way: the opposite number of the predicted recommendation parameter value is determined as the first predicted parameter value, and the first predicted parameter value is normalized to obtain the target predicted recommendation value of the media information feature sample.

[0200] As an example, the expression of the above target prediction recommendation value can be:

[0201]

[0202] Among them, p11 is used to indicate the target prediction recommendation value, s F It is used to indicate the predicted recommended parameter value, and sigmoid() is used to indicate the normalization processing function.

[0203] In some embodiments, the above-mentioned determination of the second loss value of the initial recommended parameter value prediction network based on the target predicted recommendation value can be achieved by obtaining the sample label, combining the sample label and the target predicted recommendation value, and determining the second loss value of the initial recommended parameter value prediction network.

[0204] As an example, the expression of the second loss value may be:

[0205]

[0206] in, It is used to indicate the second loss value, y is used to indicate the sample label, the value of the sample label is 0 or 1, representing negative samples and positive samples respectively, and p11 is used to indicate the target prediction recommendation value.

[0207] In step 1063, the recommended parameter value prediction network is called to predict the recommended parameter value of the media information feature based on the splicing feature to obtain a second recommended parameter value of the media information feature.

[0208] As an example, see Figure 4 , calling the recommended parameter value prediction network, based on the splicing feature F8, predicting the recommended parameter value of the media information feature, and obtaining the second recommended parameter value of the media information feature.

[0209] In this way, by calling the recommendation parameter value prediction network based on the object characteristics and the characteristics of each media information, the recommendation parameter value of each media information to be recommended is predicted, and the second recommendation parameter value of each media information to be recommended is obtained. The second recommendation parameter value is used to form an effective supplement to the first recommendation parameter value. By combining the first recommendation parameter value and the second recommendation parameter value, the recommended object is recommended, so that the recommended target media information can accurately reflect the preference degree of the media information to be recommended, thereby effectively improving the recommendation accuracy of the media information.

[0210] In some embodiments, see Fig.10 , Fig.10 is a flowchart of a method for recommending media information provided in an embodiment of the present application. Figure 3 In step 105, based on the first recommendation parameter value, target media information is selected from a plurality of media information to be recommended, which can be performed by executing Fig.10 Steps 1051 to 1052 are implemented as shown.

[0211] In step 1051, for each piece of media information to be recommended, a recommendation value of the media information to be recommended is determined in combination with the corresponding first recommendation parameter value and the second recommendation parameter value.

[0212] In some embodiments, the above step 1051 can be implemented as follows: summing the first recommended parameter value and the second recommended parameter value corresponding to the media information to be recommended to obtain a summed parameter value, and determining the opposite of the summed parameter value as the target parameter value; normalizing the target parameter value to obtain a recommended value for the media information to be recommended.

[0213] As an example, the expression of the recommendation value of the above media information to be recommended may be:

[0214]

[0215] Among them, p is used to indicate the recommended value of the media information to be recommended, s FM Used to indicate the first recommended parameter value, s DNN Used to indicate the second recommended parameter value, -(s FM +s DNN ) is used to indicate the target parameter value, and sigmoid() is used to indicate the normalization processing function.

[0216] In some embodiments, the above-mentioned normalization of the target parameter value to obtain the recommended value of the media information to be recommended can be achieved in the following way: the natural exponential power of the target parameter value is determined as the reference parameter value, the reference parameter value is added by 1 to obtain the reference recommended value, and the reciprocal of the reference recommended value is determined as the recommended value of the media information to be recommended.

[0217] In step 1052, the to-be-recommended media information having a recommendation value greater than a recommendation threshold is determined as target media information.

[0218] In other embodiments, Figure 3 The step 105 shown in which the target media information is selected from multiple media information to be recommended based on the first recommended parameter value can also be implemented in the following way: for each piece of media information to be recommended, the recommended value of the media information to be recommended is determined in combination with the corresponding first recommended parameter value and second recommended parameter value, and the media information to be recommended with the largest recommended value is determined as the target media information.

