Material recommendation method and device

Abstract

The invention discloses a material recommendation method and device, and belongs to the technical field of artificial intelligence. The material recommendation method comprises the steps of obtaining first feature data in an application program, wherein the application program comprises a to-be-recommended material; obtaining a first material list, wherein the first material list is obtained through first arrangement processing; determining second feature data according to the first feature data and the first material list; and generating a second material list through the first model according to the second feature data and the prompt template corresponding to the material type.

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and specifically relates to a material recommendation method and device thereof. Background Art

[0002] The recommendation system mainly includes recall, rough sorting, fine sorting, re-ranking, front-end display and other links. In the recommendation system, the sorting results of fine sorting are often concentrated and must be scattered after re-ranking before they can be recommended to users. Re-ranking is the last link between users and the recommendation system, and plays an immeasurable role in the recommendation system. From the system level, on the one hand, re-ranking can improve the accuracy of recommendation results; on the other hand, re-ranking increases the diversity of recommendation results and avoids the monotony of recommended content.

[0003] In order to ensure the accuracy of the recommendation results, the re-ranking needs to readjust the position and layout of the recommendation list output by the refined ranking, which inevitably requires a large amount of computing resources and data storage resources. At present, the commonly used re-ranking methods in the industry mainly include the re-ranking method based on rule-based scattering and the re-ranking method based on DPP (Determinantal Point Process). Among them, the re-ranking method based on rule-based scattering is a relatively hard-regular random scattering method. Its advantage is that it is time-consuming, and its disadvantage is that it does not take into account the correlation between contexts and is more dependent on the accuracy of the rules. The re-ranking method based on DPP is a greedy algorithm, which solves the pain point that random scattering cannot measure correlation and diversity, but it can only perform pairwise comparisons each time, and cannot take into account the overall diversity of the recommendation list. It is also more dependent on the accuracy of the embedded representation of each material.

[0004] On this basis, material recommendations based on existing rearrangement methods will have problems with poor accuracy and diversity of the rearrangement results, thereby reducing the accuracy and diversity of the recommendation results and making it difficult to meet user needs. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a material recommendation method and device thereof, which can improve the accuracy and diversity of material recommendations.

[0006] In a first aspect, an embodiment of the present application provides a material recommendation method, the method comprising: obtaining first characteristic data in an application, the application including materials to be recommended; obtaining a first material list, the first material list being obtained through a first arrangement process; determining second characteristic data based on the first characteristic data and the first material list; generating a second material list through a first model according to a prompt template corresponding to the second characteristic data and the material type.

[0007] In the second aspect, an embodiment of the present application provides a material recommendation device, which includes: an acquisition unit, used to acquire first characteristic data in an application, the application including materials to be recommended; the acquisition unit, also used to acquire a first material list, the first material list being obtained through a first arrangement process; a processing unit, used to determine second characteristic data based on the first characteristic data and the first material list; the processing unit is also used to generate a second material list through a first model according to a prompt template corresponding to the second characteristic data and the material type.

[0008] In a third aspect, an embodiment of the present application provides an electronic device, which includes a processor and a memory, wherein the memory stores programs or instructions that can be run on the processor, and when the programs or instructions are executed by the processor, the steps of the material recommendation method of the first aspect are implemented.

[0009] In a fourth aspect, an embodiment of the present application provides a readable storage medium having a program or instruction stored thereon, which, when executed by a processor, implements the steps of the material recommendation method of the first aspect.

[0010] In a fifth aspect, an embodiment of the present application provides a chip, which includes a processor and a communication interface, wherein the communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the steps of the material recommendation method of the first aspect.

[0011] In a sixth aspect, an embodiment of the present application provides a computer program product, which is stored in a storage medium and is executed by at least one processor to implement the steps of the material recommendation method of the first aspect.

[0012] In the material recommendation method provided in the embodiment of the present application, the first characteristic data in the application is obtained, and the application includes materials to be recommended; the first material list is obtained, and the first material list is obtained through the first arrangement process; the second characteristic data is determined according to the first characteristic data and the first material list; and the second material list is generated according to the second characteristic data and the prompt template corresponding to the material type through the first model. Through the above-mentioned material recommendation method, the second characteristic data is generated based on the first characteristic data of the material recommendation and the first material list after the first arrangement process, and then the second material list is generated according to the second characteristic data and the prompt template corresponding to the material type through the first model. In this way, the rearrangement task is converted into a sequence generation task of the first model, and by introducing the prompt template corresponding to the material type, a material recommendation list is generated in the case of zero samples, which can not only improve the accuracy and efficiency of material recommendation, but also ensure the diversity of recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1One of the flow charts of the material recommendation method provided in the embodiment of the present application;

[0014] Figure 2 The second flow chart of the material recommendation method provided in the embodiment of the present application;

[0015] Figure 3 An example diagram of calculating the first similarity provided in an embodiment of the present application;

[0016] Figure 4 A structural block diagram of a material recommendation device provided in an embodiment of the present application;

[0017] Figure 5 A structural block diagram of an electronic device provided in an embodiment of the present application;

[0018] Figure 6 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0020] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way are interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0021] The material recommendation method provided in the embodiment of the present application is described in detail below through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0022] like Figure 1 As shown, the embodiment of the present application provides a material recommendation method, which may include the following S102 to S108:

[0023] S102: Acquire first characteristic data in the application.

[0024] The material recommendation method proposed in the embodiment of the present application is executed by an electronic device, which may specifically be a smart electronic device such as a smart phone, a tablet computer, a laptop computer, and a smart watch, and is not specifically limited here.

[0025] Among them, the application includes the materials to be recommended.

[0026] In the material recommendation method proposed in the embodiment of the present application, the material to be recommended may be videos, news, images, wallpapers, product information, application installation packages, games, music, articles, and takeaway information, etc., without specific limitation herein.

[0027] On this basis, the above-mentioned applications include but are not limited to: video applications, news applications, wallpaper stores, shopping applications, application stores, music applications, reading applications, and food delivery applications, etc., and no specific restrictions are made here.

[0028] Furthermore, the above-mentioned first characteristic data may specifically include but is not limited to at least one of the following: user basic data, user behavior statistical data, material basic data, material behavior statistical data, user interaction data on materials, and user feedback data on materials.

[0029] In an optional embodiment, the first characteristic data may be specifically as shown in the following Table 1:

[0030] Table 1: Example table of the first characteristic data

[0031]

[0032]

[0033] Among them, CTR (Click-Through Rate) is used to measure whether the user is interested in the recommended material. CTR = (clicks / exposures) × 100%, where clicks represents the total number of clicks on the last recommended material by the user, and exposures represents the total number of exposures of the last recommended material.

[0034] Specifically, in the material recommendation method proposed in the embodiment of the present application, based on the basic information of users and materials in the application and the historical interaction information between users and materials, the user basic data, user behavior statistics, material basic data, material behavior statistics, user interaction data on materials, and user feedback data on materials in the application are obtained, thereby generating the first feature data required for recommending materials offline.

[0035] In an optional embodiment, the first characteristic data is not limited to the above content, and different applications can add more first characteristic data when recommending items according to different recommendation scenarios. For example, the user's shopping behavior for buying clothes has seasonal factors, and seasonal related features can be added when recommending clothes for the user.

[0036] S104: Obtain a first material list.

[0037] The first material list is obtained through the first arrangement process.

[0038] The first arrangement process is a refined arrangement task in the recommendation system, and the first material list is the output result of the refined arrangement module in the recommendation system.

[0039] It is understandable that the recommendation chain of the recommendation system mainly includes recall, rough sorting, fine sorting and re-ranking, and its goal is to select dozens of materials from a large number of materials and recommend them to users. Among them, recall refers to using dozens of recall channels to retrieve dozens to hundreds of materials through each channel according to user characteristics (such as historical behavior, interest preferences, etc.), quickly retrieve thousands of materials that users are potentially interested in from a large number of materials, and remove duplicates and filter the retrieved materials. Rough sorting refers to using a small-scale neural network to score and sort the thousands of materials obtained by recall, and select the materials with the top hundreds of scores. Fine sorting refers to using a large-scale neural network and more features to re-score and sort the hundreds of materials selected by rough sorting. The fine sorting score is more reliable than the rough sorting score, and the fine sorting score of the material can reflect the user's interest in the material. Re-ranking refers to breaking up the fine sorting results, evaluating the diversity and re-ranking, and selecting the top dozens of materials to recommend to users, so as to improve the accuracy and diversity of the recommendation results.

[0040] Specifically, in the material recommendation method proposed in the embodiment of the present application, in the process of recommending materials to users, based on the above-mentioned first feature data, the recall module, the rough sorting module and the fine sorting module of the recommendation system are called in sequence to obtain the output result of the upstream fine sorting task, that is, the first material list, as the data basis for material recommendation.

[0041] In an optional embodiment, before obtaining the first material list, it is also possible to determine whether the user triggers the material recommendation process. If it is determined that the user triggers the material recommendation process, the first material list is obtained and subsequent processing is performed.

[0042] Specifically, based on the browsing behavior of the user in the application, it is determined whether the user triggers the material recommendation process. In the case where the exposure rate of the last recommended multiple materials in the application is greater than a preset threshold such as 50%, it is determined that the user triggers a new material recommendation process.

[0043] For example, a wallpaper application recommends 10 wallpapers to a user each time. When the user is exposed to the first 5 wallpapers recommended last time, the wallpaper application will trigger a new round of wallpaper recommendation process and determine 10 new wallpapers to recommend to the user.

[0044] Among them, the judgment basis for the user to trigger the material recommendation process can be set by technical personnel in this field according to actual conditions, and no specific limitation is made here.

[0045] S106: Determine second characteristic data according to the first characteristic data and the first material list.

[0046] The second characteristic data is complete text information obtained by combining and arranging the first characteristic data and the first material list.

