Method and system for extracting and generating social media topics based on large model retrieval
Through the method based on big model retrieval, the problem of the lack of priori ability and knowledge integration ability of the algorithm in the existing social media topic refining scheme is solved, and in-depth analysis of social media topics and refining background information is achieved, and user behavior decisions are provided. It is recommended to ensure the continuous optimization and accuracy of the algorithm.
Patent Information
- Application Number
- CN202411408905.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-10
- Publication Date
- 2025-02-07
AI Technical Summary
The existing social media topic refining schemes have problems such as the lack of prior capabilities of algorithms, complex models, lack of knowledge integration capabilities, and neglecting social media background information, which leads to lag, unstable and difficult to maintain.
Using a method based on big model retrieval, social media data is obtained regularly for sentiment analysis, theme extraction and triple extraction, aggregate data to generate background analysis results, and update the iterative knowledge base to optimize the analysis results, and finally based on background analysis and public opinion factors Generate marketing decision recommendations.
Through the knowledge base, the big model can be enhanced, in-depth analysis of social media topics and the refinement of key background information, and user behavior decision-making suggestions are provided to ensure the continuous optimization and accuracy of the algorithm.
Smart Images
Figure CN119397009A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for extracting and generating social media topics based on large model retrieval. Background Art
[0002] Social media topic extraction refers to the process of automatically or semi-automatically extracting core topics, trends, emotions, opinions or specific background information from a large amount of user-generated content on social media platforms (such as Weibo, Tik Tok, WeChat, etc.).
[0003] Existing social media topic extraction solutions include a social media collection module, which is used to obtain hot search data from social media platforms and search engines through API interfaces and crawler tools; a data preprocessing module, which is used to perform text processing such as denoising, filtering anomalies, and word segmentation on network data, and store it in structured forms such as feature matrices and inverted indexes; an algorithm analysis module, which is used to build semantic features such as entities, topics, and vector representations around content texts, build user attribute features based on user portraits and content browsing behaviors, and construct implicit features between content collections through clustering models, graph models, and time series models; a business decision module, which is used to combine multi-dimensional algorithm features, evaluate content credibility, identify community dynamics, predict business opportunities, and build marketing strategies through decision trees.
[0004] The shortcomings of existing technologies are: 1. The algorithm lacks prior capabilities: the current solution mainly relies on supervised features. Social media covers a wide range of topics and is highly time-sensitive, making it difficult to maintain the timeliness of the supervised model (frequent data annotation is required to update the model, and the effect is delayed and unstable); 2. The number of algorithm models is complex: different features and different tasks require dedicated models, resulting in too many system modules, greatly increasing the difficulty of maintenance and updating; 3. Lack of knowledge fusion capabilities: social media topics imply external knowledge, which cannot be covered by semantic and behavioral features alone, and additional knowledge base modules are needed to enhance the system; 4. Ignoring the importance of social media background information: analysis and decisions made without the social media background are often too generalized and not referenceable in specific scenarios, while existing semantic and behavioral feature engineering cannot effectively cover social media background information, and requires additional interpretation and generation modules to complete. Summary of the invention
[0005] The purpose of the present invention is to provide a method and system for extracting and generating social media topics based on large model retrieval to solve the problems raised in the above background technology.
[0006] To achieve the above object, the present invention provides the following technical solutions:
[0007] A method for extracting and generating social media topics based on large model retrieval, the method comprising:
[0008] Regularly obtain social media data within the authorized platform; the social media data includes but is not limited to the text, comments and interactive information posted by users, and the interactive information includes the number of likes and shares;
[0009] Based on the big model, sentiment analysis, topic extraction and triple extraction are performed on social media data. There are three types of sentiment analysis results: positive, negative and neutral. The results of topic extraction are keywords, and the results of triple extraction are subject-predicate-object triple structures.
[0010] Aggregate social media data based on sentiment analysis results, topic extraction results, and triple extraction results, input into a large model, and generate background analysis results on specific topics;
[0011] Based on the background analysis results, the knowledge base is updated and iterated, the background analysis results are optimized based on the knowledge base, and the causes of the background analysis results are predicted simultaneously to obtain public opinion factors;
[0012] With the help of big models, marketing decision recommendations are generated based on background analysis results and public opinion factors.