[0219] In some embodiments, the media information recommendation method provided in the embodiments of the present application can also be implemented in the following manner: obtain a media information feature corresponding to the media information to be recommended and an object feature of the object to be recommended; perform feature cross-feature on the media information feature corresponding to the media information to be recommended and the object feature to obtain the cross-feature corresponding to the media information to be recommended, perform importance evaluation on each feature element in the cross-feature to obtain the feature importance of each feature element; determine a first recommendation parameter value for the media information to be recommended based on the feature importance of each feature element and the element value of each feature element, determine the recommendation probability of the media information to be recommended based on the first recommendation parameter value, recommend the media information to be recommended to the object to be recommended when the recommendation probability is greater than the recommendation probability threshold, and do not recommend the media information to be recommended to the object to be recommended when the recommendation probability is less than or equal to the recommendation probability threshold.

[0220] In this way, by obtaining media information features corresponding to multiple media information to be recommended and object features of the object to be recommended, the media information features corresponding to each media information to be recommended are cross-featured with the object features to obtain cross-features corresponding to each media information to be recommended. For each cross-feature, the importance of each feature element in the cross-feature is evaluated to obtain the feature importance of the feature element. The first recommendation parameter value of the media information to be recommended is determined in combination with the feature importance of the feature element and the element value of each feature element. Based on the first recommendation parameter value, target media information is selected from multiple media information to be recommended, and the target media information is recommended to the object to be recommended. In this way, by cross-feature-crossing the media information features corresponding to each media information to be recommended with the object features respectively, the cross-features corresponding to each media information to be recommended are obtained, thereby integrating the object features into each media information to be recommended, and obtaining the cross-features corresponding to each media information to be recommended, so that the cross-features can comprehensively reflect the interactive characteristics of the media information to be recommended and the object features, and by evaluating the importance of each feature element in the cross-features, the target media information recommended to the object to be recommended is determined based on the feature importance of the feature elements obtained, and because the feature importance comprehensively reflects the importance of the feature elements in the cross-features, the determined first recommended parameter value can fully consider the influence of each feature element in the cross-features on the first recommended parameter value, so that the determined first recommended parameter value is more accurate, and the selected target media information is more accurate, which effectively improves the recommendation accuracy of the media information.

[0221] The following describes an exemplary application of the embodiment of the present application in an actual audio and video recommendation application scenario.

[0222] Factorization Machines (FM) is a method of modeling feature intersections using vector inner products. Since FM can directly perform inner product operations based on the vector representation of two features to obtain the result of feature intersection, it does not depend on whether a certain feature combination exists in the sample during training. It has good generalization ability in the recommendation scenario of large-scale sparse features. In the recommendation system based on deep learning, although DNN can capture nonlinear high-order implicit intersections, the additional introduction of FM to explicitly model low-order (second-order) feature intersections can still bring significant effect improvement (DeepFM). Although FM can effectively model feature intersections based on vector inner products, there is a lack of effective trade-offs for feature intersections of different combinations. Obviously, not all feature intersections have a significant positive impact on the model's prediction, so it is necessary to model the importance of feature intersections. Through the recommendation method for media information provided by the embodiment of the present application, when evaluating the importance of feature intersections of a specific combination, the contextual information of global feature intersections can be considered at the same time, so that the importance of different feature combinations is associated to a certain extent, and then the effective feature combinations can be more accurately strengthened, while reducing the interference of invalid feature combinations, thereby improving the prediction effect of the model.