[0047] Specifically, in the material recommendation method proposed in the embodiment of the present application, after obtaining the first characteristic data and the first material list, the first characteristic data and the first material list are combined to obtain the second characteristic data required for material recommendation.

[0048] S108: Generate a second material list through the first model according to the second characteristic data and the prompt template corresponding to the material type.

[0049] Among them, material types include but are not limited to: video, news, image, wallpaper, commodity, application, game, music, article and takeaway, etc., and no specific restrictions are made here.

[0050] Furthermore, the first model can specifically be LLMs (Large Language Models), which have the capabilities of instruction following, zero-shot rearrangement, and contextual learning. Among them, instruction following refers to the ability of LLMs to perform corresponding tasks according to user requirements or instructions. Based on the instruction following ability, LLMs can understand and generate some complex operations, such as answering questions and generating text. Zero-shot rearrangement refers to the ability of LLMs to directly understand and perform rearrangement tasks without special training for specific tasks. Contextual learning refers to the ability of LLMs to learn and generate text based on a given context. Based on the contextual learning ability, LLMs can respond to user needs more accurately in specific scenarios.

[0051] Furthermore, the prompt template is Prompt, and the prompt template is used to be input to the first model to require the first model to output text content according to a specified prompt and a specified format.

[0052] It is understandable that different types of material recommendations have different display requirements for recommendation results. For example, for material recommendations with strong timeliness, such as news recommendations, it is necessary to recommend more recommended content related to user interests to users; for entertainment and leisure material recommendations, such as video recommendations, it is necessary to balance the relevance and diversity of the recommended content. Therefore, in the material recommendation method provided in the embodiment of the present application, corresponding prompt templates are designed for different types of materials, and different prompt templates have different requirements for the relevance, diversity, and discoverability of recommended materials.

[0053] Among them, relevance refers to the degree of association between two or more things. In this application, relevance refers to the degree of association between recommended materials and user interests. Diversity refers to the diversity of material recommendation results, which is usually measured by the richness of the recommended material types. Discoverability refers to the ability to further explore and discover new things under the premise of satisfying the relevance and diversity of recommended materials.

[0054] Specifically, in the material recommendation method provided in the embodiment of the present application, as shown in Table 2 below, the above-mentioned prompt templates may specifically include a relevance priority prompt template, a diversity priority prompt template, a relevance and diversity balance prompt template, and a discoverability priority prompt template.

[0055] Table 2: Example of prompt template table

[0056] Prompt Template Recommended conditions Relevance Priority Tips Template Maximize the relevance of recommended materials Diversity Priority Tips Template Maximize the diversity of recommended materials Relevance and Diversity Balance Tips Template Balance the relevance and diversity of recommended materials Discovery-first prompt template Recommend more new content related to user interests

[0057] Among them, different prompt templates correspond to different recommendation goals, namely, recommendation conditions. The recommendation condition of the relevance priority prompt template is to maximize the relevance of the recommended materials to meet the requirements of recommendation accuracy; the recommendation condition of the diversity priority prompt template is to maximize the diversity of the recommended materials to ensure the richness of the recommendation results and alleviate the information cocoon; the recommendation condition of the relevance and diversity balance prompt template is to balance the relevance and diversity of the recommended materials to take into account the recommendation accuracy and richness of the recommendation results; the recommendation condition of the discoverability priority prompt template is to recommend more new content related to user interests to improve the novelty and real-time nature of the recommendation results and tap into the user's potential interests. In this way, a variety of prompt templates corresponding to different types of materials are designed. According to different recommendation application scenarios, the zero-sample learning ability of the large language model can be used to adopt different strategies to re-rank and recommend materials, thereby improving the accuracy of the material recommendation results.

[0058] Specifically, in the material recommendation method provided in the embodiment of the present application, after obtaining the second feature data, based on the type of material to be recommended, the corresponding recommendation target, i.e., the recommendation condition, is determined, and then the corresponding prompt template is selected, and the second material list is generated through the first model according to the prompt template and the second feature data. In this way, on the one hand, there is no need to rely on a large number of samples for model training, and the instruction following ability, zero-sample rearrangement ability, and context learning ability of the large model are utilized to innovatively transform the rearrangement task into a text sequence generation task of the large model, so that a more diverse material recommendation list can be generated under zero-sample conditions. On the other hand, by introducing a feature awareness template through a prompt template corresponding to the material type, the large model can be helped to better understand the characteristics of each material to be recommended, and generate a more diverse material recommendation list, reducing the gap between the traditional recommendation method and the recommendation method based on the large model, and improving the accuracy and efficiency of the material recommendation, which not only solves the strong dependence of the rearrangement method based on rule-breaking on the specified rules, but also can coordinate the global recommendation list, solve the problem that the rearrangement method based on DPP cannot take into account the overall diversity of the recommendation list, and take into account the improvement of the accuracy and diversity of the material recommendation.

[0059] The above-mentioned material recommendation method provided in the embodiment of the present application obtains the first characteristic data in the application, and the application includes the materials to be recommended; obtains the first material list, and the first material list is obtained through the first arrangement process; determines the second characteristic data according to the first characteristic data and the first material list; generates the second material list according to the second characteristic data and the prompt template corresponding to the material type through the first model. Through the above-mentioned material recommendation method, the second characteristic data is generated based on the first characteristic data of the material recommendation and the first material list after the first arrangement process, and then the second material list is generated according to the second characteristic data and the prompt template corresponding to the material type through the first model. In this way, the rearrangement task is converted into a sequence generation task of the first model, and by introducing the prompt template corresponding to the material type, a material recommendation list is generated in the case of zero samples, which can not only improve the accuracy and efficiency of material recommendation, but also ensure the diversity of recommendation results.

[0060] In the embodiment of the present application, the first feature data includes at least one of user basic data, user behavior statistical data, material basic data, material behavior statistical data, and user interaction data on materials. The above S106 may specifically include the following S106a to S106d:

[0061] S106a: Determine first material information according to the first material list.

[0062] The first material information includes identification information of each material in the first material list and score information in the first arrangement process.

[0063] Specifically, in the material recommendation method provided in the embodiment of the present application, after obtaining the first material list, the first material list is converted into a dictionary in JSON (JavaScript Object Notation, a lightweight data exchange format) format to obtain the first material information.

[0064] Among them, the key of the dictionary in JSON format is the identification information of each material, such as the ID (IDentity, identity) information of each material, and the value is the score information of each material in the first arrangement process (fine arrangement task). An example of a dictionary in JSON format is as follows: {"pid1":"score1","pid2":"score2","pid3":"score3",...,"pidN":"scoreN"}, which means: the score of the material with ID pid1 in the fine arrangement task is score1; the score of the material with ID pid2 in the fine arrangement task is score2; the score of the material with ID pid3 in the fine arrangement task is score3; the score of the material with ID pidN in the fine arrangement task is scoreN.

[0065] S106b: Combine the user basic data and the user behavior statistical data to obtain first information.

[0066] Specifically, in the material recommendation method provided in the embodiment of the present application, the first characteristic data includes user basic data and user behavior statistical data. After obtaining the first characteristic data, the user basic data and user behavior statistical data in the first characteristic data are combined and organized into complete text information to obtain the first information.

[0067] Exemplarily, the first information may specifically be: the gender of user 10001 is male, the age is 23 years old, the city he belongs to is City A, and in the past 30 days, he has exposed 3,000 wallpapers, clicked 200 wallpapers, and liked 10 wallpapers.

[0068] S106c: Combine the identification information, score information, order information in the first material list, material basic data, material behavior statistical data and interaction data of each material in the first material list to obtain second information corresponding to each material.

[0069] Specifically, in the material recommendation method provided in the embodiment of the present application, the first characteristic data also includes basic material data, material behavior statistical data, and user interaction data on the material. After obtaining the first characteristic data and the first material list, and determining the first material information based on the first material list, the identification information, score information, order information in the first material list, basic material data, material behavior statistical data and interaction data of each material in the first material list are combined and organized into complete text information to obtain the second information corresponding to each material.

[0070] Exemplarily, the second information of material pid1 can be specifically as follows: the order of material pid1 in the fine ranking task is 1, the score is 0.98, the first-level classification is animation, the second-level classification is animated movie, the title is "Title 1", and the release time is 100 days; material pid1 has been exposed 2,000 times, clicked 100 times, and liked 10 times in the past 30 days, the number of clicks on animation classified materials in the past 30 days is 20,000 times, and the number of clicks on animation movie classified materials in the past 30 days is 800 times; users have clicked on animation classified materials 10 times in the past 30 days, and users have clicked on animation movie classified materials 2 times in the past 30 days.

[0071] Exemplarily, the second information of material pid2 can be specifically as follows: the order of material pid2 in the fine ranking task is 2, the score is 0.96, the first-level classification is pet, the second-level classification is dog, the title is "Title 2", and the release time is 2 days; material pid2 has been exposed 3,000 times, clicked 500 times, and liked 60 times in the past 30 days, pet classified materials have been clicked 50,000 times in the past 30 days, and dog classified materials have been clicked 800 times in the past 30 days; users have clicked on pet classified materials 15 times in the past 30 days, and users have clicked on dog classified materials 3 times in the past 30 days.

[0072] Exemplarily, the second information of material pidN can be specifically as follows: the order of material pidN in the fine ranking task is N, the score is 0.25, the first-level classification is people, the second-level classification is women, and the title is "Title N"; material pidN has been exposed 5,000 times, clicked 1,000 times, and liked 500 times in the past 30 days, the person classification material has been clicked 10,000 times in the past 30 days, and the women classification material has been clicked 2,000 times in the past 30 days; the user has clicked on the person classification material 10 times in the past 30 days, and the user has clicked on the women classification material once in the past 30 days.

[0073] S106d: Combine the first information and the second information corresponding to each material to obtain second characteristic data.

