Media resource recommendation method and device, electronic equipment and storage medium

Abstract

The invention relates to a media resource recommendation method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining current resource representation data corresponding to a current media resource; the current resource characterization data are input into an extended interest reasoning model for extended interest reasoning processing, predicted extended reasoning data corresponding to the current media resource are obtained, and the predicted extended reasoning data are reasoning process data of target extended interest data corresponding to the current media resource based on the current resource characterization data; the target extension interest data is data for performing interest extension on the current media resource from content correlation and / or content continuity; based on target extension interest data in the prediction extension reasoning data, recommended media resources are determined, and the recommended media resources are media resources recommended after the current media resources. According to the embodiment of the invention, on the basis of meeting the continuous consumption demand of the user, the innovativeness of the recommended media resources is improved, and the recommendation quality and effect can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular, to a method, apparatus, electronic device, and storage medium for media resource recommendation. Background Art

[0002] With the development of Internet technology, media resources such as videos have grown explosively, and the number of users and media resources on various media resource platforms have been increasing continuously. In the field of media resource recommendation, providing users with accurate and interest-compliant media resources to enhance user experience and platform user stickiness has become an important research direction for major media resource platforms.

[0003] In related technologies, traditional recommendation algorithms based on collaborative filtering and content filtering are often used for media resource recommendation. Among them, the collaborative filtering algorithm analyzes the historical behavior data of users to find user groups with similar interests to the target user; then, recommends the media resources liked by these similar users to the target user. The content filtering algorithm recommends videos similar to the content of the user's historical viewing videos based on the content features of the media resources, such as video tags, classifications, etc. However, traditional recommendation algorithms often lack a deep understanding of the content of media resources and cannot deeply understand the internal connections between media resources in terms of plot, theme, etc. For example, for a TV drama, it is impossible to accurately recommend the next episode or relevant plot extension content, and it is difficult to meet the user's continuous consumption needs; moreover, combined with historical behavior data, the recommended media resources are often relatively homogeneous, and it is difficult to discover the potential interests of users, and it cannot bring freshness and incremental content to users, resulting in a lack of innovation in the recommended media resources, poor media resource recommendation quality and recommendation effect, and inability to effectively meet the user's media resource browsing needs. Furthermore, there are also problems such as waste of system resources caused by ineffective media resource recommendations. Summary of the Invention

[0004] The present disclosure provides a method, apparatus, electronic device, and storage medium for media resource recommendation to at least solve the technical problems in related technologies that media resource recommendation cannot meet the user's continuous consumption needs, cannot bring freshness and incremental content to users, the recommended media resources lack innovation, and thus the media resource recommendation quality and recommendation effect are poor, and it is impossible to effectively meet the user's media resource browsing needs, as well as the waste of system resources caused by ineffective media resource recommendations. The technical solutions of the present disclosure are as follows:

[0005] According to the first aspect of the embodiments of the present disclosure, a method for media resource recommendation is provided, including:

[0006] Obtain current resource characterization data corresponding to the current media resource;

[0007] Input the current resource representation data into an extended interest inference model for extended interest inference processing to obtain predicted extended inference data corresponding to the current media resource. The predicted extended inference data is the inference process data for inferring target extended interest data corresponding to the current media resource based on the current resource representation data. The target extended interest data is data obtained by extending the interest in the current media resource in terms of content relevance and / or content continuity.

[0008] Based on the target extended interest data in the predicted extended inference data, determine recommended media resources. The recommended media resources are media resources recommended after the current media resource.

[0009] In an alternative embodiment, the predicted extended inference data includes a plurality of first extended inference data. The step of inputting the current resource representation data into an extended interest inference model for extended interest inference processing to obtain the predicted extended inference data corresponding to the current media resource includes:

[0010] Input the current resource representation data into the extended interest inference model multiple times for extended interest inference processing to obtain the plurality of first extended inference data corresponding to the current media resource.

[0011] Input the plurality of first extended inference data into a first quality analysis model for inference quality analysis to obtain first quality index data corresponding to each of the plurality of first extended inference data. The first quality index data represents the accuracy of the corresponding first extended inference data.

[0012] The step of determining recommended media resources based on the target extended interest data in the predicted extended inference data includes:

[0013] Based on the target extended interest data in the target extended inference data, determine the recommended media resources. The target extended inference data is the second extended inference data with the largest first quality index data among the plurality of first extended inference data.

[0014] In an alternative embodiment, the extended interest inference model is trained in the following manner:

[0015] Obtain first resource representation data corresponding to sample media resources.

[0016] Based on the first resource representation data, perform extensibility identification on the sample media resources to obtain a first extensibility identification result of the sample media resources.

[0017] In the case that the first ductility recognition result of the target media resource indicates that the target media resource has ductility, obtain at least one sample extended interest data corresponding to the target media resource, where the target media resource is any one of the sample media resources; the at least one sample extended interest data is data obtained by extending the interest in the target media resource from the aspects of content relevance and / or content continuity.

[0018] Input the first resource characterization data and the at least one sample extended interest data into a first preset large model for extended interest inference processing to obtain preset extended inference data corresponding to the sample media resource, where the preset extended inference data is inference process data for inferring the extended interest data corresponding to the sample media resource based on the first resource characterization data.

[0019] Based on the first resource characterization data and the preset extended inference data, perform extended interest inference training on the model to be trained to obtain an extended interest inference model.

[0020] In an alternative embodiment, the step of inputting the first resource characterization data and the at least one sample extended interest data into a first preset large model for extended interest inference processing to obtain preset extended inference data corresponding to the sample media resource includes:

[0021] Input the first resource characterization data, the at least one sample extended interest data, and the first ductility recognition result into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource.

[0022] In an alternative embodiment, the method further includes:

[0023] In the case that the first ductility recognition result of the target media resource indicates that the target media resource has ductility, perform ductility recognition on the at least one sample extended interest data corresponding to the target media resource to obtain a second ductility recognition result of the at least one sample extended interest data.

[0024] The step of inputting the first resource characterization data, the at least one sample extended interest data, and the first ductility recognition result into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource includes:

[0025] Input the first ductility recognition result, the second ductility recognition result, the first resource characterization data, and the at least one sample extended interest data into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource.

[0026] Among them, the preset extended inference data corresponding to the first media resource includes the at least one sample extended interest data corresponding to the first media resource; the preset extended inference data corresponding to the second media resource includes the sample extended interest data with extensibility among the at least one sample extended interest data corresponding to the second media resource, where the first media resource is a media resource in the target media resource for which all the corresponding at least one sample extended interest data has extensibility; the second media resource is a media resource in the target media resource where there is sample extended interest data that does not have extensibility.

[0027] In an optional embodiment, the obtaining the second extensibility recognition result of the at least one sample extended interest data corresponding to the target media resource by performing extensibility recognition on the at least one sample extended interest data corresponding to the target media resource includes:

[0028] Inputting the at least one sample extended interest data into a second preset large model, performing extensibility recognition on the at least one sample extended interest data, and obtaining the second extensibility recognition result.

[0029] In an optional embodiment, the at least one sample extended interest data includes at least one of: first extended interest data, second extended interest data, and custom third extended interest data; the first extended interest data is obtained in the following manner:

[0030] Inputting first resource characterization data into a third preset large model for interest extension processing to obtain the first extended interest data;

[0031] The second extended interest data is obtained in the following manner:

[0032] Obtaining search operation data of at least one historical browsing object during the process of browsing the sample media resource, where the search operation data includes at least one keyword searched by the at least one historical browsing object during the process of browsing the sample media resource;

[0033] Determining the search times corresponding to the at least one keyword;

[0034] Taking the keywords with search times greater than or equal to a preset threshold as the second extended interest data.

[0035] In an optional embodiment, the method further includes:

[0036] Based on at least one preset filtering dimension, performing filtering processing on the at least one sample extended interest data to obtain filtered extended interest data;

[0037] Inputting the first resource characterization data and the at least one sample extended interest data into a first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource includes:

[0038] Inputting the first resource characterization data and the filtered extended interest data into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource.

[0039] In an alternative embodiment, the obtaining of the first ductility recognition result of the sample media resource based on the first resource characterization data includes:

[0040] Inputting the first resource characterization data into a fourth preset large model to perform ductility recognition on the sample media resource to obtain the first ductility recognition result.

[0041] In an alternative embodiment, after training the model to be trained for extended interest inference based on the first resource characterization data and the preset extended inference data to obtain an extended interest inference model, the method further includes:

[0042] Obtaining second resource characterization data corresponding to the test media resource;

[0043] Inputting the second resource characterization data into the extended interest inference model multiple times for extended interest inference processing to obtain multiple second extended inference data corresponding to each media resource in the test media resource;

[0044] Inputting the multiple second extended inference data into a second quality analysis model for inference quality analysis to obtain second quality index data corresponding to each of the multiple second extended inference data, where the second quality index data characterizes the accuracy of the corresponding extended inference data;

[0045] Taking the extended inference data with the largest second quality index data among the multiple second extended inference data corresponding to each media resource as the positive sample data corresponding to each media resource, and taking the extended inference data with the smallest second quality index data among the multiple second extended inference data corresponding to each media resource as the negative sample data corresponding to each media resource;

[0046] Adjusting and training the extended interest inference model based on the second resource characterization data, the positive sample data, and the negative sample data.