[0223] In some embodiments, see Fig.11 , Fig.11It is a schematic diagram of the interface of the method for recommending media information provided by the embodiment of the present application. The embodiment of the present application is suitable for business scenarios such as short videos and advertisements that use deep learning systems in large quantities, such as video push. Users will receive target messages from time to time in the interface. Multiple target message types have different external forms. After the user clicks on the message, he will enter the interface and play the video sent with the target message. Assuming that the probability of clicking on the target message is now considered, the modeling of feature crossover is crucial to the prediction effect. Through data analysis, it can be observed that the student group is usually more active on weekends (age, weekends, second-order crossover), and the activity of older users in first-tier cities on weekends is significantly lower than that in the middle of the week (city, age, weekends, third-order crossover). Usually, this feature crossover is very complex, and low-order and high-order feature crossovers have a significant effect on the model prediction effect. If the value of a specific feature combination crossover is very large, but the values ​​of all remaining feature combinations are very small, then it is very likely that the value of the crossover result of this specific feature combination is a noise disturbance that deviates from the normal distribution, and its importance should not be too high (the weight of AFM is easily polarized to this noise crossover). When evaluating feature cross-weights, the result of feature cross-weights itself is also an important reference (FwFM lacks this modeling capability). For example, some features may be randomly missing. For features with two missing values, their weights should not be too high; but when both features are filled with values ​​normally, their weights may not be too small. DeepFM4 is a deep learning recommendation model that uses FM to automatically model low-order (second-order) feature cross-weights, and then uses DNN to model high-order implicit cross-weights. It can consider low-order and high-order cross-weights end-to-end at the same time. Based on DeepFM, the present invention combines SENet5 with FM cross-weights to perform adaptive importance evaluation. When evaluating the weight of a single feature combination, the cross-weight results of all feature combinations in the sample are fully considered, and the global FM cross-weight result context information is perceived (for example, the first-order and second-order results are considered at the same time). It can greatly reduce the interference of noise cross-weights, establish the association between cross-weights of different feature combinations, and more accurately model the importance of feature combinations in specific samples.

[0224] In some embodiments, see Fig.12 , Fig.12 It is a schematic diagram of the principle of the media information recommendation method provided in the embodiment of the present application. Based on DeepFM, the SENet structure is added on the FM side, the cross-results of FM are used as the summary information input of SENet, and the weights of SENet output are used as the weights of each feature combination of FM.

[0225] In some embodiments, the following describes from bottom to top Fig.12In the input processing flow, each original input feature (gender, age, whether it is the weekend, the ID of the video to be estimated, etc., corresponding to F1 / F2 / F3 in the diagram) is mapped to a vector representation through the parameter table corresponding to the Embedding layer; for variable-length sequence features (for example, the Item ID sequence played by the user for >30 seconds), each element in the sequence is first mapped to a vector representation, and then the variable-length vector sequence is subjected to a Pooling operation (summing, averaging, or using a sequence model to process, etc.) to obtain a fixed-length representation of the sequence. Assume that we have N fixed-length feature vectors, and the dimensions are all set to d. The nth feature is recorded as x. n For the FM side, consider the intersection of first-order and second-order features, where the value of the first-order feature is taken as the sum of the elements of the feature vector. For the i-th feature, the first-order value of its FM is:

[0226]

[0227] N features have a total of N first-order values.

[0228] For the second-order feature intersection, take the inner product of the two feature vectors. For the i-th feature and the j-th feature, the value of the second-order feature intersection of the FM is:

[0229]

[0230] There are N features in total The second-order feature intersection value.

[0231] Concatenate the first-order and second-order values ​​of FM to get a summary vector Z, namely:

[0232]

[0233] Its dimension is Use this vector as the summary vector of SENet and perform compression and stretching:

[0234] Squeeze: Compression dimensionality reduction operation:

[0235] R sq =f sq (Z) = a1(W sq ·Z T ) (twenty three)

[0236] Among them, R sq is the compressed intermediate result, with dimension Among them, r represents the compression multiple; the parameter matrix W sq The dimension is a1 is a nonlinear activation function.

[0237] Excitation: The excitation operation is to perform a dimensionality increase operation on the intermediate result of compression and dimensionality reduction, namely:

[0238] R ex =f ex (R sq )=a2(W ex ·[R sq ] T ) (twenty four)

[0239] Among them, R ex is the stretched d Z dimensional weight vector for d Z The first-order and second-order values ​​of FM are weighted, and the parameter matrix W ex The dimension is a2 is the tanh function, which directly obtains the feature cross weight between -1 and 1.