[0074] Specifically, in the material recommendation method provided in the embodiment of the present application, after obtaining the first information and the second information corresponding to each material, the first information and the second information corresponding to each material are combined into the second characteristic data required for recommending the material. In this way, based on the output result of the refined sorting task and the first characteristic data related to the user and the material, the second characteristic data required for material recommendation is generated, which ensures the accuracy of the second characteristic data, and further ensures the accuracy of subsequent material recommendation based on the second characteristic data.

[0075] In the above-mentioned embodiment provided by the present application, the first characteristic data includes at least one of user basic data, user behavior statistics, material basic data, material behavior statistics, and user interaction data on materials. In the process of determining the second characteristic data according to the first characteristic data and the first material list, the first material information is determined according to the first material list, and the first material information includes the identification information of each material in the first material list and the score information in the first arrangement process; the user basic data and the user behavior statistics are combined to obtain the first information; the identification information, score information, order information in the first material list, material basic data, material behavior statistics, and interaction data of each material in the first material list are combined to obtain the second information corresponding to each material; the first information and the second information corresponding to each material are combined to obtain the second characteristic data. In this way, based on the output result of the refined arrangement task and the first characteristic data related to the user and the material, the second characteristic data required for material recommendation is generated, which ensures the accuracy of the second characteristic data, and further ensures the accuracy of subsequent material recommendation based on the second characteristic data.

[0076] In the embodiment of the present application, the above S108 may specifically include the following S108a to S108c:

[0077] S108a: Determine the output target and output format according to the prompt template.

[0078] The output target is the output content of the first model, that is, the content of the second material list, which corresponds to the recommendation target corresponding to the prompt template, that is, the recommendation condition.

[0079] Furthermore, the output format is the format of the output content of the first model, that is, the content format of the second material list.

[0080] S108b: Construct a generation instruction according to the second characteristic data, the output target and the output format.

[0081] Among them, the generation instruction may specifically include the recommendation target corresponding to the prompt template, that is, the recommendation condition, the second feature data, the user's feedback data on the material in the first feature data, the output target of the first model, and the output format. In this way, when the first model is subsequently controlled to generate the second material list based on the generation instruction, by introducing the second feature data containing the first information and the second information, and introducing the feature awareness template, the first model can better understand the user and material information, reduce the gap between the traditional material recommendation method and the material recommendation method based on the large language model, thereby improving the accuracy and efficiency of material recommendation; and, by introducing the user's feedback data on the material, it is possible to achieve adaptive learning and adjustment of the material recommendation results, while ensuring the relevance and diversity of the material recommendation results, and adaptively adjust the material recommendation link, thereby improving the robustness of the material recommendation.

[0082] Specifically, in the material recommendation method provided in the embodiment of the present application, the output target and output format are determined according to the prompt template, and then the second characteristic data, the output target, the output format and the recommendation conditions corresponding to the prompt template are sorted and combined to construct a generation instruction.

[0083] S108c: Input the generation instruction into the first model, and generate a second material list conforming to the output format through the first model.

[0084] Specifically, in the material recommendation method provided in the embodiment of the present application, after the generation instruction is constructed, the generation instruction is input into the first model so that the first model performs the text sequence generation task according to the generation instruction and generates a second material list that conforms to the above output format. In this way, the instruction following ability, zero-sample rearrangement ability and context learning ability of the large model are utilized to innovatively transform the material rearrangement task into the text sequence generation task of the large model, and through a specific prompt template, the large model outputs the second material list according to a specific output target and output format, thereby ensuring the accuracy and diversity of material recommendations.

[0085] For example, the recommended material is wallpaper, and the prompt template uses a relevance priority prompt template. Assuming that the user ID is 10001, the first material list output in the refined sorting task stage includes 500 wallpapers. This time, the top 10 best wallpapers need to be recommended to the user. The specific content of the prompt template or the specific content of the generation instruction can be:

[0086] You are an excellent recommendation system rearrangement expert. You have information about user 10001 and 500 wallpapers that have been carefully sorted. Your task is to rearrange these 500 wallpapers. Your goal is to maximize the relevance of the material recommendation list. The information of the user and wallpaper is as follows: The user's information is: {User 10001 is male, 23 years old, and belongs to City A. In the past 30 days, he has viewed 3,000 wallpapers, clicked 200 wallpapers, and liked 10 wallpapers.}The recommendation system recommends 500 wallpapers to the user. The information of these wallpapers is as follows: {The order of wallpaper pid1 in the fine ranking task is 1, the score is 0.98, the first-level category is anime, the second-level category is animated movie, the title is "Title 1", and the release time is 100 days; wallpaper pid1 has been exposed 2000 times, clicked 100 times, and liked 10 times in the past 30 days, the number of clicks on anime classified wallpapers in the past 30 days is 20,000 times, and the number of clicks on anime movie classified wallpapers in the past 30 days is 800 times; the user has clicked on anime classified wallpapers 10 times in the past 30 days, and the user has clicked on anime movie classified wallpapers 2 times in the past 30 days. Wallpaper pid2 ranks 2 in the ranking task, with a score of 0.96. The first-level category is pets, the second-level category is dogs, the title is "Title 2", and the release time is 2 days. Wallpaper pid2 has been exposed 3,000 times, clicked 500 times, and liked 60 times in the past 30 days. Pets have been clicked 50,000 times, and dogs have been clicked 800 times in the past 30 days. Users have clicked on pets 15 times in the past 30 days, and users have clicked on dogs 3 times in the past 30 days. Information about wallpaper pid3 to wallpaper pid499. Wallpaper pid500 has an order of 500 in the fine sorting task, a score of 0.25, a first-level category of people, a second-level category of women, and a title of "Title 500"; wallpaper pid500 has been exposed 5,000 times, clicked 1,000 times, and liked 500 times in the past 30 days, people-classified wallpapers have been clicked 10,000 times in the past 30 days, and women-classified wallpapers have been clicked 2,000 times in the past 30 days; users have clicked on people-classified wallpapers 10 times in the past 30 days, and users have clicked on women-classified wallpapers once in the past 30 days.} In the last wallpaper recommendation, the user's CTR was 0.25. Next, based on the above information, re-sort the 500 wallpapers and output the best 10 wallpapers. Please note: The generated results must be output in the specified output format and cannot contain additional information. The output format is as follows: {pid[], the order in the refined sorting is [], and the order after re-arrangement is 1; pid[], the order in the refined sorting is [], and the order after re-arrangement is 2; pid[], the order in the refined sorting is [], and the order after re-arrangement is 3; ...; pid[], the order in the refined sorting is [], and the order after re-arrangement is 10}.

[0087] Among them, the user information is the above-mentioned first information, the wallpaper information is the above-mentioned second information, and the specific values ​​in [] are determined according to the actual output.

[0088] On this basis, the output results of the first model can be specifically as follows: {pid8, the order in the fine sorting is 8, and the order after re-arrangement is 1; pid52, the order in the fine sorting is 52, and the order after re-arrangement is 2; pid9, the order in the fine sorting is 9, and the order after re-arrangement is 3; ...; pid1, the order in the fine sorting is 1, and the order after re-arrangement is 10}.

[0089] In the above-mentioned embodiment provided by the present application, in the process of generating the second material list according to the prompt template corresponding to the second feature data and the material type through the first model, the output target and output format are determined according to the prompt template; the generation instruction is constructed according to the second feature data, the output target and the output format; the generation instruction is input into the first model, and the second material list conforming to the output format is generated through the first model. In this way, the instruction following ability, zero-sample rearrangement ability and context learning ability of the large model are utilized to innovatively transform the material rearrangement task into the text sequence generation task of the large model, and through a specific prompt template, the large model outputs the second material list according to the specific output target and output format, thereby ensuring the accuracy and diversity of material recommendations.

[0090] In the embodiment of the present application, after the above S108, the above material recommendation method may further include the following S110 to S114:

[0091] S110: Check the second material list.

[0092] Specifically, in the material recommendation method provided in the embodiment of the present application, after the first model generates the second material, the second material list will also be verified to determine whether the second material list meets the recommendation conditions corresponding to the prompt template.

[0093] S112: When the second material list does not meet the recommendation condition corresponding to the prompt template, regenerate the second material list through the first model until the generated second material list meets the recommendation condition or the number of times the second material list is generated reaches a preset number threshold.

[0094] Specifically, in the material recommendation method provided in the embodiment of the present application, when the second material list fails to pass the verification, that is, when the second material list does not meet the recommendation conditions corresponding to the prompt template, a new second material list continues to be regenerated through the first model until the generated second material list meets the recommendation conditions or the number of times the second material list is generated reaches a preset number threshold.

[0095] Among them, for the specific value of the above-mentioned preset number threshold, technical personnel in this field can set it according to actual conditions. For example, the above-mentioned preset number threshold can be 5, 6, 7, 8 and other values, and no specific limitation is made here.

[0096] S114: When the second material list meets the recommendation condition corresponding to the prompt template, display the multiple materials in the second material list.

[0097] Specifically, in the material recommendation method provided in the embodiment of the present application, when the second material list passes the verification, that is, when the second material list meets the recommendation conditions corresponding to the prompt template, multiple materials in the second material list are displayed to display the recommended multiple materials to the user at the front end. In this way, before displaying the material recommendation results to the user, the recommended materials are first verified to ensure the accuracy of the final material recommendation results.

[0098] In the above embodiment provided by the present application, after the second material list is generated, the second material list is verified; if the second material list does not meet the recommendation conditions corresponding to the prompt template, the second material list is regenerated through the first model until the generated second material list meets the recommendation conditions or the number of times the second material list is generated reaches a preset number threshold; if the second material list meets the recommendation conditions corresponding to the prompt template, multiple materials in the second material list are displayed. In this way, before displaying the material recommendation results to the user, the recommended materials are first verified to ensure the accuracy of the final material recommendation results.