[0047] According to the second aspect of the embodiments of the present disclosure, another media resource recommendation method is provided, including:

[0048] Display the current media resource on the preset page;

[0049] In response to a media resource switching instruction, display on the preset page a recommended media resource associated with target extended interest data corresponding to the current media resource;

[0050] Wherein, the target extended interest data is data obtained by extending the interest in the current media resource in terms of content relevance and / or content continuity; and the target extended interest data is included in the predicted extended inference data corresponding to the current media resource, and the predicted extended inference data is obtained by the extended interest inference model performing extended interest inference based on the current resource representation data corresponding to the current media resource, and the predicted extended inference data is the inference process data for inferring the target extended interest data based on the current resource representation data.

[0051] According to a third aspect of the embodiments of the present disclosure, there is provided a media resource recommendation device, including:

[0052] A first representation data acquisition module, configured to acquire current resource representation data corresponding to the current media resource;

[0053] A first extended interest inference module, configured to perform extended interest inference processing on the current resource representation data by inputting it into an extended interest inference model, to obtain predicted extended inference data corresponding to the current media resource, and the predicted extended inference data is the inference process data for inferring target extended interest data corresponding to the current media resource based on the current resource representation data; the target extended interest data is data obtained by extending the interest in the current media resource in terms of content relevance and / or content continuity;

[0054] A recommended media resource determination module, configured to determine a recommended media resource based on the target extended interest data in the predicted extended inference data, and the recommended media resource is a media resource recommended after the current media resource.

[0055] In an optional embodiment, the predicted extended inference data includes a plurality of first extended inference data; the first extended interest inference module includes:

[0056] A first extended interest inference unit, configured to perform extended interest inference processing on the current resource representation data by inputting it into the extended interest inference model multiple times, to obtain the plurality of first extended inference data corresponding to the current media resource;

[0057] An inference quality analysis unit, configured to perform inference quality analysis on the multiple first extended inference data by inputting them into a first quality analysis model, to obtain first quality index data corresponding to each of the multiple first extended inference data, where the first quality index data characterizes the accuracy of the corresponding first extended inference data;

[0058] The recommended media resource determination module is further configured to perform determining the recommended media resource based on the target extended interest data in the target extended inference data, where the target extended inference data is the second extended inference data with the largest first quality index data among the multiple first extended inference data.

[0059] In an alternative embodiment, the extended interest inference model is trained using the following modules:

[0060] A second characterization data acquisition module, configured to perform acquiring first resource characterization data corresponding to a sample media resource;

[0061] A first extensibility recognition module, configured to perform extensibility recognition on the sample media resource based on the first resource characterization data, to obtain a first extensibility recognition result of the sample media resource;

[0062] An extended interest data acquisition module, configured to perform acquiring at least one sample extended interest data corresponding to the target media resource in the case where the first extensibility recognition result of the target media resource indicates that the target media resource has extensibility, where the target media resource is any one of the sample media resources; the at least one sample extended interest data is data obtained by extending the interest in the target media resource from content relevance and / or content continuity;

[0063] A second extended interest inference module, configured to perform extended interest inference processing on the first resource characterization data and the at least one sample extended interest data by inputting them into a first preset large model, to obtain preset extended inference data corresponding to the sample media resource, where the preset extended inference data is inference process data for inferring the extended interest data corresponding to the sample media resource based on the first resource characterization data;

[0064] A model training module, configured to perform extended interest inference training on a model to be trained based on the first resource characterization data and the preset extended inference data, to obtain an extended interest inference model.

[0065] In an alternative embodiment, the second extended interest inference module includes:

[0066] The second extended interest inference unit is configured to perform extended interest inference processing by inputting the first resource characterization data, the at least one sample extended interest data, and the first extensibility recognition result into the first preset large model, so as to obtain the preset extended inference data corresponding to the sample media resource.

[0067] In an optional embodiment, the apparatus further includes:

[0068] The second extensibility recognition module is configured to perform extensibility recognition on the at least one sample extended interest data corresponding to the target media resource to obtain a second extensibility recognition result of the at least one sample extended interest data when the first extensibility recognition result of the target media resource indicates that the target media resource has extensibility;

[0069] The second extended interest inference unit includes:

[0070] The third extended interest inference unit is configured to perform extended interest inference processing by inputting the first extensibility recognition result, the second extensibility recognition result, the first resource characterization data, and the at least one sample extended interest data into the first preset large model, so as to obtain the preset extended inference data corresponding to the sample media resource;

[0071] Among them, the preset extended inference data corresponding to the first media resource includes the at least one sample extended interest data corresponding to the first media resource; the preset extended inference data corresponding to the second media resource includes the sample extended interest data with extensibility among the at least one sample extended interest data corresponding to the second media resource, where the first media resource is the media resource in the target media resource for which all the at least one sample extended interest data have extensibility; the second media resource is the media resource in the target media resource where there is sample extended interest data that does not have extensibility.

[0072] In an optional embodiment, the second extensibility recognition module is specifically configured to perform extensibility recognition on the at least one sample extended interest data by inputting the at least one sample extended interest data into a second preset large model to obtain the second extensibility recognition result.

[0073] In an optional embodiment, the at least one sample extended interest data includes at least one of first extended interest data, second extended interest data, and custom third extended interest data; the first extended interest data is obtained by using the following module:

[0074] The interest extension processing module is configured to perform interest extension processing by inputting the first resource characterization data into a third preset large model to obtain the first extended interest data;

[0075] The second extended interest data is obtained by using the following modules:

[0076] A search operation data acquisition module, configured to execute acquiring search operation data of at least one historical browsing object during browsing the sample media resource, where the search operation data includes at least one keyword searched by the at least one historical browsing object during browsing the sample media resource;

[0077] A search times determination module, configured to execute determining the search times corresponding to the at least one keyword;

[0078] An extended interest data determination module, configured to execute using the keyword whose search times is greater than or equal to a preset threshold as the second extended interest data.

[0079] In an alternative embodiment, the apparatus further includes:

[0080] A filtering processing module, configured to execute filtering processing on the at least one sample extended interest data based on at least one preset filtering dimension to obtain filtered extended interest data;

[0081] The second extended interest inference module is further configured to execute inputting the first resource characterization data and the filtered extended interest data into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource.

[0082] In an alternative embodiment, the first extensibility recognition module is specifically configured to execute inputting the first resource characterization data into a fourth preset large model to perform extensibility recognition on the sample media resource to obtain the first extensibility recognition result.

[0083] In an alternative embodiment, the apparatus further includes:

[0084] A third characterization data acquisition module, configured to execute acquiring second resource characterization data corresponding to a test media resource after performing extended interest inference training on a to-be-trained model based on the first resource characterization data and the preset extended inference data to obtain an extended interest inference model;

[0085] A third extended interest inference module, configured to execute inputting the second resource characterization data into the extended interest inference model multiple times for extended interest inference processing to obtain multiple second extended inference data corresponding to each media resource in the test media resource;

[0086] An inference quality analysis module, configured to perform inference quality analysis on the multiple second extended inference data by inputting them into a second quality analysis model, to obtain second quality index data corresponding to each of the multiple second extended inference data, where the second quality index data characterizes the accuracy of the corresponding extended inference data;

[0087] A positive and negative sample determination module, configured to perform taking the extended inference data with the largest second quality index data among the multiple second extended inference data corresponding to each media resource as the positive sample data corresponding to each media resource, and taking the extended inference data with the smallest second quality index data among the multiple second extended inference data corresponding to each media resource as the negative sample data corresponding to each media resource;

[0088] A model adjustment module, configured to perform adjustment training on the extended interest inference model based on the second resource characterization data, the positive sample data, and the negative sample data.

[0089] According to a fourth aspect of the embodiments of the present disclosure, there is provided another media resource recommendation device, including:

[0090] A current media resource display module, configured to perform displaying the current media resource on a preset page;

[0091] A recommended media resource display module, configured to perform displaying, in response to a media resource switching instruction, recommended media resources associated with target extended interest data corresponding to the current media resource on the preset page;

[0092] Wherein, the target extended interest data is data obtained by extending the interest in terms of content relevance and / or content continuity of the current media resource; and the target extended interest data is included in the predicted extended inference data corresponding to the current media resource, and the predicted extended inference data is obtained by the extended interest inference model performing extended interest inference based on the current resource characterization data corresponding to the current media resource, and the predicted extended inference data is the inference process data for inferring the target extended interest data based on the current resource characterization data.

[0093] According to a fifth aspect of the embodiments of the present disclosure, there is provided an electronic device, including: a processor; a memory for storing executable instructions of the processor; wherein, the processor is configured to execute the instructions to implement the method according to any one of the media resource recommendation methods in the embodiments of the present disclosure.

