Topic content recommendation method, electronic device and storage medium based on large model

Abstract

The embodiments of the present invention provide a topic content recommendation method, system, electronic device and storage medium based on a big model. The method includes: collecting user behavior data from multiple data sources in real time; analyzing the behavior data from multiple data sources through an artificial intelligence big model to construct a real-time and dynamically updated user preliminary portrait of the user; verifying the user preliminary portrait, using the verified user preliminary portrait as the user's formal portrait, marking the user preliminary portrait that has not passed the verification with questions, and iteratively updating the artificial intelligence big model with the question marking; analyzing the user's multi-dimensional preferences based on the artificial intelligence big model and the user's formal portrait, determining the correlation structure of the preferences in each dimension, and obtaining personalized topic content groups to recommend to the user. The embodiment of the present invention uses a big model to perform in-depth analysis of user behavior in real time, so that the big model can perform self-correction and learning. Improve the performance of the big model and the recommendation effect.

Description

Technical Field

[0001] The present invention relates to the field of intelligent speech, and in particular to a topic content recommendation method, system, electronic device and storage medium based on a large model. Background Art

[0002] With the popularization of the Internet and the continuous advancement of information technology, a large amount of information is being produced, disseminated and accumulated rapidly. Whether it is news information, product information, entertainment content or academic resources, they are all showing an explosive growth trend. When faced with massive amounts of information, users find it difficult to quickly and accurately find the content they really need and are interested in, and information overload has become a common problem. As people's pace of life continues to accelerate, time has become more and more precious, and users hope to obtain the information that best meets their needs in the shortest time. Traditional information retrieval methods, such as search engines, require users to actively enter keywords to search. For some users who are not clear about their needs or lack search skills, it is often difficult to obtain satisfactory results. Therefore, users' demand for personalized and accurate information recommendation services is growing.

[0003] The main steps of the recommendation system are: collecting data, analyzing user behavior, building user portrait information by analyzing user interests, preferences, behavior habits and other features, storing the extracted features in the content feature repository, and providing content information for recommendation calculations. Data collection methods include: capturing relevant data from external data sources, interacting with the interface of internal systems or third-party platforms to obtain user behavior data and content data, and collecting user information by recording user operations in the recommendation system, such as browsing, clicking, and purchasing. The construction of user portraits can generate a series of tags for users through the results of user behavior analysis to describe user characteristics. Recommendation calculation module: Contains various recommendation algorithms (such as collaborative filtering, content-based recommendations, association rule recommendations, etc.), and performs recommendation calculations based on user portraits and content features.

[0004] Specifically, the recommended solutions in this field are mainly the following: 1. Collaborative filtering solution By leveraging the past behavior of existing user groups, we can recommend content that target users have never seen but may be interested in, helping them discover their potential interests and preferences. For example, on a music platform, if a user with similar interests likes a song that you have never heard of, the system may recommend that song to you.

[0005] 2. Content-based recommendation solution Content-based recommendation is the continuation and development of information filtering technology. It makes recommendations based on the content information of the project. The recommendation results are clearly explanatory, and users can clearly understand why they receive such recommendations.

[0006] 3. Recommendation scheme based on association rules The associations between different items can be discovered from a large amount of user behavior data, so as to recommend related products or content to users. The recommendation results can be explained by the mined association rules, so that users can understand the basis of the recommendation. For example, if it is found on an e-commerce platform that users who buy mobile phones often also buy mobile phone cases, mobile phone cases can be recommended to users who buy mobile phones.

[0007] 4. Knowledge-based recommendation solutions A recommendation solution based on knowledge content is a method that uses knowledge representation methods such as knowledge graphs to evaluate the similarity between items that the user has not seen and items that the current user liked in the past, and recommends relevant knowledge content to the user based on the user's interests and needs.

[0008] In the process of implementing the present invention, the inventors found that there are at least the following problems in the related art: 1. User-based collaborative filtering solution (1) Poor real-time performance. When a user’s interests change, a certain amount of new behavior data needs to be accumulated to reflect this change, resulting in poor real-time performance of recommendations.

[0009] (2) Not universal: In practical applications, the intersection of different users’ behaviors is often small, leading to data sparsity. For example, in some niche interest areas, the number of users is small and the behavioral data is insufficient, making it difficult to find enough similar users for accurate recommendations.