[0013] As a further solution of the present invention: the steps of performing sentiment analysis, topic extraction and triple extraction on social media data based on the big model include:
[0014] Perform sentiment analysis on the acquired social media data according to the big model to identify sentiment data in each social media data, wherein the sentiment data includes positive sentiment, negative sentiment and neutral sentiment;
[0015] Extract topics from the acquired social media data and determine the discussion topics of each social media data;
[0016] Extract the subject-predicate-object triple structure from each social media data;
[0017] The extracted sentiment data, discussion topics, and triple structures are correlated.
[0018] As a further solution of the present invention: the steps of performing sentiment analysis, topic extraction and triple extraction on social media data based on the large model also include:
[0019] Obtain the sentiment data from each social media data, sort it by time, and get the sentiment change trend;
[0020] Obtain discussion topics for each social media data as public opinion keywords;
[0021] The interaction information of each social media data is obtained, the interaction amount is calculated according to the number of likes and shares in the interaction information, and the social media data corresponding to the maximum interaction amount is selected as the most popular social media data.
[0022] As a further solution of the present invention: the method further comprises:
[0023] Based on the self-learning ability of the large model, blind spots in the knowledge base are regularly identified, and the knowledge base is updated and iterated based on the blind spots; the blind spots are information not covered by the knowledge base.
[0024] As a further solution of the present invention: the step of generating marketing decision suggestions based on the background analysis results and public opinion factors by means of the big model includes:
[0025] Read the background analysis results and public opinion factors as context;
[0026] Identify context based on prior knowledge in the big model and generate marketing decision recommendations;
[0027] Open the intent receiving port, receive the intent uploaded by the user based on the intent receiving port, and filter and update the marketing decision suggestions according to the received intent.
[0028] The technical solution of the present invention also provides a system for extracting and generating social media topics based on large model retrieval, the system comprising:
[0029] A social media data acquisition module, used to periodically acquire social media data within the authorized platform; the social media data includes but is not limited to the text, comments and interactive information posted by users, and the interactive information includes the number of likes and shares;
[0030] The social media data recognition module is used to perform sentiment analysis, topic extraction, and triple extraction on social media data based on a large model. There are three types of sentiment analysis results: positive, negative, and neutral. The result of topic extraction is a keyword, and the result of triple extraction is a subject-predicate-object triple structure.
[0031] The social media data aggregation module is used to aggregate social media data based on sentiment analysis results, topic extraction results, and triple extraction results, input into the big model, and generate background analysis results on specific topics;
[0032] A cause prediction module is used to update the iterative knowledge base based on the background analysis results, optimize the background analysis results based on the knowledge base, and simultaneously predict the causes of the background analysis results to obtain public opinion factors;
[0033] The suggestion generation module is used to generate marketing decision suggestions based on background analysis results and public opinion factors with the help of a large model.
[0034] As a further solution of the present invention: the social media data identification module includes:
[0035] An emotion recognition unit is used to perform emotion analysis on the acquired social media data according to the large model, and identify the emotion data in each social media data, wherein the emotion data includes positive emotion, negative emotion and neutral emotion;
[0036] A topic identification unit is used to extract topics from the acquired social media data and determine the discussion topic of each social media data;
[0037] A triple extraction unit, used to extract the subject-predicate-object triple structure from each social media data;
[0038] The information association unit is used to associate the extracted sentiment data, discussion topics and triple structures.
[0039] As a further solution of the present invention: the social media data identification module further includes:
[0040] The situation analysis unit is used to obtain the sentiment data in each social media data, sort them according to time, and obtain the sentiment change trend;
[0041] The public opinion analysis unit is used to obtain the discussion topics of each social media data as public opinion keywords;
[0042] The social media data selection unit is used to obtain the interactive information of each social media data, calculate the interactive amount according to the number of likes and shares in the interactive information, and select the social media data corresponding to the maximum interactive amount as the most popular social media data.
[0043] As a further solution of the present invention: the working content of the system also includes:
[0044] Based on the self-learning ability of the large model, blind spots in the knowledge base are regularly identified, and the knowledge base is updated and iterated based on the blind spots; the blind spots are information not covered by the knowledge base.