[0094] According to a sixth aspect of the embodiments of the present disclosure, there is provided a computer-readable storage medium, when the instructions in the storage medium are executed by a processor of an electronic device, enabling the electronic device to execute the method according to any one of the media resource recommendation methods in the embodiments of the present disclosure.

[0095] According to a seventh aspect of the embodiments of the present disclosure, there is provided a computer program product including instructions, which when running on a computer, causes the computer to execute the method described in any one of the media resource recommendation methods of the embodiments of the present disclosure.

[0096] The technical solutions provided by the embodiments of the present disclosure at least bring the following beneficial effects:

[0097] In the process of media resource recommendation, obtain the current resource characterization data corresponding to the currently browsed media resource; and input the current resource characterization data into the extended interest inference model for extended interest inference processing to obtain the inference process data (predicted extended inference data) of the target extended interest data corresponding to the current media resource inferred based on the current resource characterization data, so that the model can infer through coherent logic the data (target extended interest data) for interest extension of the current media resource from content relevance and / or content continuity, which can perform more targeted interest extension. And based on the target extended interest data, determine the media resources to be recommended after the current media resource, which can combine the interest extension based on content continuity to meet the user's continuous consumption needs, and combine the interest extension based on content relevance to bring freshness and incremental content to the user, improve the innovation of the recommended media resources, and thus can improve the quality and effect of media resource recommendation, more effectively meet the user's media resource browsing needs, and reduce the waste of system resources caused by ineffective media resource recommendations.

[0098] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure, and do not constitute an improper limitation to the present disclosure.

[0100] Figure 1 is a schematic diagram of an application environment shown according to an exemplary embodiment;

[0101] Figure 2 is a flowchart of a media resource recommendation method shown according to an exemplary embodiment;

[0102] Figure 3 is a flowchart of a training process of an extended interest inference model shown according to an exemplary embodiment;

[0103] Figure 4 is a schematic diagram of a training process of an extended interest inference model shown according to an exemplary embodiment;

[0104] Figure 5 It is a schematic diagram showing the process of inferring target extended interest data corresponding to the current media resource based on an extended interest inference model according to an exemplary embodiment;

[0105] Figure 6 It is a flowchart of another media resource recommendation method according to an exemplary embodiment;

[0106] Figure 7 It is a block diagram of a media resource recommendation device according to an exemplary embodiment;

[0107] Figure 8 It is a block diagram of another media resource recommendation device according to an exemplary embodiment;

[0108] Figure 9 It is a block diagram of an electronic device for media resource recommendation according to an exemplary embodiment;

[0109] Figure 10 It is a block diagram of another electronic device for media resource recommendation according to an exemplary embodiment. Detailed implementation manners

[0110] In order to enable those of ordinary skill in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0111] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that such used data may be interchanged under appropriate circumstances so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0112] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in the present disclosure are all information and data authorized by the user or fully authorized by all parties.

[0113] Please refer to Figure 1 , Figure 1 It is a schematic diagram of an application environment according to an exemplary embodiment, and the application environment may include a terminal 100 and a server 200.

[0114] In an optional embodiment, the terminal 100 can be used to provide services such as media resource recommendation to users. Specifically, the terminal 100 can include, but is not limited to, electronic devices such as smart phones, desktop computers, tablet computers, laptop computers, smart speakers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, smart wearable devices, etc., and can also be software running on the above-mentioned electronic devices, such as application programs, etc. Optionally, the operating systems running on the electronic devices can include, but are not limited to, Android system, IOS system, Linux, Windows, etc.

[0115] In an optional embodiment, the server 200 can provide background services for the terminal 100. Specifically, the server 200 can be used to pre-train an extended interest inference model and push media resources to the terminal 100 based on the extended interest inference model; 200 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0116] In addition, it should be noted that Figure 1 The application environment shown is only one provided by the present disclosure. In actual applications, other application environments may also be included. For example, the terminal 100 can perform media resource recommendation locally based on the extended interest inference model.

[0117] In the embodiments of this specification, the above-mentioned terminal 100 and server 200 can be directly or indirectly connected through wired or wireless communication methods, and the present disclosure does not limit this.

[0118] Figure 2 is a flowchart of a media resource recommendation method shown according to an exemplary embodiment. This method can be applied to a server, such as Figure 2 shown, and this method can include the following steps:

[0119] In step S201, obtain the current resource characterization data corresponding to the current media resource.

[0120] In a specific embodiment, the current media resource can be the media resource currently recommended for display; specifically, the media resource includes at least one of video, picture and text, audio, text, and picture. The current resource characterization data can be data that can characterize the current media resource. Exemplarily, the current resource characterization data can include resource content and resource description information; specifically, the resource content can be the content contained in the media resource itself, such as pictures, audio, subtitles, cover images, etc. in the video; the resource description information can be information describing the media resource; for example, theme information, brief introduction information, resource category, etc.

[0121] In step S203, the current resource representation data is input into the extended interest inference model for extended interest inference processing to obtain the predicted extended inference data corresponding to the current media resource.

[0122] In a specific embodiment, the above predicted extended inference data may be the inference process data for inferring the target extended interest data corresponding to the current media resource based on the current resource representation data; the above target extended interest data may be data for extending the interest in the current media resource from the aspects of content relevance and / or content continuity. Specifically, the data for extending the interest from the aspect of content relevance may be the newly generated interest data (extended content) based on the content related to the theme of the media resource; the data for extending the interest from the aspect of content continuity may be the extended content that is continuous with the media resource in terms of time and / or plot; the predicted extended inference data may include: the current resource representation data, the key information extracted based on the current resource representation data, the interest content (possibly interested content) of the resource browsing object (user account) speculated based on the key information, the determination of whether the current media resource has extensibility and the target extended interest data based on the extension from the key information to the interest content.

[0123] In an alternative embodiment, the extended interest inference model may be a deep learning model for performing extended interest inference processing; specifically, the model structure of the extended interest inference model may be set according to the actual application, such as the Transformer model or a variant model of the Transformer model, etc.; optionally, the above extended interest inference model is trained in the following manner, as Figure 3 shown, and may include the following steps:

[0124] In step S301, the first resource representation data corresponding to the sample media resource is obtained;

[0125] In step S303, based on the first resource representation data, the extensibility of the sample media resource is identified to obtain the first extensibility identification result of the sample media resource;

[0126] In step S305, when the first extensibility identification result of the target media resource indicates that the target media resource has extensibility, at least one sample extended interest data corresponding to the target media resource is obtained;

[0127] In step S307, the first resource representation data and at least one sample extended interest data are input into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource;

[0128] In step S309, based on the first resource representation data and the preset extended reasoning data, extended interest reasoning training is performed on the model to be trained to obtain an extended interest reasoning model.

[0129] In a specific embodiment, the sample media resources may be media resources used for training an extended interest reasoning model; specifically, the sample media resources may include multiple media resources; the above-mentioned first resource representation data may be data that can represent the sample media resources (each media resource in the multiple media resources). Specifically, the specific details of the first resource representation data can refer to the specific details of the above-mentioned current resource representation data, which will not be repeated here.

[0130] In an optional embodiment, the scalability identification of the sample media resource based on the first resource representation data to obtain the first scalability identification result of the sample media resource may include:

[0131] The first resource representation data is input into a fourth preset large model, and extensibility identification is performed on the sample media resource to obtain a first extensibility identification result.

[0132] In a specific embodiment, the fourth preset large model can be a preset generative large language model, which can be set in combination with actual application requirements; specifically, during the use of the generative large language model, it is often necessary to input corresponding instructions; accordingly, in the scenario where the first resource representation data is input into the fourth preset large model and the extensibility of the sample media resource is identified, the first extension identification instruction can also be input into the fourth preset large model; specifically, the first extension identification instruction can be used to instruct the extensibility identification of the sample media resource; illustratively, the first extension identification instruction can be: help me identify the extensibility of "xx" (the name of the sample media resource and other identification information).

[0133] In a specific embodiment, the first extensibility identification result can be used to indicate whether the sample media resource (each media resource among multiple media resources) has extensibility (continuity and / or relevance). Specifically, the extensibility of the media resource can be characterized by the continuity and / or relevance of the media resource; wherein, the continuity of the media resource can be characterized by the possibility of having extended content that is continuous with it in time and / or plot; the relevance of the media resource can be characterized by the possibility of deriving new points of interest (extended content) based on the content related to the theme of the media resource.

[0134] In the above embodiment, the first resource representation data is input into the fourth preset big model to perform extensibility identification on the sample media resource. This can be combined with the generative big language model (the fourth preset big model) to greatly improve the convenience and accuracy of extensibility identification of the media resource, and then the extended interest data can be generated in a targeted manner in combination with the extensibility identification results of the media resource.

[0135] In addition, it should be noted that, in practical applications, a pre-trained resource scalability identification model (a deep learning model used for scalability identification of media resources) can also be used to perform scalability identification of sample media resources.