[0010] (3) For new users, it is difficult to find other users who are similar to them because there is not enough historical behavior data. Therefore, the recommendation effect is not good when new users just register. 2. Content-based recommendation solution (1) Difficulty in feature extraction: It is necessary to extract meaningful features from the attributes of items. If the feature extraction is inaccurate or incomplete, it will seriously affect the recommendation effect. For example, for some text content, how to accurately extract features such as keywords and topics is a challenge.

[0011] (2) Poor real-time performance: Over-reliance on users’ past behaviors and preferences while ignoring possible changes in users’ interests, and lack of adaptability to the evolution of users’ interests.

[0012] 3. Recommendation scheme based on association rules (1) Rule maintenance is difficult: The core of the algorithm is the discovery of association rules. This process requires a lot of calculations and data processing, and the computational complexity is high, which is the bottleneck of the algorithm. (2) Poor real-time performance: Over-reliance on users’ past behaviors and preferences while ignoring possible changes in users’ interests, and lack of adaptability to the evolution of users’ interests.

[0013] 4. Knowledge-based recommendation solutions (1) Difficulty in knowledge maintenance: A lot of effort is required to maintain the knowledge base, and content updates may not be timely. (2) Lack of diversity may affect user experience: it is completely dependent on the content of the knowledge base and it is difficult to discover users' potential interests.

[0014] In general, the defects in the prior art include: Not universal: The defect is that the intersection of behaviors of different users is often small, resulting in data sparsity problem.

[0015] Poor real-time performance: Due to over-reliance on users' past behaviors and preferences, it ignores possible changes in users' interests and lacks adaptability to the evolution of users' interests. As a result, it is unable to tap into users' potential interest preferences and is unable to adapt to changes in interests.

[0016] Lack of diversity: Technologies in this field usually spend a lot of effort on data collection and user behavior analysis. Data maintenance is difficult, and the analysis is based on past user behaviors, which may result in a lack of diversity in recommended content. Summary of the invention

[0017] In order to at least solve the problems of topic content recommendation in the prior art being not universal, poor in real time and lacking in diversity.

[0018] In a first aspect, an embodiment of the present invention provides a topic content recommendation method based on a large model, comprising: Collect user behavior data from multiple data sources in real time, where the user behavior from multiple data sources includes: browsing history, search history, click behavior, dwell time, purchase history, and social interaction; The behavioral data from the multiple data sources are analyzed by the artificial intelligence big model according to the dimensions including usage scenarios, time series and usage frequency, so as to construct a user profile of the user that is dynamically updated in real time; Verifying the user's preliminary portrait, taking the verified preliminary portrait as the user's formal portrait, marking problems for the preliminary portrait that fails the verification, and iteratively updating the artificial intelligence big model using the problem markings; Based on the artificial intelligence big model and the user's formal portrait, the user's multi-dimensional preferences are analyzed, the correlation structure of the preferences in each dimension is determined, and a personalized topic content group is obtained to recommend to the user.

[0019] In a second aspect, an embodiment of the present invention provides a topic content recommendation system based on a large model, comprising: A data collection module is used to collect user behavior data from multiple data sources in real time, wherein the user behavior from multiple data sources includes: browsing history, search history, click behavior, dwell time, purchase history, and social interaction; A portrait determination module is used to analyze the behavior data of the multiple data sources according to the dimensions including usage scenarios, time series and usage frequency through an artificial intelligence big model, and construct a user preliminary portrait of the user that is dynamically updated in real time; A verification module, used to verify the user's preliminary portrait, use the verified user's preliminary portrait as the user's formal portrait, mark problems for the user's preliminary portrait that fails the verification, and iteratively update the artificial intelligence big model using the problem marking; The recommendation module is used to analyze the multi-dimensional preferences of the user based on the artificial intelligence big model and the user's formal portrait, determine the correlation structure of the preferences in each dimension, and obtain personalized topic content groups to recommend to the user.

[0020] According to a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the large model-based topic content recommendation method of any embodiment of the present invention.

[0021] In a fourth aspect, an embodiment of the present invention provides a storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the steps of the large model-based topic content recommendation method of any embodiment of the present invention are implemented.

[0022] In a fifth aspect, an embodiment of the present invention provides a computer program product, including a computer program / instructions, characterized in that when the computer program / instructions are executed by a processor, the steps of the large model-based topic content recommendation method of any embodiment of the present invention are implemented.