[0240] The final output value of the FM side is:

[0241]

[0242] For the DNN side that models high-order implicit interactions, we concatenate the N feature vectors to obtain the final sample input vector, i.e.

[0243] X=[…,x n ,…] (26)

[0244] Among them, the dimension of X is d X =Nd, the output value of the high-order implicit cross on the DNN side (without activation function) is:

[0245] s DNN =h(X) (27)

[0246] Among them, h(·) represents the processing process of the DNN network, and the final estimated probability value of the model is:

[0247]

[0248] Let the label of the sample be y (with a value of 0 or 1, representing a negative sample and a positive sample), and the loss function of its training is:

[0249]

[0250] In this way, by obtaining media information features corresponding to multiple media information to be recommended and object features of the object to be recommended, the media information features corresponding to each media information to be recommended are cross-featured with the object features to obtain cross-features corresponding to each media information to be recommended. For each cross-feature, the importance of each feature element in the cross-feature is evaluated to obtain the feature importance of the feature element. The first recommendation parameter value of the media information to be recommended is determined in combination with the feature importance of the feature element and the element value of each feature element. Based on the first recommendation parameter value, target media information is selected from multiple media information to be recommended, and the target media information is recommended to the object to be recommended. In this way, by cross-feature-crossing the media information features corresponding to each media information to be recommended with the object features respectively, the cross-features corresponding to each media information to be recommended are obtained, thereby integrating the object features into each media information to be recommended, and obtaining the cross-features corresponding to each media information to be recommended, so that the cross-features can comprehensively reflect the interactive characteristics of the media information to be recommended and the object features, and by evaluating the importance of each feature element in the cross-features, the target media information recommended to the object to be recommended is determined based on the feature importance of the feature elements obtained, and because the feature importance comprehensively reflects the importance of the feature elements in the cross-features, the determined first recommended parameter value can fully consider the influence of each feature element in the cross-features on the first recommended parameter value, so that the determined first recommended parameter value is more accurate, and the selected target media information is more accurate, which effectively improves the recommendation accuracy of the media information.

[0251] It is understandable that in the embodiments of the present application, when it involves data related to media information to be recommended, when the embodiments of the present application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0252] The following is a description of an exemplary structure of the media information recommendation device 455 provided in the embodiment of the present application implemented as a software module. In some embodiments, for example, Figure 2As shown, the software modules in the media information recommendation device 455 stored in the memory 450 may include: an acquisition module, used to acquire media information features corresponding to a plurality of media information to be recommended, and object features of an object to be recommended; a feature intersection module, used to perform feature intersection of the media information features corresponding to each of the media information to be recommended with the object features, to obtain cross-features corresponding to each of the media information to be recommended, wherein the cross-features include at least two feature elements; an evaluation module, used to perform importance evaluation on each feature element in the cross-features corresponding to each of the media information to be recommended, to obtain the feature importance of each feature element, and to determine a first recommendation parameter value for the media information to be recommended in combination with the feature importance of each feature element and the element value of each feature element; a recommendation module, used to select target media information from the plurality of media information to be recommended based on the first recommendation parameter value, and recommend the target media information to the object to be recommended.

[0253] In some embodiments, the above-mentioned media information recommendation device also includes: a prediction module, which is used to predict the recommended parameter value of each media information to be recommended based on the object characteristics and the characteristics of each media information, and obtain the second recommended parameter value of each media information to be recommended; the above-mentioned recommendation module is also used to determine the recommended value of the media information to be recommended for each media information to be recommended, in combination with the corresponding first recommended parameter value and the second recommended parameter value; the media information to be recommended whose recommendation value is greater than the recommendation threshold is determined as the target media information.

[0254] In some embodiments, the above-mentioned prediction module is also used to perform the following processing for each of the media information features: feature splicing the object feature and the media information feature to obtain a splicing feature corresponding to the media information feature; obtaining a recommended parameter value prediction network, calling the recommended parameter value prediction network, and predicting the recommended parameter value of the media information feature based on the splicing feature to obtain a second recommended parameter value of the media information feature.