[0136] In a specific embodiment, the target media resource may be any media resource among the sample media resources (media resources with extensibility); at least one sample extended interest data may be data for extending the interest of the target media resource from the perspective of content relevance and / or content continuity.

[0137] In an optional embodiment, the at least one sample extended interest data may include: at least one of: first extended interest data, second extended interest data and customized third extended interest data; Optionally, the first extended interest data is obtained in the following manner: inputting the first resource representation data into the third preset large model for interest extension processing to obtain the first extended interest data;

[0138] In a specific embodiment, the first extended interest data may be extended interest data of a sample media resource generated based on the third preset macro model; specifically, the first extended interest data may include one or more extended interest data;

[0139] In a specific embodiment, the third preset big model can be a preset generative big language model, which can be set in combination with actual application requirements; specifically, in a scenario where the first resource representation data is input into the third preset big model for interest extension processing, the interest extension instruction can also be input into the third preset big model; specifically, the interest extension instruction can be used to instruct interest extension processing of sample media resources; exemplarily, the interest extension instruction can be: help me generate an extended interest search term (extended interest data) of "xx" (name of the sample media resource and other identification information).

[0140] In an optional embodiment, the second extended interest data may be obtained in the following manner:

[0141] Acquire search operation data of at least one historical browsing object in a process of browsing the sample media resource, where the search operation data includes at least one keyword searched by the at least one historical browsing object in the process of browsing the sample media resource;

[0142] Determine the number of searches corresponding to at least one keyword;

[0143] Keywords with a search count greater than or equal to a preset threshold are used as second extended interest data.

[0144] In a specific embodiment, the second extended interest data may be extended interest data determined based on the corresponding number of searches; specifically, at least one historical browsing object may be a user account that has browsed the sample media resource; at least one keyword may include a search keyword independently input by the historical browsing object and / or a search keyword recommended by the platform; the number of searches corresponding to any keyword may be the number of searches based on the keyword performed by at least one historical browsing object corresponding to each media resource in the sample media resource during the browsing of the media resource.

[0145] In a specific embodiment, the third extended interest data may be pre-defined extended interest data; specifically, the customized extended interest data may be determined based on content relevance and / or content continuity in combination with actual application requirements.

[0146] In a specific embodiment, taking a media resource in the sample media resources as an explanation work (such as a video) of the nth episode in a collection of explanation works of a certain drama xx as an example, the first extended interest data may include search keywords for content related to a certain plot in the nth episode of drama xx; the second extended interest data may include movie xx (a movie with the same name as the drama) and at least one search keyword related to a certain plot in the nth episode of drama xx; the third extended interest data may include the sequel to a certain plot in the nth episode of drama xx, the n+1th episode of drama xx, and the collection of drama xx.

[0147] In the above embodiment, combining the generative large language model (the third preset large model), customization, and search operation data of historical browsing objects of media resources can reflect the relationship between media resources and extended interests from different angles, greatly improving the diversity and richness of extended interests of media resources, and further providing users with more novel and interesting media resource recommendations, enriching users' media resource browsing needs.

[0148] In a specific embodiment, the first preset big model can be a preset generative big language model, which can be set specifically in combination with actual application requirements; in the scenario where the first resource representation data and at least one sample extended interest data are input into the first preset big model for extended interest reasoning processing, the extended interest reasoning instruction can also be input into the first preset big model; specifically, the extended interest reasoning instruction can be used to instruct the interest extension reasoning process for the sample media resource; exemplarily, the extended interest reasoning instruction can be: help me generate the reasoning process data of the extended interest data of "xx" (the name of the sample media resource and other identification information).

[0149] In a specific embodiment, the preset extended inference data may be the inference process data for inferring the extended interest data corresponding to the sample media resource based on the first resource representation data. Specifically, the preset extended inference data corresponding to any media resource may include the resource representation data corresponding to the media resource, the key information extracted based on the resource representation data corresponding to the media resource, the interest content (possible interested content) of the resource browsing object (user account) speculated based on the key information, the extension from the key information to the interest content, and the judgment as to whether the media resource has extensibility and the extended interest data.

[0150] In a specific embodiment, taking the explanation work (media resource) of the nth episode in the explanation collection of a certain drama xx as an example, the preset extended inference data may include: the resource representation data of the explanation work (video) of the nth episode (such as video title, video cover image, actors in the video, video category, audio text of the video, video subtitles, etc.), the key information extracted based on the resource representation data (such as plot 1, plot 2, and plot 3); the content that the resource browsing object may be interested in speculated based on the key information (such as the follow-up of plot 1, search content related to the theme of plot 2, search content related to the theme of plot 3), the extension from the key information to the interested content, and the judgment as to whether the media resource has extensibility (for example, combining plot 1 with the follow-up of plot 1 can determine that there is a clear plot development, and users may be interested in the follow-up development of plot 1. Therefore, this video has continuity; assuming that plot 2 involved in the video is the life of a certain character at a certain stage, it can be inferred that users may be interested in the personal growth and emotional changes of this character. Therefore, this video has relevance) and the inferred extended interest data (such as the extended interest data of continuity: the follow-up of plot 1 in the nth episode of drama xx, the (n + 1)th episode of drama xx; the extended interest data of relevance: the collection of drama xx, at least one search keyword related to a certain plot in the nth episode of drama xx, movie xx (a movie with the same name as the drama)).

[0151] In addition, it should be noted that the extended interest data corresponding to the media resource without extensibility in the sample media resource may be empty (that is, there is no extended interest data). Correspondingly, the preset extended inference data corresponding to the media resource without extensibility in the sample media resource may also include the resource representation data corresponding to the media resource, the key information extracted based on the resource representation data corresponding to the media resource, the content that the resource browsing object (user account) may be interested in speculated based on the key information, the extension from the key information to the interested content, the judgment as to whether the media resource has extensibility, and the inferred extended interest data; among them, the extended interest data is empty.

[0152] In an optional embodiment, inputting the first resource characterization data and at least one sample extended interest data into a first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource may include:

[0153] Input the first resource characterization data, at least one sample extended interest data, and the first extensibility recognition result into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource.

[0154] In a specific embodiment, inputting the first extensibility recognition result together into the first preset large model for extended interest inference processing can enable the generative large language model to clarify whether the corresponding media resource has extensibility during the extended interest inference process.

[0155] In the above embodiment, when inputting the first resource characterization data, at least one sample extended interest data, and the first extensibility recognition result together into the first preset large model for extended interest inference processing, the generative large language model can clarify whether the corresponding media resource has extensibility during the extended interest inference process, thereby better improving the effectiveness of interest extension inference and the effectiveness of extended interest data.

[0156] In an optional embodiment, the extended interest data corresponding to any media resource in the target media resource may be at least one sample extended interest data corresponding to the media resource. Optionally, it is also possible to filter the sample extended interest data that does not have extensibility in combination with the generative large language model for extended interest inference processing, that is, the extended interest data corresponding to any media resource in the target media resource may be the sample extended interest data with extensibility among at least one sample extended interest data corresponding to the media resource; correspondingly, the above method may further include:

[0157] When the first extensibility recognition result of the target media resource indicates that the target media resource has extensibility, perform extensibility recognition on at least one sample extended interest data corresponding to the target media resource to obtain a second extensibility recognition result of at least one sample extended interest data;

[0158] Correspondingly, the above inputting the first resource characterization data, at least one sample extended interest data, and the first extensibility recognition result into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource may include:

[0159] Input the first extensibility recognition result, the second extensibility recognition result, the first resource characterization data, and at least one sample extended interest data into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource;

[0160] In an optional embodiment, the ductility recognition of at least one sample extended interest data corresponding to the target media resource, and the second ductility recognition result of at least one sample extended interest data obtained includes:

[0161] Input at least one sample extended interest data into a second preset large model, perform ductility recognition on at least one sample extended interest data, and obtain a second ductility recognition result.

[0162] In a specific embodiment, the second preset large model can be a preset generative large language model, which can be specifically set according to actual application requirements. Specifically, the second ductility recognition result can be used to indicate whether the corresponding sample extended interest data has ductility (continuity and / or relevance). Specifically, the fact that the sample extended interest data has ductility can characterize that the sample extended interest data has continuity and / or relevance; among them, the fact that the sample extended interest data has continuity can characterize that the extended content corresponding to the sample extended interest data has continuity in time and / or plot with the corresponding media resource; the fact that the sample extended interest data has relevance can characterize that the extended content corresponding to the sample extended interest data is a derivative content related to the theme of the corresponding media resource.

[0163] In the above embodiment, inputting at least one sample extended interest data into the second preset large model and performing ductility recognition on at least one sample extended interest data can greatly improve the convenience and accuracy of the recognition of sample extended interest data, and then more targeted interest extension can be carried out in combination with the ductility recognition result of the sample extended interest data.

[0164] In addition, it should be noted that in actual applications, the ductility recognition of sample media resources can also be combined with a pre-trained interest data ductility recognition model (a deep learning model for performing ductility recognition on extended interest data).