[0023] The beneficial effects of the embodiments of the present invention are as follows: the multi-dimensional data collection and analysis of this method can be applied to more comprehensive recommendations, and the big model is used to conduct in-depth analysis of user behavior in real time, combining multiple factors such as the time series, frequency, and context of the behavior to build a multi-dimensional user portrait. The big model can be self-corrected and learned. Through continuous iteration and optimization, the diversity, accuracy, and reliability of the user portrait constructed by the big model can be improved. The construction and integration of user portraits can be optimized to improve the performance of the big model and the recommendation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 is a flow chart of a topic content recommendation method based on a large model provided by an embodiment of the present invention; Figure 2 This is a specific implementation flow chart of a topic content recommendation method based on a large model provided by an embodiment of the present invention; Figure 3 It is a structural diagram of a topic content recommendation system based on a large model provided by an embodiment of the present invention; Figure 4 A schematic structural diagram of an embodiment of an electronic device for recommending topic content based on a large model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] like Figure 1 The flowchart of a topic content recommendation method based on a large model provided by an embodiment of the present invention includes the following steps: S11: Collecting user behavior data from multiple data sources in real time, wherein the user behavior from the multiple data sources includes: browsing history, search history, click behavior, dwell time, purchase history, and social interaction; S12: Analyze the behavior data from multiple data sources according to the dimensions of usage scenarios, time series, and usage frequency through an artificial intelligence big model to construct a user profile of the user that is dynamically updated in real time; S13: Verify the user's preliminary portrait, use the verified preliminary portrait as the formal portrait of the user, mark problems for the preliminary portrait that fails the verification, and iteratively update the artificial intelligence model using the problem markings; S14: Analyze the multi-dimensional preferences of the user based on the artificial intelligence big model and the formal portrait of the user, determine the correlation structure of the preferences in each dimension, and obtain a personalized topic content group to recommend to the user.

[0028] In this embodiment, it is considered that the content recommendation of the prior art is a combination of multiple recommendation algorithms, such as combining content-based recommendation and collaborative filtering. Content-based recommendation can make recommendations based on the user's unique preference characteristics, and collaborative filtering can make recommendations based on the behavior data of other users. By combining the results of the two methods, the accuracy of recommendations for users with unique preferences can be improved. A more complete user feedback mechanism is established to encourage users with unique preferences to provide feedback on the recommendation results, such as likes, collections, comments, and ratings. The recommendation strategy is adjusted in a timely manner according to user feedback to improve the accuracy and satisfaction of the recommendation. However, this method needs to consider how to combine multiple recommendation algorithms, and the more recommendation algorithms are combined, the higher the difficulty cost index of later maintenance, and maintenance is more difficult.

[0029] Therefore, in order to avoid the combination and superposition of algorithms, this application constructs another content recommendation method, which uses the artificial intelligence big model LLM (hereinafter referred to as the big model) to automatically build user portraits. Through the continuous learning of the big model, it analyzes user behavior, automatically builds user portraits, extracts user behavior characteristics, and explores users' potential interests and hobbies; combined with user trajectories and retrieved real-time messages, it provides users with more accurate content recommendations.

[0030] The intention of this method is to achieve efficient data collection and user portrait analysis based on big models through big model technology, and to generate personalized recommended content in real time, ensuring that users can obtain the latest and most relevant content in every interaction. This solves the problems of content recommendation being non-universal, poor in real time, and lacking in diversity in existing technologies.

[0031] For step S11, data collection is the basis of this method. In order to recommend topic content, this application not only selects conventional data sources, such as user browsing records, click records, social media data, etc. At the same time, interactive data such as voice interaction records and smart device usage records are further introduced. In other words, this application not only focuses on the user's own data, but also focuses on the interaction data between the user and other devices, laying the foundation for building multi-dimensional preferences in subsequent steps. Among them, the collection of data can utilize the real-time processing capabilities of large models to ensure that data can be collected and processed in real time, thereby improving the response speed of the system. Taking into account the privacy of users, when using their own data, users are unwilling to transmit their personal regular data to the server. Therefore, when users refuse to share regular data, the interactive data is obtained by the interaction between the user and the server, so the server uses it as the basis for user portraits.

[0032] After obtaining the data, perform data cleaning and preprocessing: While collecting data, use the big model to perform preliminary data cleaning and preprocessing to ensure the accuracy of the early data quality. The subsequent big model can build a more accurate user portrait in the self-service learning process, laying the foundation for subsequent analysis.