[0255] In some embodiments, the above-mentioned prediction module is also used to obtain object feature samples and media information feature samples, and perform feature splicing on the object feature samples and the media information feature samples to obtain spliced ​​feature samples; obtain an initial recommended parameter value prediction network and a label recommended parameter value, and call the initial recommended parameter value prediction network to predict the recommended parameter value of the media information feature sample based on the spliced ​​feature sample to obtain the predicted recommended parameter value of the media information feature sample; train the initial recommended parameter value prediction network in combination with the label recommended parameter value and the predicted recommended parameter value to obtain the recommended parameter value prediction network.

[0256] In some embodiments, the above-mentioned recommendation module is also used to perform the following processing for each of the media information to be recommended: summing the first recommendation parameter value and the second recommendation parameter value corresponding to the media information to be recommended to obtain a summed parameter value, and determining the opposite of the summed parameter value as the target parameter value; normalizing the target parameter value to obtain the recommended value of the media information to be recommended.

[0257] In some embodiments, the above-mentioned feature intersection includes independent feature intersection and mixed feature intersection. The above-mentioned feature intersection module is also used to perform the following processing respectively for the media information features corresponding to each of the media information to be recommended: perform independent feature intersection on the media information features and the object features respectively to obtain a first cross feature value and a second cross feature value; perform mixed feature intersection on the media information features and the object features to obtain a third cross feature value; perform feature value splicing on the first cross feature value, the second cross feature value and the third cross feature value to obtain the cross feature corresponding to the media information to be recommended.

[0258] In some embodiments, the feature cross module is further used to sum the element values ​​of each feature element in the media information feature to obtain the first cross feature value; and sum the element values ​​of each feature element in the object feature to obtain the second cross feature value.

[0259] In some embodiments, the feature intersection module is also used to obtain a standard feature dimension, and to perform dimension standardization on the media information feature and the object feature respectively according to the standard feature dimension, so as to obtain a target media information feature and a target object feature of the standard feature dimension; and to determine the inner product of the target media information feature and the target object feature as the third intersection feature.

[0260] In some embodiments, the above-mentioned feature intersection module is also used to obtain the feature dimension of the media information feature, and when the feature dimension of the media information feature is not equal to the standard feature dimension, adjust the feature dimension of the media information feature to obtain the target media information feature of the standard feature dimension; obtain the feature dimension of the object feature, and when the feature dimension of the object feature is not equal to the standard feature dimension, adjust the feature dimension of the object feature to obtain the target object feature of the standard feature dimension.

[0261] In some embodiments, the above-mentioned evaluation module is also used to compress the cross-feature to obtain a compressed cross-feature, wherein the number of feature elements in the compressed cross-feature is less than the number of feature elements in the cross-feature; stretch the compressed cross-feature to obtain an importance feature, wherein the number of feature elements in the importance feature is equal to the number of feature elements in the cross-feature, and the feature elements in the importance feature correspond one-to-one with the feature elements in the cross-feature; for each feature element in the importance feature, the element value of the feature element in the importance feature is determined as the feature importance of the corresponding feature element in the cross-feature.

[0262] In some embodiments, the above-mentioned evaluation module is also used to obtain a compressed feature for reducing the dimension of the cross feature, and multiply the cross feature and the compressed feature to obtain a reference compressed feature, wherein the number of feature elements in the reference compressed feature is less than the number of feature elements in the cross feature; and perform feature transformation on the reference compressed feature to obtain the compressed cross feature.

[0263] In some embodiments, the above-mentioned evaluation module is also used to obtain a stretched feature for upgrading the dimension of the compressed cross feature, and multiply the stretched feature and the compressed cross feature to obtain a reference importance feature, wherein the number of feature elements in the reference importance feature is equal to the number of feature elements in the cross feature; and perform feature transformation on the reference importance feature to obtain the importance feature.

[0264] The embodiment of the present application provides a computer program product, which includes a computer program or a computer executable instruction, and the computer program or the computer executable instruction is stored in a computer-readable storage medium. The processor of the electronic device reads the computer executable instruction from the computer-readable storage medium, and the processor executes the computer executable instruction, so that the electronic device executes the media information recommendation method described in the embodiment of the present application.