[0099] In the embodiment of the present application, the prompt template includes a first prompt template, and the first prompt template corresponds to the first recommendation condition. The above S110 may specifically include the following S116 to S122:

[0100] S116: Embedding the user and each item in the second item list to obtain a first vector for the user and a second vector for each item in the second item list.

[0101] The prompt template includes a first prompt template, and the first prompt template corresponds to the first recommendation condition.

[0102] Furthermore, the first prompt template is a relevance-first prompt template, and the first recommendation condition is to maximize the relevance of the recommended material.

[0103] On this basis, in the material recommendation method provided in the embodiment of the present application, after obtaining the second material list based on the first prompt template, in the process of verifying the second material list, the embedding representation generation module of the recommendation system is called to obtain the embedding representation of the user and each material in the second material list, and obtain the first vector corresponding to the user and the second vector corresponding to each material in the second material list.

[0104] Among them, the embedding representation is a vector of fixed dimension, and the embedding representations of users and materials are in the same vector space dimension. The vector space dimension refers to the vector representation that converts text into a high-dimensional space.

[0105] Exemplarily, assuming that the dimension of the embedding representation is 10, the embedding representation of the user and the material can be: the embedding representation of the user, that is, the first vector is: [0.11, 0.22, 0.32, 0.21, 0.98, 4.21, 0.81, 0.35, 0.99, 1.25]; the embedding representation of the material pid1, that is, the second vector is: [0.23, 0.18, 0.52, 0.98, 0.18, 9.12, 0.92, 0.15, 0.52, 2.14].

[0106] S118: Calculate a first similarity between the user and each material based on the first vector and the second vector of each material.

[0107] Among them, the first similarity is the cosine similarity, which is between -1 and 1. The larger the cosine similarity between the user and the material, the stronger the correlation between the user's interest and the material.

[0108] Specifically, in the material recommendation method provided in the embodiment of the present application, after obtaining the first vector corresponding to the user and the second vector corresponding to each material in the second material list, as follows Figure 3 As shown, the first similarity between the user and each material is calculated based on the first vector and the second vector of each material.

[0109] Among them, the first vector corresponding to the user is represented by A, and the elements in A are [A1, A2, A3, A4, A5, A6, A7, A8, A9, A10], and the second vector corresponding to the material pidI (I = 1, 2, 3, 4, 5, 6, 7, 8, 9, 10) is represented by B, and the elements in B are [B1, B2, B3, B4, B5, B6, B7, B8, B9, B10]. The cosine similarity between the user and the material pidI, that is, the first similarity, can be calculated by the following formula:

[0110]

[0111] Among them, similarI represents the cosine similarity between the user and the material pidI, that is, the first similarity, A i represents the i-th element in vector A, B i represents the i-th element in vector B, n represents the vector dimension, n=10.

[0112] S120: Counting the number of first materials whose first similarity is greater than or equal to a first threshold.

[0113] Specifically, in the material recommendation method provided in the embodiment of the present application, after calculating the first similarity between the user and each material, the first similarity corresponding to each material in the second material table is compared with the first threshold, and the number of first materials whose first similarity is greater than or equal to the first threshold is counted.

[0114] Among them, the specific value of the above-mentioned first threshold can be set by those skilled in the art according to actual conditions. For example, the above-mentioned first threshold can be a value such as 0.85, and no specific limitation is made here.

[0115] S122: When a first ratio of the first material quantity to the total number of materials in the second material list is greater than a second threshold, it is determined that the second material list meets the first recommendation condition; when the first ratio is less than or equal to the second threshold, it is determined that the second material list does not meet the first recommendation condition.

[0116] Specifically, in the material recommendation method provided in the embodiment of the present application, when the first ratio of the first material quantity to the total number of materials in the second material list is greater than the second threshold, it is determined that the second material list meets the first recommendation condition, and when the first ratio is less than or equal to the second threshold, it is determined that the second material list does not meet the first recommendation condition. In this way, the corresponding verification method is configured for the relevance priority prompt template to ensure the accuracy of the verification of the second material list.

[0117] Among them, the specific value of the second threshold can be set by those skilled in the art according to actual conditions. For example, the second threshold can be 70% or other values, and no specific limitation is made here.

[0118] In the above-mentioned embodiment provided by the present application, the prompt template includes a first prompt template, the first prompt template corresponds to the first recommendation condition, and in the process of verifying the second material list, the user and each material in the second material list are embedded and represented to obtain the first vector of the user and the second vector of each material in the second material list; the first similarity between the user and each material is calculated according to the first vector and the second vector of each material; the number of first materials whose first similarity is greater than or equal to the first threshold is counted; when the first ratio of the number of first materials to the total number of materials in the second material list is greater than the second threshold, it is determined that the second material list meets the first recommendation condition, and when the first ratio is less than or equal to the second threshold, it is determined that the second material list does not meet the first recommendation condition. In this way, the corresponding verification method is configured for the relevance priority prompt template to ensure the accuracy of the verification of the second material list.

[0119] In the embodiment of the present application, the prompt template includes a second prompt template, and the second prompt template corresponds to the second recommendation condition. The above S110 may specifically include the following S124 to S130:

[0120] S124: Determine a first material category set preferred by the user based on user behavior statistical data.

[0121] The prompt template includes a second prompt template, and the second prompt template corresponds to the second recommendation condition.

[0122] Furthermore, the second prompt template is a diversity-first prompt template, and the second recommendation condition is to maximize the diversity of recommended materials.

[0123] On this basis, in the material recommendation method provided in the embodiment of the present application, after obtaining the second material list based on the second prompt template, during the process of verifying the second material list, the first material category set preferred by the user is determined based on user behavior statistical data.

[0124] Among them, the first material category set can specifically indicate the user's preference for first-level classification materials, the user's preference for second-level classification materials, the user's preference for material creators, etc., and no specific restrictions are made here.

[0125] S126: Obtain a second material type set included in the second material list.

[0126] The second material type set is at the same level as the user preference indicated by the first material type set. That is, when the first material type set indicates the user's preference for secondary classified materials, the second material type set is the set of secondary classified materials included in the second material list.

[0127] S128: Calculate a second similarity between the user and the second material list according to an intersection-and-union ratio of the first material category set and the second material category set.

[0128] The second similarity is the Jaccard similarity, which is between 0 and 1. The Jaccard similarity is used to compare the differences between two sets. The smaller the Jaccard similarity, the greater the difference between the two sets.

[0129] Specifically, in the material recommendation method provided in the embodiment of the present application, after obtaining the first material type set and the second material type set, the intersection and union ratio of the first material type set and the second material type set is calculated according to the following formula to obtain the second similarity between the user and the second material list.

[0130]

[0131] Among them, J(A,B) represents the Jaccard similarity between the user and the second material list, that is, the second similarity, A represents the first material type set, B represents the second material type set, AIB represents the intersection of A and B, and A∪B represents the union of A and B.

[0132] Exemplarily, the first material category set indicates the user's preference for secondary classification materials, the first material category set is [animation, sports, entertainment, funny, news, animals, scenery], the set of secondary classification materials contained in the second material list, that is, the second material category set, is [animation, sports, furniture, education], the intersection of the first material category set and the second material category set is [animation, sports], and the union of the first material category set and the second material category set is [animation, sports, entertainment, funny, news, animals, scenery, furniture, education]. At this time, the Jaccard similarity between the user and the second material list, that is, the second similarity = 2 / 9 = 0.22.

[0133] S130: When the second similarity is less than a third threshold, determine that the second material list meets the second recommendation condition; when the second similarity is greater than or equal to the third threshold, determine that the second material list does not meet the second recommendation condition.

[0134] Specifically, in the material recommendation method provided in the embodiment of the present application, after calculating the Jaccard similarity between the user and the second material list, that is, the second similarity, if the second similarity is less than the third threshold, it is determined that the second material list meets the second recommendation condition, and if the second similarity is greater than or equal to the third threshold, it is determined that the second material list does not meet the second recommendation condition. In this way, the corresponding verification method is configured for the diversity priority prompt template to ensure the accuracy of the verification of the second material list.

[0135] Among them, the specific value of the third threshold can be set by those skilled in the art according to actual conditions. For example, the third threshold can be a value such as 0.65, and no specific limitation is made here.

[0136] In the above-mentioned embodiment provided by the present application, the prompt template includes a second prompt template, and the second prompt template corresponds to the second recommendation condition. In the process of verifying the second material list, the first material type set preferred by the user is determined based on the user behavior statistical data; the second material type set included in the second material list is obtained; the second similarity between the user and the second material list is calculated based on the intersection and union ratio of the first material type set and the second material type set; when the second similarity is less than the third threshold, it is determined that the second material list meets the second recommendation condition, and when the second similarity is greater than or equal to the third threshold, it is determined that the second material list does not meet the second recommendation condition. In this way, the corresponding verification method is configured for the diversity priority prompt template to ensure the accuracy of the verification of the second material list.

[0137] In the embodiment of the present application, the prompt template includes a third prompt template, and the third prompt template corresponds to a third recommendation condition. The above S110 may specifically include the following S132 to S138:

[0138] S132: Calculate a first similarity between the user and each material in the second material list, and calculate a second similarity between the user and the second material list.

[0139] The prompt template includes a third prompt template, and the third prompt template corresponds to a third recommendation condition.

[0140] Furthermore, the third prompt template is a relevance and diversity balance prompt template, and the third recommendation condition is to balance the relevance and diversity of the recommended materials.

[0141] On this basis, in the material recommendation method provided in the embodiment of the present application, after obtaining the second material list based on the third prompt template, in the process of verifying the second material list, the first similarity between the user and each material in the second material list is calculated in the above manner, and the second similarity between the user and the second material list is calculated.

[0142] S134: Perform weighted processing on the first similarity and the second similarity to obtain a third similarity corresponding to each material.