[0033] For step S12, user portrait is the core of the recommendation system. This method uses large model technology to achieve more sophisticated and accurate user portrait construction.

[0034] By using the natural language processing and multimodal processing capabilities of the big model, we can integrate browsing records, click records, social media data, voice interaction records, and smart device usage records. Since the sources of data are too broad and difficult to integrate, we need to analyze these data sources separately according to usage scenarios, time series, and usage frequency.

[0035] For example, for conventional data: users’ browsing history and click history can be used to query what content users like to browse and what content they are more inclined to click on through usage scenarios, thereby analyzing the user’s preference interest portrait; and time series can indicate what users click on and when. For example, from 7 to 9 a.m., users usually like to click on news content. From 7 to 10 p.m., users usually like to click on entertainment content. Therefore, even if users have preferences, there are differences in usage time. Furthermore, the frequency of use can further highlight the frequency of use of each preference interest by users, and determine the priority of each preference interest of users. The dimensions of the preferences include: interest preference, consumption preference, age level, gender, region, and occupation.

[0036] Similarly, the interaction data can also reflect the user's preferences and interests. Even if the user does nothing, it can be reflected that the user is not interested in voice interaction. Marking the user's preferences and interests and not liking interaction can also be used to adjust the frequency of subsequent recommendations. For example, build a comprehensive preliminary portrait of user A: during the day - news, music; at night - entertainment, radio; news> entertainment> music> radio (at the same time, music and radio have a certain correlation); the user does not like voice interaction, and the frequency of content recommendation decreases. As users continue to use it, the data source is constantly updated, and the user's preliminary portrait is updated in real time. In general, the continuous learning ability of the large model is used to automatically analyze user behavior, interests and preferences. The user portrait should be updated and optimized in real time as the user behavior changes. Through user interaction and feedback data, the user portrait is automatically adjusted and optimized to ensure the accuracy and relevance of the portrait.

[0037] For step S13, after obtaining the user's preliminary portrait, further verification can be performed. For example, a large model can be used to screen and verify by setting rules. Considering the accuracy rate, it can also be manually verified, configured by the developer, or the user can build his own portrait. However, it should be noted that it does not mean that the user's preliminary portrait cannot be used without verification.

[0038] Two scenarios are considered here. The first is to base precise recommendations on actual needs. The other is to consider overly accurate portraits. Although the accuracy of the recommendations is guaranteed, the recommended content is too limited. For example, although the user likes to listen to Zhou's music, the use of the user's formal portrait can ensure that the recommended content is Zhou's music. However, the user will get tired of listening to it over and over again. Therefore, the user's preliminary portrait can also be used to occasionally recommend some other content appropriately, but the frequency is lower than that of the first scenario.

[0039] As an implementation method, the problem annotation is used to iteratively update the user portrait construction strategy of the artificial intelligence big model. In this implementation method, based on accuracy, the user preliminary portrait that has passed the verification is used as the user's formal portrait, and the user preliminary portrait that has not passed the verification is annotated with problems (AI annotation, developer annotation, user's own annotation, etc.). The artificial intelligence big model uses the annotation as training data and iteratively updates it, so that the user portrait construction strategy is more accurate when the portrait is constructed next time.

[0040] For step S14, personalized topic recommendation is achieved through the user's formal portrait and real-time data generated by the large model. Specifically, it includes: Obtaining user behavior context information of the user from the formal user profile; Using the artificial intelligence big model to call the search engine to obtain real-time information on the behavior trajectory of the user behavior context information; The association structure of preferences in each dimension is predicted through the behavior trajectory, and a personalized topic content group is obtained to recommend to the user.

[0041] The user's formal portrait is analyzed through a large model to determine the user's behavioral context information, such as what videos did you watch yesterday and what news did you click on in the morning. Further, the user's behavioral trajectory is determined through context information. The user's entire behavioral trajectory from morning to night is determined. This is used for: for example, when recommending content to users in the morning, the content recommended is the morning content in the behavioral trajectory. It is not just based on the user's preferences, randomly recommending related content. Through the behavioral trajectory, it is determined that the user likes news in the morning period (news is associated with morning TV news and night radio news) and entertainment-related content in the night period (which singer does the specific user like? What kind of songs do you like will not be repeated here.) are also associated. Since there is a huge amount of content in the data, the association structure of the user's preferences in various dimensions is constructed at this time. The user's information is brought in (time, historical data, or the content currently input by the user) to obtain a personalized topic content group to recommend to the user.