[0265] The present application embodiment provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the processor will be caused to execute the media information recommendation method provided by the present application embodiment, for example, Figure 3 A recommended method of displaying media information.

[0266] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEP ROM, flash memory, magnetic surface storage, optical disk, or CD-ROM; or it may be various electronic devices including one or any combination of the above memories.

[0267] In some embodiments, computer executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0268] As an example, computer-executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file storing other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions).

[0269] As an example, computer executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one site, or on multiple electronic devices distributed at multiple sites and interconnected by a communication network.

[0270] In summary, the embodiments of the present application have the following beneficial effects:

[0271] (1) By obtaining media information features corresponding to a plurality of media information to be recommended and object features of an object to be recommended, the media information features corresponding to each piece of media information to be recommended are cross-featured with the object features to obtain cross-features corresponding to each piece of media information to be recommended; for each cross-feature, importance evaluation is performed on each feature element in the cross-feature to obtain feature importance of the feature element; a first recommendation parameter value of the media information to be recommended is determined in combination with the feature importance of the feature element and the element value of each feature element; based on the first recommendation parameter value, target media information is selected from the plurality of media information to be recommended, and the target media information is recommended to the object to be recommended. In this way, by cross-feature-crossing the media information features corresponding to each media information to be recommended with the object features respectively, the cross-features corresponding to each media information to be recommended are obtained, thereby integrating the object features into each media information to be recommended, and obtaining the cross-features corresponding to each media information to be recommended, so that the cross-features can comprehensively reflect the interactive characteristics of the media information to be recommended and the object features, and by evaluating the importance of each feature element in the cross-features, the target media information recommended to the object to be recommended is determined based on the feature importance of the feature elements obtained, and because the feature importance comprehensively reflects the importance of the feature elements in the cross-features, the determined first recommended parameter value can fully consider the influence of each feature element in the cross-features on the first recommended parameter value, so that the determined first recommended parameter value is more accurate, and the selected target media information is more accurate, which effectively improves the recommendation accuracy of the media information.

[0272] (2) By accurately obtaining the media information features corresponding to the multiple media information to be recommended and the object features of the object to be recommended, it is convenient to accurately determine the accurate target media information from the multiple media information to be recommended, so as to recommend the target media information to the object to be recommended, and provide accurate data support for the subsequent recommendation of media information, thereby effectively improving the recommendation accuracy of media information.

[0273] (3) The cross-features corresponding to each media information to be recommended are obtained by respectively cross-feature-crossing the media information features corresponding to each media information to be recommended with the object features, thereby integrating the object features into each media information to be recommended, thereby obtaining the cross-features corresponding to each media information to be recommended. The cross-features can comprehensively reflect the interactive characteristics of the media information to be recommended and the object features, and media information recommendation is performed based on the cross-features, so that the recommended target media information can accurately reflect the preference of the media information to be recommended, thereby effectively improving the recommendation accuracy of the media information.

[0274] (4) Based on the object characteristics and the characteristics of each media information, the recommendation parameter value prediction network is called respectively to predict the recommendation parameter value of each media information to be recommended, and the second recommendation parameter value of each media information to be recommended is obtained. The second recommendation parameter value is used to form an effective supplement to the first recommendation parameter value. By combining the first recommendation parameter value and the second recommendation parameter value, the recommended object is recommended, so that the recommended target media information can accurately reflect the preference of the recommended object for the media information, thereby effectively improving the recommendation accuracy of the media information.

[0275] (5) Through the media information recommendation method provided in the embodiment of the present application, when evaluating the feature cross-importance of a specific combination, the contextual information of the global feature cross-importance can be considered simultaneously, so that the importance of different feature combinations is correlated to a certain extent, thereby more accurately strengthening the effective feature combinations and reducing the interference of invalid feature combinations, thereby improving the prediction effect of the model.