[0143] Specifically, in the material recommendation method provided in the embodiment of the present application, after calculating the first similarity between the user and each material in the second material list, and calculating the second similarity between the user and the second material list, the first similarity and each second similarity are weighted according to the following formula to obtain the third similarity corresponding to each material.

[0144] scoreI=α×similarI+(1-α)(J(A,B))

[0145] Among them, scoreI represents the third similarity corresponding to material pidI, and α represents the weighting coefficient.

[0146] S136: Counting the number of second materials whose third similarity is within a preset range.

[0147] Specifically, in the material recommendation method provided in the embodiment of the present application, after calculating the third similarity between the user and each material, the third similarity corresponding to each material in the second material table is compared with the preset range, and the number of second materials whose third similarity is within the preset range is counted.

[0148] Among them, for the specific value range of the above preset range, those skilled in the art can set it according to actual conditions. For example, the above preset range can be [0.5, 0.8], and no specific limitation is made here.

[0149] S138: When a second ratio of the second material quantity to the total number of materials in the second material list is greater than a fourth threshold, it is determined that the second material list meets the third recommendation condition; when the second ratio is less than or equal to the fourth threshold, it is determined that the second material list does not meet the third recommendation condition.

[0150] Specifically, in the material recommendation method provided in the embodiment of the present application, when the second ratio of the second material quantity to the total number of materials in the second material list is greater than the fourth threshold, it is determined that the second material list meets the third recommendation condition, and when the second ratio is less than or equal to the fourth threshold, it is determined that the second material list does not meet the third recommendation condition. In this way, the corresponding verification method is configured for the relevance and diversity balance prompt template to ensure the accuracy of the verification of the second material list.

[0151] Among them, the specific value of the fourth threshold can be set by those skilled in the art according to actual conditions. For example, the fourth threshold can be 60% or other values, and no specific limitation is made here.

[0152] In the above-mentioned embodiment provided by the present application, the prompt template includes a third prompt template, and the third prompt template corresponds to the third recommendation condition. In the process of verifying the second material list, the first similarity between the user and each material in the second material list is calculated, and the second similarity between the user and the second material list is calculated; the first similarity and the second similarity are weighted to obtain the third similarity corresponding to each material; the number of second materials whose third similarity is within a preset range is counted; when the second ratio of the number of second materials to the total number of materials in the second material list is greater than the fourth threshold, it is determined that the second material list meets the third recommendation condition, and when the second ratio is less than or equal to the fourth threshold, it is determined that the second material list does not meet the third recommendation condition. In this way, the corresponding verification method is configured for the relevance and diversity balance prompt template to ensure the accuracy of the verification of the second material list.

[0153] In the embodiment of the present application, the prompt template includes a fourth prompt template, and the fourth prompt template corresponds to a fourth recommendation condition. The above S110 may specifically include the following S140 to S146:

[0154] S140: Obtain the release time of each material in the second material list.

[0155] The prompt template includes a fourth prompt template, and the fourth prompt template corresponds to a fourth recommendation condition.

[0156] Furthermore, the fourth prompt template is a discoverability-first prompt template, and the fourth recommendation condition is to recommend more new content related to the user's interests.

[0157] S142: Calculate the first similarity between the user and each material.

[0158] Specifically, in the material recommendation method provided in the embodiment of the present application, after obtaining the second material list based on the fourth prompt template, in the process of verifying the second material list, the release time of each material in the second material list is obtained, and the first similarity between the user and each material is calculated in the above manner.

[0159] S144: Counting the number of third materials whose first similarity is greater than or equal to the first threshold and whose release time is less than a preset time.

[0160] Specifically, in the material recommendation method provided in the embodiment of the present application, after obtaining the release time of each material in the second material list and calculating the first similarity between the user and each material, the first similarity corresponding to each material in the second material list is compared with the first threshold, and the number of third materials whose first similarity is greater than or equal to the first threshold and whose release time is less than the preset time is counted.

[0161] Among them, the specific value of the above-mentioned preset time can be set by technical personnel in this field according to actual conditions. For example, the above-mentioned preset time can be a value such as 30 days, and no specific limitation is made here.

[0162] S146: When a third ratio of the third material quantity to the total number of materials in the second material list is greater than a fifth threshold, it is determined that the second material list meets the fourth recommendation condition; when the third ratio is less than or equal to the fifth threshold, it is determined that the second material list does not meet the fourth recommendation condition.

[0163] Specifically, in the material recommendation method provided in the embodiment of the present application, when the third ratio of the third material quantity to the total number of materials in the second material list is greater than the fifth threshold, it is determined that the second material list meets the fourth recommendation condition, and when the third ratio is less than or equal to the fifth threshold, it is determined that the second material list does not meet the fourth recommendation condition. In this way, the corresponding verification method is configured for the discoverability priority prompt template to ensure the accuracy of the verification of the second material list.

[0164] Among them, the specific value of the fifth threshold can be set by those skilled in the art according to actual conditions. For example, the fifth threshold can be 50% or other values, and no specific limitation is made here.

[0165] In the above-mentioned embodiment provided by the present application, the prompt template includes a fourth prompt template, and the fourth prompt template corresponds to the fourth recommendation condition. In the process of verifying the second material list, the release time of each material in the second material list is obtained; the first similarity between the user and each material is calculated; the number of third materials whose first similarity is greater than or equal to the first threshold and whose release time is less than the preset time is counted; when the third ratio of the number of third materials to the total number of materials in the second material list is greater than the fifth threshold, it is determined that the second material list meets the fourth recommendation condition; when the third ratio is less than or equal to the fifth threshold, it is determined that the second material list does not meet the fourth recommendation condition. In this way, the corresponding verification method is configured for the discoverability priority prompt template to ensure the accuracy of the verification of the second material list.

[0166] In the embodiment of the present application, after the above S114, the above material recommendation method may further include the following S148 to S152:

[0167] S148: Recording user behavior data on multiple materials.

[0168] Among them, the above-mentioned behavioral data is used to indicate the user's exposure, clicks, likes and other behaviors on the recommended multiple materials.

[0169] S150: Determine user feedback data on multiple materials based on the behavior data.

[0170] The feedback data is the CTR of the user for the recommended multiple materials. In this way, the introduction of user feedback data on the materials makes it easier to adaptively adjust and optimize the material recommendation results based on the user's feedback on the previous recommended materials.

[0171] S152: Update the first feature data according to the behavior data and the feedback data.

[0172] Specifically, in the material recommendation method provided in the embodiment of the present application, after displaying multiple materials in the second material list, the user's exposure, clicks, and likes on the multiple materials are recorded to obtain the user's behavior data on the multiple materials, and the user's feedback data on the multiple materials is determined based on the behavior data. On this basis, the first feature data is updated based on the behavior data and the feedback data. In this way, on the one hand, by introducing the user's feedback data on the material, the material recommendation results can be adaptively learned and adjusted, thereby improving the accuracy of the material recommendation; on the other hand, the recommendation system is closed-loop, realizing end-to-end learning of material recommendation.

[0173] Among them, end-to-end learning refers to the process in which the model is not trained step by step or module during the learning process, but directly from data input to model learning.

[0174] In an optional embodiment, for consumable materials such as takeaways, commodities, rewarded videos, paid videos, paid music, and paid wallpapers, the user's consumption data of multiple materials can also be recorded, and the first characteristic data can be updated according to the consumption data. In this way, the user's consumption data of materials is introduced to facilitate adaptive adjustment and optimization of the material recommendation results based on the user's consumption data of the previously recommended materials, thus realizing a closed loop of the material recommendation link.

[0175] In the above embodiment provided by the present application, after displaying multiple materials in the second material list, the user's behavior data on the multiple materials is recorded; the user's feedback data on the multiple materials is determined based on the behavior data; and the first feature data is updated based on the behavior data and the feedback data. In this way, on the one hand, by introducing the user's feedback data on the materials, the material recommendation results can be adaptively learned and adjusted, thereby improving the accuracy of the material recommendation; on the other hand, the recommendation system is closed-loop, realizing end-to-end learning of material recommendations.

[0176] In summary, if Figure 2 As shown, the material recommendation method provided in the embodiment of the present application may specifically include the following S202 to S218:

[0177] S202: Generate first feature data offline.

[0178] S204: Determine: trigger the material recommendation process, if yes, execute S206, if not, end the process.

[0179] S206: Obtain a first material list.

[0180] S208: Combine the first characteristic data and the first material list to obtain second characteristic data.

[0181] S210: Introduce a feature awareness template and select a prompt template.

[0182] S212: Generate a second material list through the first model according to the second feature data and the prompt template.

[0183] S214: Determine: The second material list verification passes, if yes, execute S216, if no, execute S210.

[0184] S216: Display multiple materials in the second material list.

[0185] S218: Obtaining user behavior data on the multiple materials, and determining user feedback data on the multiple materials based on the behavior data.

[0186] After S218 is executed, S202 will continue to be executed to update the first feature data in real time according to the behavior data and feedback data.

[0187] The material recommendation method provided in the embodiment of the present application can be executed by a material recommendation device. In the embodiment of the present application, the material recommendation device provided in the embodiment of the present application is described by taking the material recommendation device executing the above material recommendation method as an example.

[0188] like Figure 4 As shown, an embodiment of the present application provides a material recommendation device 400, which may include the following acquisition unit 402 and processing unit 404.

[0189] An acquisition unit 402 is used to acquire first characteristic data in an application, where the application includes a material to be recommended;

[0190] The acquisition unit 402 is further used to acquire a first material list, where the first material list is obtained through the first arrangement process;

[0191] The processing unit 404 is used to determine the second characteristic data according to the first characteristic data and the first material list;

[0192] The processing unit 404 is further configured to generate a second material list according to the second characteristic data and a prompt template corresponding to the material type through the first model.