[0042] As an implementation manner, after the personalized topic content group is recommended to the user, the method further includes: Based on the user's feedback on the recommendation, the topic recommendation strategy of the artificial intelligence big model is self-corrected.

[0043] The user's feedback on the recommendation includes: active feedback and passive feedback, wherein the passive feedback includes: click rate and stay time.

[0044] In this embodiment, if the user actively gives feedback on the recommendation, such as clicking on the content or responding with voice, then it can be said that the recommendation is effective. Similarly, the user's passive feedback can also be collected. For example, if the user stays on the recommended content for a long time, it can be said that the recommendation is also effective. Therefore, it is not necessarily that the user has done something in response to the recommendation. Even if the user enjoys the recommended content itself, it can be regarded as passive feedback. If the feedback shows that the user does not like it, the topic recommendation strategy will be self-corrected to ensure that the next recommendation can make the user more interested, thereby improving the recommendation accuracy and user satisfaction.

[0045] As an embodiment, the method further includes: Based on the artificial intelligence big model, the user preliminary portrait and the user formal portrait, the multi-dimensional preference of the user is analyzed, the correlation structure of the preferences in each dimension is determined, and a personalized complex topic content group is obtained to recommend to the user.

[0046] This method further uses the user's preliminary profile, which has a high degree of overlap with the user's formal profile in some content, and can determine the user's most accurate preference interests. Content that does not overlap can be used as an extended topic content recommendation with low frequency.

[0047] In general, the embodiments of the present method are as follows Figure 2 As shown, the steps are as follows: 1. User behavior data collection Comprehensively collect user behavior data through multiple channels and methods, including but not limited to user browsing history, search history, click behavior, dwell time, purchase history, social interaction, etc. Ensure the accuracy, completeness and timeliness of the data, and provide a solid foundation for subsequent analysis and portrait construction.

[0048] (II) Use big models to conduct in-depth analysis of user behavior and build accurate user portraits We use advanced big model technology to conduct in-depth analysis of the collected user behavior data. We not only consider the characteristics of individual behaviors, but also combine the time series, frequency, context and other factors of the behaviors to build a more comprehensive and accurate user portrait. The portrait content includes multiple dimensions such as user interests, consumption preferences, age level, gender, region, occupation, etc.

[0049] (III) Storing the generated user profiles in the preliminary database and conducting preliminary screening The constructed user portraits are stored in the preparatory database. At the same time, the portraits in the preparatory database are preliminarily screened to remove obviously unreasonable or abnormal data to improve the efficiency and accuracy of subsequent manual verification.

[0050] (IV) Strictly verify the generated user profile data manually to promote self-correction and learning of large models Strict manual verification is performed on the user portraits in the preparatory database. Reasonable data is stored in the official database; unreasonable data is marked with detailed problems and fed back to the big model for self-correction and learning. Through continuous iteration and optimization, the accuracy and reliability of the user portrait constructed by the big model are improved.

[0051] (V) Personalized content recommendations based on user profiles and relevant background knowledge to achieve highly relevant and accurate recommendations Combine user portraits and rich background knowledge to make personalized content recommendations. For example, if a user is following "The Strange Stories of the Tang Dynasty" recently, the search engine can integrate relevant information about "The Strange Stories of the Tang Dynasty" for recommendation. It can also recommend similar film and television works, novels, games, etc. based on the user's interests and historical behaviors. At the same time, considering the user's consumption preferences and regional factors, it recommends related peripheral products, offline activities, etc., to achieve all-round, highly correlated and accurate recommendations.

[0052] (VI) Collect user feedback on recommended content and continuously optimize the re-construction and integration of user portraits Actively collect user feedback on recommended content through various means, such as user clicks, browsing, collections, comments, ratings, etc. According to user feedback, timely adjust recommendation strategies and optimize the construction and integration of user portraits. At the same time, use user feedback as important training data to further improve the performance of large models and recommendation effects.

[0053] (VII) Regularly update and maintain user profiles and recommendation systems As user behavior continues to change and new data continues to be generated, user portraits are regularly updated and maintained. At the same time, the recommendation system is optimized and upgraded to ensure that users are always provided with the highest quality and most personalized content recommendation services.