[0276] The above is only an embodiment of the present application and is not intended to limit the protection scope of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. A method for recommending media information, characterized in that: The method comprises: Acquire media information features corresponding to the plurality of media information to be recommended, and object features of the object to be recommended; Performing feature crossover on the media information features corresponding to each of the media information to be recommended and the object features to obtain crossover features corresponding to each of the media information to be recommended, wherein the crossover features include at least two feature elements; For each cross-feature corresponding to each of the media information to be recommended, an importance evaluation is performed on each feature element in the cross-feature to obtain a feature importance of each feature element, and a first recommendation parameter value of the media information to be recommended is determined in combination with the feature importance of each feature element and an element value of each feature element; Based on the first recommendation parameter value, target media information is selected from the plurality of media information to be recommended, and the target media information is recommended to the object to be recommended.

2. The method according to claim 1, characterized in that Before selecting target media information from the plurality of media information to be recommended based on the first recommendation parameter value, the method further includes: Based on the object characteristics and the characteristics of each of the media information, respectively predicting a recommended parameter value for each of the media information to be recommended, to obtain a second recommended parameter value for each of the media information to be recommended; The selecting target media information from the plurality of media information to be recommended based on the first recommendation parameter value includes: For each piece of media information to be recommended, determining a recommendation value of the media information to be recommended in combination with the corresponding first recommendation parameter value and the second recommendation parameter value; The to-be-recommended media information whose recommendation value is greater than the recommendation threshold is determined as the target media information.

3. The method according to claim 2, characterized in that The predicting of the recommended parameter value for each of the media information to be recommended based on the object feature and each of the media information features to obtain the second recommended parameter value for each of the media information to be recommended includes: The following processing is performed for each of the media information features: Performing feature splicing on the object feature and the media information feature to obtain a splicing feature corresponding to the media information feature; A recommended parameter value prediction network is obtained, the recommended parameter value prediction network is called, and based on the splicing feature, a recommended parameter value prediction is performed on the media information feature to obtain a second recommended parameter value of the media information feature.

4. The method according to claim 3, characterized in that The obtaining of the recommended parameter value prediction network includes: Acquire an object feature sample and a media information feature sample, and perform feature splicing on the object feature sample and the media information feature sample to obtain a spliced ​​feature sample; Acquire an initial recommended parameter value prediction network and a tag recommended parameter value, and call the initial recommended parameter value prediction network to predict the recommended parameter value of the media information feature sample based on the spliced ​​feature sample to obtain a predicted recommended parameter value of the media information feature sample; The initial recommended parameter value prediction network is trained in combination with the label recommended parameter value and the predicted recommended parameter value to obtain the recommended parameter value prediction network.

5. The method according to claim 2, characterized in that: The determining, for each piece of the media information to be recommended, the recommended value of the media information to be recommended in combination with the corresponding first recommended parameter value and the second recommended parameter value, includes: The following processing is performed for each piece of media information to be recommended: Summing the first recommendation parameter value and the second recommendation parameter value corresponding to the to-be-recommended media information to obtain a summed parameter value, and determining the opposite of the summed parameter value as a target parameter value; The target parameter value is normalized to obtain a recommendation value of the media information to be recommended.

6. The method according to claim 1, characterized in that The feature crossover includes independent feature crossover and mixed feature crossover, and the media information features corresponding to each of the media information to be recommended are crossovered with the object features to obtain crossover features corresponding to each of the media information to be recommended, including: The following processing is performed respectively for each media information feature corresponding to the media information to be recommended: Performing independent feature crossover on the media information feature and the object feature respectively to obtain a first crossover feature value and a second crossover feature value; Performing mixed feature crossover on the media information feature and the object feature to obtain a third crossover feature value; The first cross feature value, the second cross feature value, and the third cross feature value are concatenated to obtain a cross feature corresponding to the media information to be recommended.

7. The method according to claim 6, characterized in that The performing independent feature crossover on the media information feature and the object feature to obtain a first crossover feature value and a second crossover feature value comprises: Summing the element values ​​of each characteristic element in the media information feature to obtain the first cross characteristic value; The element values ​​of each feature element in the object feature are summed to obtain the second cross feature value.