[0193] The material recommendation device 400 provided in the embodiment of the present application obtains the first characteristic data in the application, the application includes the materials to be recommended; obtains the first material list, the first material list is obtained through the first arrangement process; determines the second characteristic data according to the first characteristic data and the first material list; generates the second material list according to the second characteristic data and the prompt template corresponding to the material type through the first model. Through the above-mentioned material recommendation device 400, the second characteristic data is generated based on the first characteristic data of the material recommendation and the first material list after the first arrangement process, and then the second material list is generated according to the second characteristic data and the prompt template corresponding to the material type through the first model. In this way, the rearrangement task is converted into a sequence generation task of the first model, and by introducing the prompt template corresponding to the material type, a material recommendation list is generated in the case of zero samples, which can not only improve the accuracy and efficiency of material recommendation, but also ensure the diversity of recommendation results.

[0194] In an embodiment of the present application, the first characteristic data includes at least one of user basic data, user behavior statistics data, material basic data, material behavior statistics data, and user interaction data on the material. The processing unit 404 is specifically used to: determine the first material information according to the first material list, the first material information including the identification information of each material in the first material list and the score information in the first arrangement processing; combine the user basic data and the user behavior statistics data to obtain the first information; combine the identification information, score information, order information in the first material list, material basic data, material behavior statistics data and interaction data of each material in the first material list to obtain the second information corresponding to each material; combine the first information and the second information corresponding to each material to obtain the second characteristic data.

[0195] In the above-mentioned embodiment provided by the present application, the first characteristic data includes at least one of user basic data, user behavior statistics, material basic data, material behavior statistics, and user interaction data on materials. In the process of determining the second characteristic data according to the first characteristic data and the first material list, the first material information is determined according to the first material list, and the first material information includes the identification information of each material in the first material list and the score information in the first arrangement process; the user basic data and the user behavior statistics are combined to obtain the first information; the identification information, score information, order information in the first material list, material basic data, material behavior statistics, and interaction data of each material in the first material list are combined to obtain the second information corresponding to each material; the first information and the second information corresponding to each material are combined to obtain the second characteristic data. In this way, based on the output result of the refined arrangement task and the first characteristic data related to the user and the material, the second characteristic data required for material recommendation is generated, which ensures the accuracy of the second characteristic data, and further ensures the accuracy of subsequent material recommendation based on the second characteristic data.

[0196] In the embodiment of the present application, the processing unit 404 is specifically used to: determine the output target and output format according to the prompt template; construct a generation instruction according to the second characteristic data, the output target and the output format; input the generation instruction into the first model, and generate a second material list that conforms to the output format through the first model.

[0197] In the above-mentioned embodiment provided by the present application, in the process of generating the second material list according to the prompt template corresponding to the second feature data and the material type through the first model, the output target and output format are determined according to the prompt template; the generation instruction is constructed according to the second feature data, the output target and the output format; the generation instruction is input into the first model, and the second material list conforming to the output format is generated through the first model. In this way, the instruction following ability, zero-sample rearrangement ability and context learning ability of the large model are utilized to innovatively transform the material rearrangement task into the text sequence generation task of the large model, and through a specific prompt template, the large model outputs the second material list according to the specific output target and output format, thereby ensuring the accuracy and diversity of material recommendations.

[0198] In the embodiment of the present application, after the second material list is generated, the processing unit 404 is also used to: verify the second material list; if the second material list does not meet the recommendation conditions corresponding to the prompt template, regenerate the second material list through the first model until the generated second material list meets the recommendation conditions or the number of times the second material list is generated reaches a preset number threshold; the material recommendation device 400 also includes: a display unit 406, which is used to display multiple materials in the second material list when the second material list meets the recommendation conditions corresponding to the prompt template.

[0199] In the above embodiment provided by the present application, after the second material list is generated, the second material list is verified; if the second material list does not meet the recommendation conditions corresponding to the prompt template, the second material list is regenerated through the first model until the generated second material list meets the recommendation conditions or the number of times the second material list is generated reaches a preset number threshold; if the second material list meets the recommendation conditions corresponding to the prompt template, multiple materials in the second material list are displayed. In this way, before displaying the material recommendation results to the user, the recommended materials are first verified to ensure the accuracy of the final material recommendation results.

[0200] In an embodiment of the present application, the prompt template includes a first prompt template, the first prompt template corresponds to the first recommendation condition, and the processing unit 404 is specifically used to: embed the user and each material in the second material list to obtain the first vector of the user and the second vector of each material in the second material list; calculate the first similarity between the user and each material according to the first vector and the second vector of each material; count the number of first materials whose first similarity is greater than or equal to the first threshold; when the first ratio of the number of first materials to the total number of materials in the second material list is greater than the second threshold, determine that the second material list meets the first recommendation condition, and when the first ratio is less than or equal to the second threshold, determine that the second material list does not meet the first recommendation condition.

[0201] In the above-mentioned embodiment provided by the present application, the prompt template includes a first prompt template, the first prompt template corresponds to the first recommendation condition, and in the process of verifying the second material list, the user and each material in the second material list are embedded and represented to obtain the first vector of the user and the second vector of each material in the second material list; the first similarity between the user and each material is calculated according to the first vector and the second vector of each material; the number of first materials whose first similarity is greater than or equal to the first threshold is counted; when the first ratio of the number of first materials to the total number of materials in the second material list is greater than the second threshold, it is determined that the second material list meets the first recommendation condition, and when the first ratio is less than or equal to the second threshold, it is determined that the second material list does not meet the first recommendation condition. In this way, the corresponding verification method is configured for the relevance priority prompt template to ensure the accuracy of the verification of the second material list.

[0202] In an embodiment of the present application, the prompt template includes a second prompt template, the second prompt template corresponds to the second recommendation condition, and the processing unit 404 is specifically used to: determine the first material type set preferred by the user based on user behavior statistical data; obtain the second material type set included in the second material list; calculate the second similarity between the user and the second material list based on the intersection and union ratio of the first material type set and the second material type set; when the second similarity is less than a third threshold, determine that the second material list meets the second recommendation condition, and when the second similarity is greater than or equal to the third threshold, determine that the second material list does not meet the second recommendation condition.

[0203] In the above-mentioned embodiment provided by the present application, the prompt template includes a second prompt template, and the second prompt template corresponds to the second recommendation condition. In the process of verifying the second material list, the first material type set preferred by the user is determined based on the user behavior statistical data; the second material type set included in the second material list is obtained; the second similarity between the user and the second material list is calculated based on the intersection and union ratio of the first material type set and the second material type set; when the second similarity is less than the third threshold, it is determined that the second material list meets the second recommendation condition, and when the second similarity is greater than or equal to the third threshold, it is determined that the second material list does not meet the second recommendation condition. In this way, the corresponding verification method is configured for the diversity priority prompt template to ensure the accuracy of the verification of the second material list.

[0204] In an embodiment of the present application, the prompt template includes a third prompt template, the third prompt template corresponds to the third recommendation condition, and the processing unit 404 is specifically used to: calculate the first similarity between the user and each material in the second material list, and calculate the second similarity between the user and the second material list; perform weighted processing on the first similarity and the second similarity to obtain the third similarity corresponding to each material; count the number of second materials whose third similarity is within a preset range; when the second ratio of the number of second materials to the total number of materials in the second material list is greater than a fourth threshold, determine that the second material list meets the third recommendation condition, and when the second ratio is less than or equal to the fourth threshold, determine that the second material list does not meet the third recommendation condition.

[0205] In the above-mentioned embodiment provided by the present application, the prompt template includes a third prompt template, and the third prompt template corresponds to the third recommendation condition. In the process of verifying the second material list, the first similarity between the user and each material in the second material list is calculated, and the second similarity between the user and the second material list is calculated; the first similarity and the second similarity are weighted to obtain the third similarity corresponding to each material; the number of second materials whose third similarity is within a preset range is counted; when the second ratio of the number of second materials to the total number of materials in the second material list is greater than the fourth threshold, it is determined that the second material list meets the third recommendation condition, and when the second ratio is less than or equal to the fourth threshold, it is determined that the second material list does not meet the third recommendation condition. In this way, the corresponding verification method is configured for the relevance and diversity balance prompt template to ensure the accuracy of the verification of the second material list.

[0206] In an embodiment of the present application, the prompt template includes a fourth prompt template, the fourth prompt template corresponds to the fourth recommendation condition, and the processing unit 404 is specifically used to: obtain the release time of each material in the second material list; calculate the first similarity between the user and each material; count the number of third materials whose first similarity is greater than or equal to the first threshold and whose release time is less than the preset time; when a third ratio of the number of third materials to the total number of materials in the second material list is greater than a fifth threshold, determine that the second material list meets the fourth recommendation condition; when the third ratio is less than or equal to the fifth threshold, determine that the second material list does not meet the fourth recommendation condition.

[0207] In the above-mentioned embodiment provided by the present application, the prompt template includes a fourth prompt template, and the fourth prompt template corresponds to the fourth recommendation condition. In the process of verifying the second material list, the release time of each material in the second material list is obtained; the first similarity between the user and each material is calculated; the number of third materials whose first similarity is greater than or equal to the first threshold and whose release time is less than the preset time is counted; when the third ratio of the number of third materials to the total number of materials in the second material list is greater than the fifth threshold, it is determined that the second material list meets the fourth recommendation condition; when the third ratio is less than or equal to the fifth threshold, it is determined that the second material list does not meet the fourth recommendation condition. In this way, the corresponding verification method is configured for the discoverability priority prompt template to ensure the accuracy of the verification of the second material list.

[0208] In an embodiment of the present application, after displaying multiple materials in the second material list, the processing unit 404 is also used to: record the user's behavior data on the multiple materials; determine the user's feedback data on the multiple materials based on the behavior data; and update the first feature data based on the behavior data and the feedback data.