[0165] In a specific embodiment, the second ductility recognition result can be used to help the generative large language model (the first preset large model) filter out sample extended interest data that does not have ductility during the extended interest inference process; correspondingly, the preset extended inference data corresponding to the first media resource in the target media resource includes at least one sample extended interest data corresponding to the first media resource; the preset extended inference data corresponding to the second media resource in the target media resource includes the sample extended interest data with ductility among at least one sample extended interest data corresponding to the second media resource. The first media resource is a media resource in the target media resource for which all corresponding at least one sample extended interest data has ductility; the second media resource is a media resource in the target media resource for which there is sample extended interest data that does not have ductility.

[0166] In the above embodiments, by inputting the first ductility recognition result, the second ductility recognition result, the first resource characterization data, and at least one sample extended interest data into the first preset large model for extended interest inference processing, on the basis of enabling the generative large language model to determine whether the corresponding media resource has ductility, the sample extended interest data that does not have ductility in the input can be filtered, greatly improving the effectiveness and quality of the extended interest data used for extended interest inference processing, and thus also better improving the effectiveness of interest extension inference.

[0167] In an alternative embodiment, the above method may further include:

[0168] Filtering at least one sample extended interest data based on at least one preset filtering dimension to obtain filtered extended interest data;

[0169] Correspondingly, the above step of inputting the first resource characterization data and at least one sample extended interest data into the first preset large model for extended interest inference processing to obtain preset extended inference data corresponding to the sample media resource may include:

[0170] Inputting the first resource characterization data and the filtered extended interest data into the first preset large model for extended interest inference processing to obtain preset extended inference data corresponding to the sample media resource.

[0171] In a specific embodiment, at least one preset filtering dimension can be set in combination with the actual application. Optionally, at least one preset filtering dimension can include at least one filtering dimension among the invalid character filtering dimension, the single part-of-speech filtering dimension, the first overlap filtering dimension, the empty content filtering dimension, and the second overlap filtering dimension. Specifically, in the process of filtering at least one sample extended interest data based on the invalid character filtering dimension, in combination with a preset invalid character library (including at least one preset invalid character), the preset invalid characters in at least one sample extended interest data can be filtered out. In the process of filtering at least one sample extended interest data based on the single part-of-speech filtering dimension, the sample extended interest data with a single part-of-speech (less part-of-speech types or high-frequency repetition of a certain part-of-speech) can be filtered out. In the process of filtering at least one sample extended interest data based on the first overlap filtering dimension, in combination with the semantic similarity between the sample extended interest data and the resource media content (resource representation data), the sample extended interest data with a semantic similarity greater than the first preset similarity threshold can be filtered out. In the process of filtering at least one sample extended interest data based on the empty content filtering dimension, in combination with a preset empty content recognition model (a deep learning model for identifying empty and non-fine-grained content), the empty content recognition can be performed on at least one sample extended interest data, and the sample extended interest data indicated as belonging to the empty content in the recognition result can be filtered out. In the process of filtering at least one sample extended interest data based on the second overlap filtering dimension, in combination with the semantic similarity between the sample extended interest data corresponding to the same media resource, the deduplication process can be performed on at least two sample extended interest data with a semantic similarity greater than the second preset similarity threshold.

[0172] In addition, it should be noted that in the case where at least one preset filtering dimension includes multiple preset filtering dimensions, the execution order corresponding to the multiple preset filtering dimensions can be set in combination with the actual application.

[0173] In a specific embodiment, for the specific refinement of inputting the first resource representation data and the filtered extended interest data into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource, reference can be made to the above specific refinement of inputting the first resource representation data and at least one sample extended interest data into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource, which will not be elaborated here.

[0174] In the above embodiments, based on at least one preset filtering dimension, filtering processing is performed on at least one sample extended interest data, and extended interest inference processing is performed in combination with the filtered extended interest data, which can greatly improve the effectiveness and quality of the extended interest data used for extended interest inference processing, and thus can also better improve the effectiveness of interest extension inference.

[0175] In a specific embodiment, the preset extended inference data is the inference process data of the extended interest data corresponding to the sample media resource inferred by the (generative large language model) based on the first resource representation data;

[0176] In a specific embodiment, the model to be trained can be an extended interest inference model to be trained; specifically, based on the first resource representation data and the preset extended inference data, performing extended interest inference training on the model to be trained to obtain an extended interest inference model may include:

[0177] Determine the media resource of the current training round from the sample media resources; input the first resource representation data corresponding to the media resource of the current training round into the model to be trained for extended interest inference processing to obtain the sample extended inference data corresponding to the media resource of the current training round; based on the sample extended inference data corresponding to the media resource of the current training round and the preset extended inference data corresponding to the media resource of the current training round, determine the inference loss information corresponding to the model to be trained; update the model parameters of the model to be trained based on the inference loss information, and repeat the above cyclic iterative training steps from determining the media resource of the current training round from the sample media resources to updating the model parameters of the model to be trained based on the inference loss information for the updated model to be trained until the preset convergence condition is satisfied, and use the model to be trained when the preset convergence condition is satisfied as the extended interest inference model.

[0178] In a specific embodiment, the media resource of the current training round can be the media resource in the current cyclic iterative training process; specifically, the media resource of the current training round can be a part of the sample media resources. Specifically, the media resource of the current training round can be randomly selected, or the media resource of the current training round can be determined in combination with a preset selection rule.

[0179] In a specific embodiment, the sample extended inference data corresponding to the media resource of the current training round and the preset extended inference data corresponding to the media resource of the current training round can be substituted into a preset loss function to determine the inference loss information; the inference loss information can characterize the extended interest inference performance of the current model to be trained. Specifically, the preset loss function can be set in combination with the actual application. Specifically, in the process of updating the model parameters, the gradient descent method can be combined. Specifically, the above-mentioned satisfaction of the preset convergence condition can be that the inference loss information is less than or equal to a preset loss threshold, or the number of training iteration steps reaches a preset number, etc.; specifically, the preset loss threshold and the preset number can be set in combination with the model accuracy and training speed requirements in the actual application.

[0180] In the above embodiment, during the training process of the extended interest inference model, based on the first resource characterization data corresponding to the sample media resource, the extensibility of the sample media resource is identified to obtain the first extensibility identification result of the sample media resource, and at least one sample extended interest data corresponding to the target media resource with extensibility is obtained, and the first resource characterization data and the at least one sample extended interest data are input into the first preset large model for extended interest inference processing together. The preset extended inference data for training the extended interest inference model can be generated by combining the generative large language model (the first preset large model), the training data for extended interest inference training (the first resource characterization data and the preset extended inference data) is constructed, and based on the first resource characterization data and the preset extended inference data, the model to be trained is trained for extended interest inference, which can greatly improve the convenience and effectiveness of the extended interest inference model training. Furthermore, in the subsequent media resource recommendation process, in combination with the extended interest inference model, extended content that is continuous with the currently browsed media resource in terms of time and plot can be provided for the user, and the relevance between the recommended media resource and the currently browsed media resource by the user can also be improved, realizing the expansion of the user's new interest points and better improving the user's media resource browsing experience.

[0181] In an alternative embodiment, after training the model to be trained for extended interest inference based on the first resource characterization data and the preset extended inference data to obtain the extended interest inference model, the above method may further include:

[0182] Obtain the second resource characterization data corresponding to the test media resource;

[0183] Input the second resource characterization data into the extended interest inference model multiple times for extended interest inference processing to obtain multiple second extended inference data corresponding to each media resource in the test media resource;

[0184] Input multiple second extended inference data into the second quality analysis model for inference quality analysis to obtain the second quality index data corresponding to each of the multiple second extended inference data;

[0185] Take the extended inference data with the largest second quality index data among the multiple second extended inference data corresponding to each media resource as the positive sample data corresponding to each media resource, and take the extended inference data with the smallest second quality index data among the multiple second extended inference data corresponding to each media resource as the negative sample data corresponding to each media resource;

[0186] Based on the second resource characterization data, positive sample data, and negative sample data, adjust and train the extended interest inference model.

[0187] In a specific embodiment, the test media resource can be a media resource used to adjust the extended interest inference model. The test media resource can include multiple media resources. Optionally, the test media resource can be different from the sample media resource or at least partially the same; the second resource characterization data can be data that can characterize the resource media resource (each media resource among the multiple media resources). Specifically, for the specific refinement of the second resource characterization data, reference can be made to the specific refinement of the current resource characterization data above, which will not be elaborated here.

[0188] In a specific embodiment, the second quality analysis model can be a deep learning model for performing extended inference data quality analysis. Specifically, the above-mentioned second quality index data characterizes the accuracy of the corresponding extended inference data;

[0189] In a specific embodiment, for the specific refinement of adjusting and training the extended interest inference model based on the second resource characterization data, positive sample data, and negative sample data, reference can be made to the specific refinement of obtaining the extended interest inference model by performing extended interest inference training on the model to be trained based on the first resource characterization data and the preset extended inference data above, which will not be elaborated here; that is, replace the preset extended inference data with the positive sample data and negative sample data.