[0054] It can be seen from this implementation that the multi-dimensional data collection and analysis of this method can be applied to more comprehensive recommendations. The big model is used to conduct in-depth analysis of user behavior in real time, and a multi-dimensional user profile is constructed by combining multiple factors such as the time series, frequency, and context of the behavior. The big model can be self-corrected and learned. Through continuous iteration and optimization, the diversity, accuracy, and reliability of the user profile constructed by the big model can be improved. The construction and integration of user profiles can be optimized to improve the performance of the big model and the recommendation effect.

[0055] like Figure 3 The figure is a schematic diagram of the structure of a topic content recommendation system based on a large model provided by an embodiment of the present invention. The system can execute the topic content recommendation method based on a large model described in any of the above embodiments and is configured in a terminal.

[0056] The present embodiment provides a topic content recommendation system 10 based on a large model, including: a data collection module 11, a portrait determination module 12, a verification module 13 and a recommendation module 14.

[0057] Among them, the data collection module 11 is used to collect user behavior data from multiple data sources in real time, wherein the user behavior from the multiple data sources includes: browsing history, search history, click behavior, dwell time, purchase record, and social interaction; the portrait determination module 12 is used to analyze the behavior data from the multiple data sources according to the dimensions including: usage scenario, time series, and usage frequency through the artificial intelligence big model, and construct a real-time dynamically updated user preliminary portrait of the user; the verification module 13 is used to verify the user preliminary portrait, use the verified user preliminary portrait as the user's formal portrait, mark the user preliminary portrait that has not passed the verification with problems, and iteratively update the artificial intelligence big model using the problem marking; the recommendation module 14 is used to analyze the multi-dimensional preferences of the user based on the artificial intelligence big model and the user's formal portrait, determine the correlation structure of the preferences in each dimension, and obtain personalized topic content groups for recommendation to the user.

[0058] The embodiment of the present invention further provides a non-volatile computer storage medium, the computer storage medium stores computer executable instructions, and the computer executable instructions can execute the topic content recommendation method based on a large model in any of the above method embodiments; As an implementation mode, the non-volatile computer storage medium of the present invention stores computer executable instructions, and the computer executable instructions are configured as follows: Collect user behavior data from multiple data sources in real time, where the user behavior from multiple data sources includes: browsing history, search history, click behavior, dwell time, purchase history, and social interaction; The behavioral data from the multiple data sources are analyzed by the artificial intelligence big model according to the dimensions including usage scenarios, time series and usage frequency, so as to construct a user profile of the user that is dynamically updated in real time; Verifying the user's preliminary portrait, taking the verified preliminary portrait as the user's formal portrait, marking problems for the preliminary portrait that fails the verification, and iteratively updating the artificial intelligence big model using the problem markings; Based on the artificial intelligence big model and the user's formal portrait, the user's multi-dimensional preferences are analyzed, the correlation structure of the preferences in each dimension is determined, and a personalized topic content group is obtained to recommend to the user.

[0059] As a non-volatile computer-readable storage medium, it can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the method in the embodiment of the present invention. One or more program instructions are stored in the non-volatile computer-readable storage medium, and when executed by the processor, the topic content recommendation method based on the large model in any of the above method embodiments is executed.

[0060] Figure 4 FIG. 1 is a schematic diagram of the hardware structure of an electronic device of a topic content recommendation method based on a large model provided by another embodiment of the present application, such as Figure 4 As shown, the device includes: One or more processors 410 and memory 420, Figure 4 A processor 410 is taken as an example. The apparatus of the topic content recommendation method based on a large model may further include: an input device 430 and an output device 440.

[0061] The processor 410, the memory 420, the input device 430 and the output device 440 may be connected via a bus or other means. Figure 4 The example of connecting through bus is taken in the following.

[0062] The memory 420, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the topic content recommendation method based on a large model in the embodiment of the present application. The processor 410 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 420, that is, the topic content recommendation method based on a large model in the above method embodiment is implemented.

[0063] The memory 420 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data, etc. In addition, the memory 420 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 420 may optionally include a memory remotely arranged relative to the processor 410, and these remote memories may be connected to the mobile device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0064] The input device 430 can receive input digital or character information. The output device 440 can include a display device such as a display screen.

[0065] The one or more modules are stored in the memory 420, and when executed by the one or more processors 410, the topic content recommendation method based on the large model in any of the above method embodiments is executed.