[0209] In the above embodiment provided by the present application, after displaying multiple materials in the second material list, the user's behavior data on the multiple materials is recorded; the user's feedback data on the multiple materials is determined based on the behavior data; and the first feature data is updated based on the behavior data and the feedback data. In this way, on the one hand, by introducing the user's feedback data on the materials, the material recommendation results can be adaptively learned and adjusted, thereby improving the accuracy of the material recommendation; on the other hand, the recommendation system is closed-loop, realizing end-to-end learning of material recommendations.

[0210] The material recommendation device 400 in the embodiment of the present application can be an electronic device or a component in an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal or other devices other than a terminal. Exemplarily, the electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, a vehicle-mounted electronic device, a mobile Internet device (Mobile Internet Device, MID), an augmented reality (augmented reality, AR) / virtual reality (virtual reality, VR) device, a robot, a wearable device, an ultra-mobile personal computer (ultra-mobile personal computer, UMPC), a netbook or a personal digital assistant (personal digital assistant, PDA), etc. It can also be a server, a network attached storage (Network Attached Storage, NAS), a personal computer (personal computer, PC), a television (television, TV), a teller machine or a self-service machine, etc., which is not specifically limited in the embodiment of the present application.

[0211] The material recommendation device 400 in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0212] The material recommendation device 400 provided in the embodiment of the present application can achieve Figure 1 and Figure 2 To avoid repetition, the various processes implemented by the method embodiment are not described here.

[0213] Alternatively, if Figure 5 As shown, an embodiment of the present application further provides an electronic device 500, including a processor 502 and a memory 504, wherein the memory 504 stores programs or instructions that can be executed on the processor 502, and when the program or instructions are executed by the processor 502, the various steps of the above-mentioned material recommendation method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, they are not described here.

[0214] It should be noted that the electronic devices in the embodiments of the present application include the above-mentioned mobile electronic devices and non-mobile electronic devices.

[0215] Figure 6 A schematic diagram of the hardware structure of an electronic device to implement an embodiment of the present application.

[0216] The electronic device 600 includes but is not limited to: a radio frequency unit 601, a network module 602, an audio output unit 603, an input unit 604, a sensor 605, a display unit 606, a user input unit 607, an interface unit 608, a memory 609, a processor 610 and other components.

[0217] Those skilled in the art will appreciate that the electronic device 600 may also include a power source (such as a battery) for supplying power to each component, and the power source may be logically connected to the processor 610 through a power management system, thereby implementing functions such as managing charging, discharging, and power consumption management through the power management system. Figure 6 The electronic device structure shown in the figure does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently, which will not be described in detail here.

[0218] The processor 610 is used to obtain first characteristic data in an application, where the application includes materials to be recommended.

[0219] The processor 610 is further configured to obtain a first material list, where the first material list is obtained through a first arrangement process.

[0220] The processor 610 is further configured to determine the second characteristic data according to the first characteristic data and the first material list.

[0221] The processor 610 is further configured to generate a second material list according to the second feature data and a prompt template corresponding to the material type through the first model.

[0222] In an embodiment of the present application, first characteristic data in an application is obtained, and the application includes materials to be recommended; a first material list is obtained, and the first material list is obtained through a first arrangement process; second characteristic data is determined based on the first characteristic data and the first material list; and a second material list is generated through a first model according to the second characteristic data and a prompt template corresponding to the material type. In an embodiment of the present application, second characteristic data is generated based on the first characteristic data of material recommendation and the first material list after the first arrangement process, and then a second material list is generated through the first model according to the second characteristic data and a prompt template corresponding to the material type. In this way, the rearrangement task is converted into a sequence generation task of the first model, and by introducing a prompt template corresponding to the material type, a material recommendation list is generated in the case of zero samples, which can not only improve the accuracy and efficiency of material recommendations, but also ensure the diversity of recommendation results.

[0223] Optionally, the first characteristic data includes at least one of user basic data, user behavior statistics data, material basic data, material behavior statistics data and user interaction data on the material, and the processor 610 is specifically used to: determine the first material information according to the first material list, the first material information including the identification information of each material in the first material list and the score information in the first arrangement processing; combine the user basic data and the user behavior statistics data to obtain the first information; combine the identification information, score information, order information in the first material list, material basic data, material behavior statistics data and interaction data of each material in the first material list to obtain the second information corresponding to each material; combine the first information and the second information corresponding to each material to obtain the second characteristic data.

[0224] In the above-mentioned embodiment provided by the present application, the first characteristic data includes at least one of user basic data, user behavior statistics, material basic data, material behavior statistics, and user interaction data on materials. In the process of determining the second characteristic data according to the first characteristic data and the first material list, the first material information is determined according to the first material list, and the first material information includes the identification information of each material in the first material list and the score information in the first arrangement process; the user basic data and the user behavior statistics are combined to obtain the first information; the identification information, score information, order information in the first material list, material basic data, material behavior statistics, and interaction data of each material in the first material list are combined to obtain the second information corresponding to each material; the first information and the second information corresponding to each material are combined to obtain the second characteristic data. In this way, based on the output result of the refined arrangement task and the first characteristic data related to the user and the material, the second characteristic data required for material recommendation is generated, which ensures the accuracy of the second characteristic data, and further ensures the accuracy of subsequent material recommendation based on the second characteristic data.

[0225] Optionally, the processor 610 is specifically used to: determine the output target and output format according to the prompt template; construct a generation instruction according to the second characteristic data, the output target and the output format; input the generation instruction into the first model, and generate a second material list that conforms to the output format through the first model.

[0226] In the above-mentioned embodiment provided by the present application, in the process of generating the second material list according to the prompt template corresponding to the second feature data and the material type through the first model, the output target and output format are determined according to the prompt template; the generation instruction is constructed according to the second feature data, the output target and the output format; the generation instruction is input into the first model, and the second material list conforming to the output format is generated through the first model. In this way, the instruction following ability, zero-sample rearrangement ability and context learning ability of the large model are utilized to innovatively transform the material rearrangement task into the text sequence generation task of the large model, and through a specific prompt template, the large model outputs the second material list according to the specific output target and output format, thereby ensuring the accuracy and diversity of material recommendations.

[0227] Optionally, after generating the second material list, the processor 610 is further used to: verify the second material list; if the second material list does not meet the recommendation conditions corresponding to the prompt template, regenerate the second material list through the first model until the generated second material list meets the recommendation conditions or the number of times the second material list is generated reaches a preset number threshold; the display unit 606 is used to display multiple materials in the second material list if the second material list meets the recommendation conditions corresponding to the prompt template.

[0228] In the above embodiment provided by the present application, after the second material list is generated, the second material list is verified; if the second material list does not meet the recommendation conditions corresponding to the prompt template, the second material list is regenerated through the first model until the generated second material list meets the recommendation conditions or the number of times the second material list is generated reaches a preset number threshold; if the second material list meets the recommendation conditions corresponding to the prompt template, multiple materials in the second material list are displayed. In this way, before displaying the material recommendation results to the user, the recommended materials are first verified to ensure the accuracy of the final material recommendation results.

[0229] Optionally, the prompt template includes a first prompt template, the first prompt template corresponds to the first recommendation condition, and the processor 610 is specifically used to: embed the user and each material in the second material list to obtain the first vector of the user and the second vector of each material in the second material list; calculate the first similarity between the user and each material according to the first vector and the second vector of each material; count the number of first materials whose first similarity is greater than or equal to a first threshold; when a first ratio of the number of first materials to the total number of materials in the second material list is greater than a second threshold, determine that the second material list meets the first recommendation condition, and when the first ratio is less than or equal to the second threshold, determine that the second material list does not meet the first recommendation condition.

[0230] In the above-mentioned embodiment provided by the present application, the prompt template includes a first prompt template, the first prompt template corresponds to the first recommendation condition, and in the process of verifying the second material list, the user and each material in the second material list are embedded and represented to obtain the first vector of the user and the second vector of each material in the second material list; the first similarity between the user and each material is calculated according to the first vector and the second vector of each material; the number of first materials whose first similarity is greater than or equal to the first threshold is counted; when the first ratio of the number of first materials to the total number of materials in the second material list is greater than the second threshold, it is determined that the second material list meets the first recommendation condition, and when the first ratio is less than or equal to the second threshold, it is determined that the second material list does not meet the first recommendation condition. In this way, the corresponding verification method is configured for the relevance priority prompt template to ensure the accuracy of the verification of the second material list.

[0231] Optionally, the prompt template includes a second prompt template, the second prompt template corresponds to a second recommendation condition, and the processor 610 is specifically used to: determine a first material type set preferred by the user based on user behavior statistical data; obtain a second material type set included in the second material list; calculate a second similarity between the user and the second material list based on an intersection and union ratio of the first material type set and the second material type set; when the second similarity is less than a third threshold, determine that the second material list meets the second recommendation condition, and when the second similarity is greater than or equal to the third threshold, determine that the second material list does not meet the second recommendation condition.

[0232] In the above-mentioned embodiment provided by the present application, the prompt template includes a second prompt template, and the second prompt template corresponds to the second recommendation condition. In the process of verifying the second material list, the first material type set preferred by the user is determined based on the user behavior statistical data; the second material type set included in the second material list is obtained; the second similarity between the user and the second material list is calculated based on the intersection and union ratio of the first material type set and the second material type set; when the second similarity is less than the third threshold, it is determined that the second material list meets the second recommendation condition, and when the second similarity is greater than or equal to the third threshold, it is determined that the second material list does not meet the second recommendation condition. In this way, the corresponding verification method is configured for the diversity priority prompt template to ensure the accuracy of the verification of the second material list.