[0190] In the above embodiments, after performing extended interest inference training on the model to be trained based on the first resource characterization data and the preset extended inference data to obtain an extended interest inference model, the second resource characterization data corresponding to the test media resource is input into the extended interest inference model multiple times for extended interest inference processing, obtaining multiple second extended inference data corresponding to each media resource in the test media resource. Then, in combination with the second quality analysis model, the inference quality of the multiple second extended inference data corresponding to each media resource is analyzed to obtain the second quality index data corresponding to each of the multiple second extended inference data. Furthermore, based on the second quality index data corresponding to the multiple second extended inference data, positive and negative sample data can be selected from the multiple second extended inference data corresponding to the same media resource, and the extended interest inference model can be adjusted in combination with the positive and negative sample data, which can better improve the extended interest inference performance of the extended interest inference model. Thus, it is possible to better provide the user with extended content that is continuous with the currently viewed media resource in terms of time and plot, and also improve the relevance between the recommended media resource and the currently viewed media resource by the user, realizing the expansion of the user's new interest points and better enhancing the user's media resource browsing experience.

[0191] In a specific embodiment, as Figure 4 shown, Figure 4 is a schematic diagram of a training process of an extended interest inference model shown according to an exemplary embodiment; combining Figure 4 it can be seen that the extensibility of the media resource can be identified in combination with the first resource characterization data corresponding to the sample media resource. Optionally, in the case where the media resource does not have extensibility, the corresponding first resource characterization data and the recognition result that the media resource does not have extensibility (the first extensibility recognition result) can be directly input into the first preset large model for extended interest inference processing to obtain the corresponding preset extended inference data. In the case where the media resource has extensibility, the corresponding sample extended interest data can be constructed, and the constructed sample extended interest data can be filtered to obtain the filtered extended interest data. Optionally, the extensibility of the filtered extended interest data can be identified. Further, regardless of whether the filtered extended interest data has extensibility, the extensibility recognition result (the second extensibility recognition result) corresponding to the filtered extended interest data, the filtered extended interest data, the recognition result that the media resource has extensibility (the first extensibility recognition result), and the corresponding first resource characterization data can be input into the first preset large model for extended interest inference processing to obtain the corresponding preset extended inference data. Further, based on the first resource characterization data corresponding to the sample media resource and the corresponding preset extended inference data, the model to be trained can be subjected to extended interest inference training to obtain an extended interest inference model.

[0192] In step S205, based on the target extended interest data in the predicted extended inference data, the recommended media resource is determined.

[0193] In a specific embodiment, the recommended media resource may be a media resource recommended after the current media resource. Specifically, a media resource matching the target extended interest data may be selected from the media resources of the media resource platform as the recommended media resource; optionally, the media resource matching the target extended interest data may be the media resource with the highest semantic similarity to the target extended interest data; it may also be any media resource with a semantic similarity to the target extended interest data greater than or equal to a third preset similarity threshold.

[0194] In an alternative embodiment, the server may push the recommended media resource to the terminal corresponding to the browsing object when the browsing object of the current media resource triggers a media resource switching instruction.

[0195] In an alternative embodiment, the above-mentioned predicted extended inference data may be an extended inference data, or may include multiple first extended inference data; correspondingly, the above-mentioned inputting the current resource representation data into the extended interest inference model for extended interest inference processing to obtain the predicted extended inference data corresponding to the current media resource may include:

[0196] Inputting the current resource representation data into the extended interest inference model multiple times for extended interest inference processing to obtain multiple first extended inference data corresponding to the current media resource;

[0197] Inputting the multiple first extended inference data into the first quality analysis model for inference quality analysis to obtain the first quality index data corresponding to each of the multiple first extended inference data;

[0198] Correspondingly, the above-mentioned determining the recommended media resource based on the target extended interest data in the predicted extended inference data includes:

[0199] Determining the recommended media resource based on the target extended interest data in the target extended inference data.

[0200] In a specific embodiment, the multiple first extended inference data may be the extended interest data corresponding to the current media resource output by the extended interest inference model multiple times.

[0201] In a specific embodiment, the first quality analysis model may be a deep learning model for performing quality analysis of extended inference data. Specifically, the above-mentioned first quality index data represents the accuracy of the corresponding first extended inference data.

[0202] In a specific embodiment, the above-mentioned target extended inference data is the second extended inference data corresponding to the largest first quality index data among multiple first extended inference data. Specifically, for the specific refinement of determining the recommended media resources based on the target extended interest data in the target extended inference data, reference can be made to the above relevant description, which will not be elaborated here.

[0203] In the above embodiment, the current resource representation data is input into the extended interest inference model multiple times for extended interest inference processing to obtain multiple first extended inference data corresponding to the current media resource. Then, in combination with the first quality analysis model, the inference quality of the multiple first extended inference data is analyzed. Furthermore, the extended interest data in the extended inference data with the best quality can be selected by combining the first quality index data corresponding to each of the multiple first extended inference data, which can greatly improve the effectiveness of the extended interest data, better meet the user's browsing needs for continuous and relevant media resources during the media resource browsing process, and then improve the adaptability between the recommended media resources and the user's interests, and better meet the user's media resource browsing needs.

[0204] In addition, it should be noted that the model structure of the deep learning model in the embodiments of the present application can be set in combination with actual application requirements.

[0205] In a specific embodiment, assume that the current media resource is a video introducing a specific mechanical device, such as Figure 5 shown, Figure 5 FIG. is a schematic diagram showing the process of inferring the target extended interest data corresponding to the current media resource based on the extended interest inference model according to an exemplary embodiment; specifically, after the current resource representation data corresponding to the current media resource is input into the extended interest inference model, the extended interest inference model can combine the current resource representation data, extract the key information 501, and combine the content 502 that the resource browsing object (user account) inferred based on the key information 501 may be interested in. Based on the extension from the key information to the interested content, it is judged whether the video has continuity and relevance, and in combination with the previous inference process, the target extended interest data 503 is inferred.

[0206] According to the technical solution provided by the embodiments of the present application above, in the process of media resource recommendation, the embodiments of the present application obtain the current resource characterization data corresponding to the currently browsed media resource, and input the current resource characterization data into the extended interest inference model for extended interest inference processing, obtaining the inference process data (predicted extended inference data) of the target extended interest data corresponding to the current media resource inferred based on the current resource characterization data, enabling the model to infer, through coherent logic, the data (target extended interest data) for interest extension of the current media resource from content relevance and / or content continuity. This can perform interest extension more pertinently, and based on the target extended interest data, determine the media resources to be recommended after the current media resource. On the basis of meeting the user's continuous consumption needs through interest extension based on content continuity, it can bring freshness and incremental content to the user by combining interest extension based on content relevance, enhancing the innovation of the recommended media resources, thereby improving the quality and effect of media resource recommendation, more effectively meeting the user's media resource browsing needs, and reducing the waste of system resources caused by ineffective media resource recommendations.

[0207] Figure 6 is a flowchart of another media resource recommendation method shown according to an exemplary embodiment. This method can be applied to a terminal, such as Figure 6 shown, and this method may include the following steps:

[0208] In step S601, display the current media resource on a preset page;

[0209] In step S603, in response to a media resource switching instruction, display the recommended media resources associated with the target extended interest data corresponding to the current media resource on the preset page;

[0210] In a specific embodiment, the target extended interest data is the data for interest extension of the current media resource from content relevance and / or content continuity; and the target extended interest data is included in the predicted extended inference data corresponding to the current media resource. The predicted extended inference data is obtained by the extended interest inference model performing extended interest inference based on the current resource characterization data corresponding to the current media resource, and the predicted extended inference data is the inference process data for inferring the target extended interest data based on the current resource characterization data.

[0211] In a specific embodiment, for the specific refinement of determining the recommended media resources, reference can be made to the relevant content above, and details will not be elaborated here.

[0212] As can be seen from the technical solutions provided in the embodiments of this specification above, in the process of media resource recommendation in this specification, the reasoning process data (predicted extended reasoning data) for inferring the target extended interest data corresponding to the current media resource based on the current resource representation data corresponding to the currently browsed media resource is combined with the extended interest reasoning model, enabling the model to infer, through coherent logical reasoning, the data (target extended interest data) for interest extension of the current media resource in terms of content relevance and / or content continuity. This allows for more targeted interest extension, and the media resources that match the target extended interest data are used as the media resources to be recommended after the current media resource. On the basis of meeting the user's continuous consumption needs through interest extension based on content continuity, interest extension based on content relevance can bring freshness and incremental content to the user, enhancing the innovation of the recommended media resources, thereby improving the quality and effect of media resource recommendation, more effectively meeting the user's media resource browsing needs, and reducing the waste of system resources caused by ineffective media resource recommendations.