8. The method according to claim 6, characterized in that The performing mixed feature crossover on the media information feature and the object feature to obtain a third crossover feature value includes: Acquire a standard feature dimension, and perform dimension standardization on the media information feature and the object feature according to the standard feature dimension, respectively, to obtain a target media information feature and a target object feature of the standard feature dimension; The inner product of the target media information feature and the target object feature is determined as the third intersection feature.

9. The method according to claim 8, characterized in that The step of respectively dimensionally normalizing the media information features and the object features to obtain target media information features and target object features of the standard feature dimensions includes: Acquire a feature dimension of the media information feature, and when the feature dimension of the media information feature is not equal to the standard feature dimension, adjust the feature dimension of the media information feature to obtain a target media information feature of the standard feature dimension; The feature dimension of the object feature is obtained, and when the feature dimension of the object feature is not equal to the standard feature dimension, the feature dimension of the object feature is adjusted to obtain the target object feature of the standard feature dimension.

10. The method according to claim 1, characterized in that The step of evaluating the importance of each feature element in the cross-feature to obtain the feature importance of each feature element includes: Compressing the intersection feature to obtain a compressed intersection feature, wherein the number of feature elements in the compressed intersection feature is less than the number of feature elements in the intersection feature; Performing stretching processing on the compressed cross feature to obtain an importance feature, the number of feature elements in the importance feature is equal to the number of feature elements in the cross feature, and the feature elements in the importance feature correspond to the feature elements in the cross feature one by one; For each feature element in the importance feature, the element value of the feature element in the importance feature is determined as the feature importance of the corresponding feature element in the cross feature.

11. The method according to claim 10, characterized in that The compressing the cross features to obtain compressed cross features includes: Acquire a compression feature for reducing the dimension of the cross feature, and multiply the cross feature and the compression feature to obtain a reference compression feature, wherein the number of feature elements in the reference compression feature is less than the number of feature elements in the cross feature; Perform feature transformation on the reference compression feature to obtain the compression cross feature.

12. The method according to claim 10, characterized in that The stretching process is performed on the compressed cross feature to obtain the importance feature, including: Acquire a stretch feature for increasing the dimension of the compressed cross feature, and multiply the stretch feature by the compressed cross feature to obtain a reference importance feature, wherein the number of feature elements in the reference importance feature is equal to the number of feature elements in the cross feature; Performing feature transformation on the reference importance feature to obtain the importance feature.

13. A media information recommendation device, characterized in that: The device comprises: An acquisition module, used to acquire media information features respectively corresponding to a plurality of media information to be recommended, and object features of an object to be recommended; A feature crossover module, used to perform feature crossover on the media information features corresponding to each of the media information to be recommended and the object features, respectively, to obtain crossover features corresponding to each of the media information to be recommended, wherein the crossover features include at least two feature elements; An evaluation module, configured to evaluate the importance of each characteristic element in the cross-characteristic corresponding to each of the media information to be recommended, obtain the characteristic importance of each characteristic element, and determine the first recommended parameter value of the media information to be recommended in combination with the characteristic importance of each characteristic element and the element value of each characteristic element; The recommendation module is configured to select target media information from the plurality of media information to be recommended based on the first recommendation parameter value, and recommend the target media information to the object to be recommended.

14. An electronic device, characterized in that: The electronic device comprises: A memory for storing computer executable instructions or computer programs; A processor, configured to implement the method for recommending media information according to any one of claims 1 to 12 when executing the computer executable instructions or computer programs stored in the memory.

15. A computer-readable storage medium storing computer-executable instructions, characterized in that: When the computer executable instructions are executed by a processor, the method for recommending media information according to any one of claims 1 to 12 is implemented.

16. A computer program product comprising a computer program or computer executable instructions, characterized in that When the computer program or computer executable instructions are executed by a processor, the method for recommending media information according to any one of claims 1 to 12 is implemented.

Images