[0233] Optionally, the prompt template includes a third prompt template, the third prompt template corresponds to the third recommendation condition, and the processor 610 is specifically used to: calculate a first similarity between the user and each material in the second material list, and calculate a second similarity between the user and the second material list; weighted processing is performed on the first similarity and the second similarity to obtain a third similarity corresponding to each material; counting the number of second materials whose third similarity is within a preset range; and determining that the second material list meets the third recommendation condition when a second ratio of the number of second materials to the total number of materials in the second material list is greater than a fourth threshold, and determining that the second material list does not meet the third recommendation condition when the second ratio is less than or equal to the fourth threshold.

[0234] In the above-mentioned embodiment provided by the present application, the prompt template includes a third prompt template, and the third prompt template corresponds to the third recommendation condition. In the process of verifying the second material list, the first similarity between the user and each material in the second material list is calculated, and the second similarity between the user and the second material list is calculated; the first similarity and the second similarity are weighted to obtain the third similarity corresponding to each material; the number of second materials whose third similarity is within a preset range is counted; when the second ratio of the number of second materials to the total number of materials in the second material list is greater than the fourth threshold, it is determined that the second material list meets the third recommendation condition, and when the second ratio is less than or equal to the fourth threshold, it is determined that the second material list does not meet the third recommendation condition. In this way, the corresponding verification method is configured for the relevance and diversity balance prompt template to ensure the accuracy of the verification of the second material list.

[0235] Optionally, the prompt template includes a fourth prompt template, the fourth prompt template corresponds to the fourth recommendation condition, and the processor 610 is specifically used to: obtain the release time of each material in the second material list; calculate the first similarity between the user and each material; count the number of third materials whose first similarity is greater than or equal to the first threshold and whose release time is less than a preset time; when a third ratio of the number of third materials to the total number of materials in the second material list is greater than a fifth threshold, determine that the second material list meets the fourth recommendation condition; when the third ratio is less than or equal to the fifth threshold, determine that the second material list does not meet the fourth recommendation condition.

[0236] In the above-mentioned embodiment provided by the present application, the prompt template includes a fourth prompt template, and the fourth prompt template corresponds to the fourth recommendation condition. In the process of verifying the second material list, the release time of each material in the second material list is obtained; the first similarity between the user and each material is calculated; the number of third materials whose first similarity is greater than or equal to the first threshold and whose release time is less than the preset time is counted; when the third ratio of the number of third materials to the total number of materials in the second material list is greater than the fifth threshold, it is determined that the second material list meets the fourth recommendation condition; when the third ratio is less than or equal to the fifth threshold, it is determined that the second material list does not meet the fourth recommendation condition. In this way, the corresponding verification method is configured for the discoverability priority prompt template to ensure the accuracy of the verification of the second material list.

[0237] Optionally, after displaying multiple materials in the second material list, the processor 610 is further used to: record user behavior data on the multiple materials; determine user feedback data on the multiple materials based on the behavior data; and update the first feature data based on the behavior data and the feedback data.

[0238] In the above embodiment provided by the present application, after displaying multiple materials in the second material list, the user's behavior data on the multiple materials is recorded; the user's feedback data on the multiple materials is determined based on the behavior data; and the first feature data is updated based on the behavior data and the feedback data. In this way, on the one hand, by introducing the user's feedback data on the materials, the material recommendation results can be adaptively learned and adjusted, thereby improving the accuracy of the material recommendation; on the other hand, the recommendation system is closed-loop, realizing end-to-end learning of material recommendations.

[0239] It should be understood that in the embodiment of the present application, the input unit 604 may include a graphics processor (GPU) 6041 and a microphone 6042, and the graphics processor 6041 processes the image data of a static picture or video obtained by an image capture device (such as a camera) in a video capture mode or an image capture mode. The display unit 606 may include a display panel 6061, and the display panel 6061 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc. The user input unit 607 includes a touch panel 6071 and at least one of other input devices 6072. The touch panel 6071 is also called a touch screen. The touch panel 6071 may include two parts: a touch detection device and a touch controller. Other input devices 6072 may include, but are not limited to, a physical keyboard, function keys (such as a volume control button, a switch button, etc.), a trackball, a mouse, and a joystick, which will not be repeated here.

[0240] The memory 609 can be used to store software programs and various data. The memory 609 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, an application program or instructions required for at least one function (such as a sound playback function, an image playback function, etc.), etc. In addition, the memory 609 may include a volatile memory or a non-volatile memory, or the memory 609 may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), a static random access memory (SRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDRSDRAM), an enhanced synchronous dynamic random access memory (ESDRAM), a synchronous link dynamic random access memory (SLDRAM) and a direct memory bus random access memory (DRRAM). The memory 609 in the embodiment of the present application includes but is not limited to these and any other suitable types of memory.

[0241] The processor 610 may include one or more processing units; optionally, the processor 610 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communication signals, such as a baseband processor. It is understandable that the modem processor may not be integrated into the processor 610.

[0242] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, each process of the above-mentioned material recommendation method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0243] The processor is the processor in the electronic device in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0244] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned material recommendation method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0245] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0246] An embodiment of the present application provides a computer program product, which is stored in a storage medium. The program product is executed by at least one processor to implement the various processes of the above-mentioned material recommendation method embodiment and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0247] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0248] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods of each embodiment of the present application.

[0249] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A material recommendation method, characterized in that: include: Acquire first characteristic data in an application, wherein the application includes a material to be recommended; Obtaining a first material list, where the first material list is obtained through a first arrangement process; Determine second characteristic data according to the first characteristic data and the first material list; Generate a second material list through the first model according to the second characteristic data and the prompt template corresponding to the material type.

2. The material recommendation method according to claim 1, characterized in that: The generating a second material list according to the second feature data and a prompt template corresponding to the material type by the first model includes: Determine the output target and output format according to the prompt template; constructing a generation instruction according to the second characteristic data, the output target and the output format; The generation instruction is input into the first model, and the second material list conforming to the output format is generated through the first model.

3. The material recommendation method according to claim 1 or 2, characterized in that: After the second material list is generated, the material recommendation method further includes: Verifying the second material list; If the second material list does not meet the recommendation condition corresponding to the prompt template, regenerate the second material list through the first model until the generated second material list meets the recommendation condition or the number of times the second material list is generated reaches a preset number threshold; When the second material list meets the recommendation condition corresponding to the prompt template, multiple materials in the second material list are displayed.

4. The material recommendation method according to claim 3, characterized in that: The prompt template includes a first prompt template, the first prompt template corresponds to a first recommendation condition, and the verifying the second material list includes: Embedding the user and each item in the second item list to obtain a first vector for the user and a second vector for each item in the second item list; Calculate a first similarity between the user and each material according to the first vector and the second vector of each material; Counting the number of first materials whose first similarity is greater than or equal to a first threshold; When a first ratio of the first material quantity to the total number of materials in the second material list is greater than a second threshold, it is determined that the second material list meets the first recommendation condition; when the first ratio is less than or equal to the second threshold, it is determined that the second material list does not meet the first recommendation condition.

5. The material recommendation method according to claim 3, characterized in that: The prompt template includes a second prompt template, the second prompt template corresponds to a second recommendation condition, and the verifying the second material list includes: Determine a first material category set preferred by the user based on user behavior statistical data; Obtain a second material type set included in the second material list; Calculating a second similarity between the user and the second material list according to an intersection-and-union ratio of the first material category set and the second material category set; When the second similarity is less than a third threshold, it is determined that the second material list meets the second recommendation condition; when the second similarity is greater than or equal to the third threshold, it is determined that the second material list does not meet the second recommendation condition.

6. A material recommendation device, characterized in that: include: An acquisition unit, configured to acquire first characteristic data from an application program, wherein the application program includes a material to be recommended; The acquisition unit is further used to acquire a first material list, where the first material list is obtained through the first arrangement process; a processing unit, configured to determine second characteristic data according to the first characteristic data and the first material list; The processing unit is further used to generate a second material list according to the second characteristic data and the prompt template corresponding to the material type through the first model.

7. The material recommendation device according to claim 6, characterized in that: The processing unit is specifically used for: Determine the output target and output format according to the prompt template; constructing a generation instruction according to the second characteristic data, the output target and the output format; The generation instruction is input into the first model, and the second material list conforming to the output format is generated through the first model.

8. The material recommendation device according to claim 6 or 7, characterized in that: After generating the second material list, the processing unit is further used for: Verifying the second material list; If the second material list does not meet the recommendation condition corresponding to the prompt template, regenerate the second material list through the first model until the generated second material list meets the recommendation condition or the number of times the second material list is generated reaches a preset number threshold; The material recommendation device further includes: The display unit is used to display multiple materials in the second material list when the second material list meets the recommendation condition corresponding to the prompt template.

9. The material recommendation device according to claim 8, characterized in that: The prompt template includes a first prompt template, the first prompt template corresponds to a first recommendation condition, and the processing unit is specifically configured to: Embedding the user and each item in the second item list to obtain a first vector for the user and a second vector for each item in the second item list; Calculate a first similarity between the user and each material according to the first vector and the second vector of each material; Counting the number of first materials whose first similarity is greater than or equal to a first threshold; When a first ratio of the first material quantity to the total number of materials in the second material list is greater than a second threshold, it is determined that the second material list meets the first recommendation condition; when the first ratio is less than or equal to the second threshold, it is determined that the second material list does not meet the first recommendation condition.

10. The material recommendation device according to claim 8, characterized in that: The prompt template includes a second prompt template, the second prompt template corresponds to a second recommendation condition, and the processing unit is specifically configured to: Determine a first material category set preferred by the user based on user behavior statistical data; Obtain a second material type set included in the second material list; Calculating a second similarity between the user and the second material list according to an intersection-and-union ratio of the first material category set and the second material category set; When the second similarity is less than a third threshold, it is determined that the second material list meets the second recommendation condition; when the second similarity is greater than or equal to the third threshold, it is determined that the second material list does not meet the second recommendation condition.

Images