[0213] Figure 7 is a block diagram of a media resource recommendation device shown according to an exemplary embodiment. Referring to Figure 7 , the device includes:

[0214] A first representation data acquisition module 710, configured to acquire the current resource representation data corresponding to the current media resource;

[0215] A first extended interest reasoning module 720, configured to input the current resource representation data into the extended interest reasoning model for extended interest reasoning processing to obtain the predicted extended reasoning data corresponding to the current media resource. The predicted extended reasoning data is the reasoning process data for inferring the target extended interest data corresponding to the current media resource based on the current resource representation data; the target extended interest data is the data for interest extension of the current media resource in terms of content relevance and / or content continuity;

[0216] A recommended media resource determination module 730, configured to determine the recommended media resource based on the target extended interest data in the predicted extended reasoning data. The recommended media resource is the media resource to be recommended after the current media resource.

[0217] In an optional embodiment, the predicted extended reasoning data includes multiple first extended reasoning data; the first extended interest reasoning module 720 includes:

[0218] A first extended interest reasoning unit, configured to input the current resource representation data into the extended interest reasoning model for extended interest reasoning processing multiple times to obtain multiple first extended reasoning data corresponding to the current media resource;

[0219] An inference quality analysis unit, configured to perform inference quality analysis on multiple first extended inference data by inputting them into a first quality analysis model, to obtain first quality index data corresponding to each of the multiple first extended inference data, where the first quality index data characterizes the accuracy of the corresponding first extended inference data;

[0220] The recommended media resource determination module 730 is further configured to perform determining a recommended media resource based on the target extended interest data in the target extended inference data, where the target extended inference data is the second extended inference data with the largest first quality index data among the multiple first extended inference data.

[0221] In an optional embodiment, the extended interest inference model is trained using the following modules:

[0222] A second characterization data acquisition module, configured to perform acquiring first resource characterization data corresponding to a sample media resource;

[0223] A first extensibility recognition module, configured to perform extensibility recognition on the sample media resource based on the first resource characterization data, to obtain a first extensibility recognition result of the sample media resource;

[0224] An extended interest data acquisition module, configured to perform acquiring at least one sample extended interest data corresponding to the target media resource in the case where the first extensibility recognition result of the target media resource indicates that the target media resource has extensibility, where the target media resource is any one of the sample media resources; the at least one sample extended interest data is data obtained by extending the interest in the target media resource from the aspects of content relevance and / or content continuity;

[0225] A second extended interest inference module, configured to perform extended interest inference processing on the first resource characterization data and the at least one sample extended interest data by inputting them into a first preset large model, to obtain preset extended inference data corresponding to the sample media resource, where the preset extended inference data is the inference process data for inferring the extended interest data corresponding to the sample media resource based on the first resource characterization data;

[0226] A model training module, configured to perform extended interest inference training on the model to be trained based on the first resource characterization data and the preset extended inference data, to obtain an extended interest inference model.

[0227] In an optional embodiment, the second extended interest inference module includes:

[0228] A second extended interest inference unit, configured to perform extended interest inference processing on the first resource characterization data, the at least one sample extended interest data, and the first extensibility recognition result by inputting them into a first preset large model, to obtain preset extended inference data corresponding to the sample media resource.

[0229] In an alternative embodiment, the above device further includes:

[0230] A second ductility recognition module, configured to perform ductility recognition on at least one sample extended interest data corresponding to the target media resource to obtain a second ductility recognition result of the at least one sample extended interest data when the first ductility recognition result of the target media resource indicates that the target media resource has ductility;

[0231] The second extended interest inference unit includes:

[0232] A third extended interest inference unit, configured to perform extended interest inference processing on the first ductility recognition result, the second ductility recognition result, the first resource characterization data, and at least one sample extended interest data by inputting them into a first preset large model to obtain preset extended inference data corresponding to the sample media resource;

[0233] Wherein, the preset extended inference data corresponding to the first media resource includes at least one sample extended interest data corresponding to the first media resource; the preset extended inference data corresponding to the second media resource includes the sample extended interest data with ductility among at least one sample extended interest data corresponding to the second media resource, the first media resource is the media resource in the target media resource where all the corresponding at least one sample extended interest data has ductility; the second media resource is the media resource in the target media resource where there is sample extended interest data that does not have ductility.

[0234] In an alternative embodiment, the second ductility recognition module is specifically configured to perform ductility recognition on at least one sample extended interest data by inputting the at least one sample extended interest data into a second preset large model to obtain a second ductility recognition result.

[0235] In an alternative embodiment, the at least one sample extended interest data includes at least one of first extended interest data, second extended interest data, and custom third extended interest data; the first extended interest data is obtained by using the following module:

[0236] An interest extension processing module, configured to perform interest extension processing on the first resource characterization data by inputting it into a third preset large model to obtain first extended interest data;

[0237] The second extended interest data is obtained by using the following module:

[0238] A search operation data acquisition module, configured to acquire search operation data of at least one historical browsing object during the process of browsing the sample media resource, and the search operation data includes at least one keyword searched by the at least one historical browsing object during the process of browsing the sample media resource;

[0239] A search times determination module, configured to determine the search times corresponding to at least one keyword;

[0240] An extended interest data determination module, configured to use keywords with search times greater than or equal to a preset threshold as second extended interest data.

[0241] In an optional embodiment, the above device further includes:

[0242] A filtering processing module, configured to perform filtering processing on at least one sample extended interest data based on at least one preset filtering dimension to obtain filtered extended interest data;

[0243] The second extended interest inference module is further configured to input the first resource representation data and the filtered extended interest data into a first preset large model for extended interest inference processing to obtain preset extended inference data corresponding to the sample media resource.

[0244] In an optional embodiment, the first extensibility recognition module is specifically configured to input the first resource representation data into a fourth preset large model to perform extensibility recognition on the sample media resource to obtain a first extensibility recognition result.

[0245] In an optional embodiment, the above device further includes:

[0246] A third representation data acquisition module, configured to, after performing extended interest inference training on a to-be-trained model based on the first resource representation data and the preset extended inference data to obtain an extended interest inference model, acquire second resource representation data corresponding to a test media resource;

[0247] A third extended interest inference module, configured to input the second resource representation data into the extended interest inference model multiple times for extended interest inference processing to obtain multiple second extended inference data corresponding to each media resource in the test media resource;

[0248] An inference quality analysis module, configured to input the multiple second extended inference data into a second quality analysis model for inference quality analysis to obtain second quality index data corresponding to each of the multiple second extended inference data, where the second quality index data represents the accuracy of the corresponding extended inference data;

[0249] A positive and negative sample determination module, configured to use the extended inference data with the largest second quality index data among the multiple second extended inference data corresponding to each media resource as the positive sample data corresponding to each media resource, and use the extended inference data with the smallest second quality index data among the multiple second extended inference data corresponding to each media resource as the negative sample data corresponding to each media resource;

[0250] A model adjustment module, configured to perform adjustment training on the extended interest inference model based on second resource characterization data, positive sample data, and negative sample data.

[0251] Regarding the device in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0252] Figure 8 It is a block diagram of another media resource recommendation device shown according to an exemplary embodiment. Referring to Figure 8 , the device includes:

[0253] A current media resource display module 810, configured to perform displaying the current media resource on a preset page;

[0254] A recommended media resource display module 820, configured to perform displaying, in response to a media resource switching instruction, recommended media resources associated with target extended interest data corresponding to the current media resource on the preset page;

[0255] Wherein, the target extended interest data is data obtained by extending the interest in the current media resource in terms of content relevance and / or content continuity; and the target extended interest data is included in the predicted extended inference data corresponding to the current media resource, and the predicted extended inference data is obtained by the extended interest inference model performing extended interest inference based on the current resource characterization data corresponding to the current media resource, and the predicted extended inference data is the inference process data for inferring the target extended interest data based on the current resource characterization data.

[0256] Regarding the device in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0257] Figure 9 It is a block diagram of an electronic device for media resource recommendation shown according to an exemplary embodiment. The electronic device may be a terminal, and its internal structure diagram may be as Figure 9As shown. The electronic device includes a processor, a memory, a network interface, a display screen, and an input device connected via a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements a media resource recommendation method. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen. The input device of the electronic device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the electronic device, or an external keyboard, a touchpad, or a mouse, etc.

[0258] Figure 10 is a block diagram of another electronic device for media resource recommendation shown according to an exemplary embodiment. The electronic device can be a server, and its internal structure diagram can be as Figure 10 shown. The electronic device includes a processor, a memory, and a network interface connected via a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements a media resource recommendation method.

[0259] Those skilled in the art can understand that Figure 9 or Figure 10 the structures shown in are only block diagrams of some structures related to the solution of the present disclosure, and do not constitute a limitation on the electronic devices to which the solution of the present disclosure is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0260] In an exemplary embodiment, there is also provided an electronic device, including: a processor; a memory for storing executable instructions of the processor; wherein, the processor is configured to execute the instructions to implement the media resource recommendation method as in the embodiments of the present disclosure.

[0261] In an exemplary embodiment, there is also provided a computer-readable storage medium. When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device can execute the media resource recommendation method in the embodiments of the present disclosure.