[0066] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.

[0067] The non-volatile computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the device, etc. In addition, the non-volatile computer-readable storage medium may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the non-volatile computer-readable storage medium may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0068] An embodiment of the present invention also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the large model-based topic content recommendation method of any embodiment of the present invention.

[0069] The electronic device of the embodiment of the present application exists in various forms, including but not limited to: (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.

[0070] (2) Ultra-mobile personal computer devices: These devices fall into the category of personal computers, have computing and processing capabilities, and generally also have mobile Internet access features. These terminals include: PDAs, MIDs, and UMPC devices, such as tablet computers.

[0071] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0072] (4) Other electronic devices with data processing functions.

[0073] In this article, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include" and "comprise" include not only those elements, but also other elements not explicitly listed, or also include elements inherent to such processes, methods, articles or equipment. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the existence of other identical elements in the process, method, article or equipment that includes the elements.

[0074] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0075] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A topic content recommendation method based on a large model, comprising: Collect user behavior data from multiple data sources in real time, where the user behavior from multiple data sources includes: browsing history, search history, click behavior, dwell time, purchase history, and social interaction; The behavioral data from the multiple data sources are analyzed by the artificial intelligence big model according to the dimensions including usage scenarios, time series and usage frequency, so as to construct a user profile of the user that is updated dynamically in real time; Verifying the user's preliminary portrait, taking the verified preliminary portrait as the formal portrait of the user, marking problems for the preliminary portrait that fails the verification, and iteratively updating the artificial intelligence big model using the problem markings; Based on the artificial intelligence big model and the user's formal portrait, the user's multi-dimensional preferences are analyzed, the correlation structure of the preferences in each dimension is determined, and a personalized topic content group is obtained to recommend to the user.

2. The method according to claim 1, wherein: The marking of problems for the user preliminary portraits that have not passed the verification, and iteratively updating the artificial intelligence big model using the problem markings include: Iteratively updating the user portrait construction strategy of the artificial intelligence large model by using the question annotation; After the personalized topic content group is recommended to the user, the method further includes: Based on the user's feedback on the recommendation, the topic recommendation strategy of the artificial intelligence big model is self-corrected.

3. The method according to claim 1, wherein: After marking the problem of the user preliminary portrait that has not passed the verification, the method further includes: Based on the artificial intelligence big model, the user preliminary portrait and the user formal portrait, the multi-dimensional preference of the user is analyzed, the correlation structure of the preferences in each dimension is determined, and a personalized complex topic content group is obtained to recommend to the user.

4. The method according to claim 1, wherein: The dimensions of the preferences include: interest preference, consumption preference, age level, gender, region, and occupation.

5. The method according to claim 1, wherein: The multi-dimensional preference analysis of the user based on the artificial intelligence big model and the user formal portrait includes: Obtaining user behavior context information of the user from the formal user profile; Using the artificial intelligence big model to call the search engine to obtain real-time information on the behavior trajectory of the user behavior context information; The association structure of preferences in each dimension is predicted through the behavior trajectory, and a personalized topic content group is obtained to recommend to the user.

6. The method according to claim 2, wherein: The user's feedback on the recommendation includes: active feedback and passive feedback, wherein the passive feedback includes: click rate and stay time.

7. A topic content recommendation system based on a large model, comprising: A data collection module is used to collect user behavior data from multiple data sources in real time, wherein the user behavior from multiple data sources includes: browsing history, search history, click behavior, dwell time, purchase history, and social interaction; A portrait determination module is used to analyze the behavior data of the multiple data sources according to the dimensions including usage scenarios, time series and usage frequency through an artificial intelligence big model, and construct a user preliminary portrait of the user that is dynamically updated in real time; A verification module, used to verify the user's preliminary portrait, use the verified user's preliminary portrait as the user's formal portrait, mark problems for the user's preliminary portrait that fails the verification, and iteratively update the artificial intelligence big model using the problem marking; The recommendation module is used to analyze the multi-dimensional preferences of the user based on the artificial intelligence big model and the user's formal portrait, determine the correlation structure of the preferences in each dimension, and obtain personalized topic content groups to recommend to the user.

8. A storage medium having a computer program product stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

9. A computer program product having instructions embedded on a storage medium, wherein the instructions implement the steps of the method according to any one of claims 1 to 6.

10. An electronic device, comprising: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the method described in any one of claims 1 to 6.

Images