[0262] In an exemplary embodiment, a computer program product including instructions is also provided. When it runs on a computer, it causes the computer to execute the media resource recommendation method in the embodiments of the present disclosure.

[0263] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0264] After considering the specification and practicing the invention disclosed herein, those skilled in the art will readily conceive of other embodiments of the present disclosure. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include known common knowledge or conventional technical means in the technical field not disclosed in the present disclosure. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of the present disclosure are pointed out by the following claims.

[0265] It should be understood that the present disclosure is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is only limited by the appended claims.

Claims

1. A media resource recommendation method, characterized in that, Including: Obtain the current resource characterization data corresponding to the current media resource; Input the current resource characterization data into an extended interest inference model for extended interest inference processing to obtain the predicted extended inference data corresponding to the current media resource. The predicted extended inference data is the inference process data for inferring the target extended interest data corresponding to the current media resource based on the current resource characterization data. The target extended interest data is data for extending the interest in the current media resource from content relevance and / or content continuity; Based on the target extended interest data in the predicted extended inference data, determine the recommended media resource. The recommended media resource is the media resource recommended after the current media resource.

2. The media resource recommendation method according to claim 1, wherein The predicted extended inference data includes multiple first extended inference data. The step of inputting the current resource characterization data into an extended interest inference model for extended interest inference processing to obtain the predicted extended inference data corresponding to the current media resource includes: Input the current resource characterization data into the extended interest inference model multiple times for extended interest inference processing to obtain the multiple first extended inference data corresponding to the current media resource; Input the multiple first extended inference data into a first quality analysis model for inference quality analysis to obtain the first quality index data corresponding to each of the multiple first extended inference data. The first quality index data characterizes the accuracy of the corresponding first extended inference data; The step of determining the recommended media resource based on the target extended interest data in the predicted extended inference data includes: Based on the target extended interest data in the target extended inference data, determine the recommended media resource. The target extended inference data is the second extended inference data with the largest first quality index data among the multiple first extended inference data.

3. The media resource recommendation method according to any one of claims 1 or 2, characterized in that, The extended interest inference model is trained in the following manner: Obtain the first resource characterization data corresponding to the sample media resource; Based on the first resource characterization data, perform ductility identification on the sample media resource to obtain the first ductility identification result of the sample media resource; When the first ductility identification result of the target media resource indicates that the target media resource has ductility, obtain at least one sample extended interest data corresponding to the target media resource. The target media resource is any one of the sample media resources. The at least one sample extended interest data is data for extending the interest in the target media resource from content relevance and / or content continuity; Input the first resource characterization data and the at least one sample extended interest data into a first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource. The preset extended inference data is the inference process data for inferring the extended interest data corresponding to the sample media resource based on the first resource characterization data; Based on the first resource characterization data and the preset extended inference data, perform extended interest inference training on the model to be trained to obtain the extended interest inference model.

4. The media resource recommendation method according to claim 3, wherein Inputting the first resource characterization data and the at least one sample extended interest data into a first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource includes: Inputting the first resource characterization data, the at least one sample extended interest data, and the first extensibility recognition result into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource.

5. The media resource recommendation method according to claim 4, wherein The method further includes: When the first extensibility recognition result of the target media resource indicates that the target media resource has extensibility, performing extensibility recognition on the at least one sample extended interest data corresponding to the target media resource to obtain a second extensibility recognition result of the at least one sample extended interest data; The inputting the first resource characterization data, the at least one sample extended interest data, and the first extensibility recognition result into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource includes: Inputting the first extensibility recognition result, the second extensibility recognition result, the first resource characterization data, and the at least one sample extended interest data into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource; Among them, the preset extended inference data corresponding to the first media resource includes the at least one sample extended interest data corresponding to the first media resource; the preset extended inference data corresponding to the second media resource includes the sample extended interest data with extensibility among the at least one sample extended interest data corresponding to the second media resource, the first media resource is the media resource in the target media resource where all the at least one sample extended interest data corresponding thereto have extensibility; the second media resource is the media resource in the target media resource where there is sample extended interest data that does not have extensibility.

6. The media resource recommendation method according to claim 5, wherein The performing extensibility recognition on the at least one sample extended interest data corresponding to the target media resource to obtain a second extensibility recognition result of the at least one sample extended interest data includes: Inputting the at least one sample extended interest data into a second preset large model to perform extensibility recognition on the at least one sample extended interest data to obtain the second extensibility recognition result.

7. The media resource recommendation method according to claim 3, wherein The at least one sample extended interest data includes at least one of: first extended interest data, second extended interest data, and custom third extended interest data; the first extended interest data is obtained by the following method: Inputting the first resource characterization data into a third preset large model for interest extension processing to obtain the first extended interest data; The second extended interest data is obtained by the following method: Obtaining search operation data of at least one historical browsing object during the process of browsing the sample media resource, the search operation data including at least one keyword searched by the at least one historical browsing object during the process of browsing the sample media resource; Determining the search times corresponding to the at least one keyword; Keywords with the number of searches greater than or equal to a preset threshold are used as the second extended interest data.

8. The media resource recommendation method according to claim 3, wherein The method further includes: Based on at least one preset filtering dimension, filtering the at least one sample extended interest data to obtain filtered extended interest data; The step of inputting the first resource representation data and the at least one sample extended interest data into a first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource includes: Inputting the first resource representation data and the filtered extended interest data into the first preset large model for extended interest inference processing to obtain the preset extended inference data corresponding to the sample media resource.

9. The media resource recommendation method according to claim 3, wherein The step of performing extensibility recognition on the sample media resource based on the first resource representation data to obtain the first extensibility recognition result of the sample media resource includes: Inputting the first resource representation data into a fourth preset large model to perform extensibility recognition on the sample media resource to obtain the first extensibility recognition result.

10. The media resource recommendation method according to claim 3, wherein After training the model to be trained for extended interest inference based on the first resource representation data and the preset extended inference data to obtain an extended interest inference model, the method further includes: Obtaining second resource representation data corresponding to the test media resource; Inputting the second resource representation data into the extended interest inference model multiple times for extended interest inference processing to obtain multiple second extended inference data corresponding to each media resource in the test media resource; Inputting the multiple second extended inference data into a second quality analysis model for inference quality analysis to obtain second quality index data corresponding to each of the multiple second extended inference data, where the second quality index data characterizes the accuracy of the corresponding extended inference data; Taking the extended inference data with the largest second quality index data among the multiple second extended inference data corresponding to each media resource as the positive sample data corresponding to each media resource, and taking the extended inference data with the smallest second quality index data among the multiple second extended inference data corresponding to each media resource as the negative sample data corresponding to each media resource; Adjusting and training the extended interest inference model based on the second resource representation data, the positive sample data, and the negative sample data.

11. A media resource recommendation method, characterized in that, Including: Displaying the current media resource on a preset page; In response to a media resource switching instruction, displaying recommended media resources associated with target extended interest data corresponding to the current media resource on the preset page; Wherein, the target extended interest data is data obtained by extending the interest in the current media resource in terms of content relevance and / or content continuity; and the target extended interest data is included in the predicted extended inference data corresponding to the current media resource, and the predicted extended inference data is obtained by the extended interest inference model through extended interest inference based on the current resource representation data corresponding to the current media resource, and the predicted extended inference data is the inference process data for inferring the target extended interest data based on the current resource representation data.

12. A media resource recommendation device, characterized in that, Including: The first characterization data acquisition module is configured to acquire current resource characterization data corresponding to the current media resource; The first extended interest inference module is configured to input the current resource characterization data into an extended interest inference model for extended interest inference processing to obtain predicted extended inference data corresponding to the current media resource. The predicted extended inference data is the inference process data for inferring the target extended interest data corresponding to the current media resource based on the current resource characterization data; the target extended interest data is data for extending the interest in the current media resource from the aspects of content relevance and / or content continuity. The recommended media resource determination module is configured to determine a recommended media resource based on the target extended interest data in the predicted extended inference data. The recommended media resource is a media resource recommended after the current media resource.

13. A media resource recommendation device, characterized in that, Comprising: The current media resource display module is configured to display the current media resource on a preset page; The recommended media resource display module is configured to, in response to a media resource switching instruction, display a recommended media resource associated with the target extended interest data corresponding to the current media resource on the preset page; Wherein, the target extended interest data is data for extending the interest in the current media resource from the aspects of content relevance and / or content continuity; and the target extended interest data is included in the predicted extended inference data corresponding to the current media resource. The predicted extended inference data is obtained by the extended interest inference model performing extended interest inference based on the current resource characterization data corresponding to the current media resource. The predicted extended inference data is the inference process data for inferring the target extended interest data based on the current resource characterization data.

14. An electronic device, characterized in that, Comprising: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to execute the instructions to implement the media resource recommendation method according to any one of claims 1 to 11.

15. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is enabled to execute the media resource recommendation method according to any one of claims 1 to 11.

Images