[0045] As a further solution of the present invention: the suggestion generation module includes:
[0046] A context reading unit, used for reading background analysis results and public opinion factors as contexts;
[0047] Prior recognition unit, used to identify context based on prior knowledge in the large model and generate marketing decision suggestions;
[0048] The intent receiving unit is used to open an intent receiving port, receive the intent uploaded by the user based on the intent receiving port, and filter and update the marketing decision suggestions according to the received intent.
[0049] Compared with the prior art, the present invention has the following beneficial effects: the present invention enhances the big model through the knowledge base, conducts in-depth analysis of social media topics through semantic understanding capabilities, extracts key background information, and provides user behavior decision suggestions based on this. At the same time, the big model is used to extract knowledge from implicit background information, so that the system has self-learning and knowledge base update functions, ensuring the continuous optimization and accuracy of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.
[0051] Figure 1 A flowchart of the method for extracting and generating social media topics based on large model retrieval. DETAILED DESCRIPTION
[0052] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0053] In an embodiment of the present invention, a method for extracting and generating social media topics based on large model retrieval is provided, the method comprising:
[0054] Regularly obtain social media data within the authorized platform; the social media data includes but is not limited to the text, comments and interactive information posted by users, and the interactive information includes the number of likes and shares;
[0055] Based on the big model, sentiment analysis, topic extraction and triple extraction are performed on social media data. There are three types of sentiment analysis results: positive, negative and neutral. The results of topic extraction are keywords, and the results of triple extraction are subject-predicate-object triple structures.
[0056] Aggregate social media data based on sentiment analysis results, topic extraction results, and triple extraction results, input into a large model, and generate background analysis results on specific topics;
[0057] Based on the background analysis results, the knowledge base is updated and iterated, the background analysis results are optimized based on the knowledge base, and the causes of the background analysis results are predicted simultaneously to obtain public opinion factors;
[0058] With the help of big models, marketing decision recommendations are generated based on background analysis results and public opinion factors.
[0059] The present invention provides a method for analyzing and summarizing the background of social media topics using a large model, aiming to enhance the large model through the knowledge base, conduct in-depth analysis of social media topics through semantic understanding capabilities, extract key background information, and provide user behavior decision suggestions based on this. At the same time, the large model is used to extract knowledge from implicit background information, so that the system has self-learning and knowledge base update functions, ensuring the continuous optimization and accuracy of the algorithm.
[0060] Specifically, the technical solution of the present invention is built around four core functions:
[0061] 1. Use large models for background analysis, including social media feature extraction, topic classification, and background summary generation (which can save the complexity of training multiple supervised models and produce more semantic and accurate social media data features);
[0062] 2. Update and iterate the knowledge base, extract knowledge based on background analysis, and automatically update the system knowledge base;
[0063] 3. Self-refinement and attribution analysis of the big model: through prompt engineering guidance and knowledge base / RAG enhancement, the big model can optimize or correct the content it generates, and make realistic guesses about the causes of the topic;
[0064] 4. Combining social media data characteristics and user intentions, the system provides accurate marketing and behavioral decision-making suggestions to help users achieve their goals.
[0065] Based on the core functions, the implementation of the entire system is as follows:
[0066] 1. The system triggers collection tasks regularly and continuously collects social media data from mainstream platforms;
[0067] 2. Perform two parts of preprocessing on social media data: text cleaning and large model extraction, and perform content statistical analysis based on the content extracted from the large model;
[0068] 3. In the algorithm part, the topic background is aggregated and initially analyzed, and then the big model is self-reflected and optimized. The background content is then refined through the big model to iterate the knowledge base. The background content and statistical features are then combined to allow the big model to determine the cause of the topic background, thereby helping to generate decision-making recommendations.
[0069] The instructions for each step are as follows:
[0070] The process of acquiring social media data mostly relies on calling Qingbo API and using crawler technology to collect data from mainstream social media platforms (such as Weibo, Xiaohongshu, Douyin, WeChat and Baidu); these data include but are not limited to various interactive information such as texts, comments, likes and shares posted by users; the process of acquiring social media data generally adopts a timed trigger method, and the real-time nature of the acquired social media data can be adjusted through the frequency, trying to ensure that the acquired social media data is relatively new data, so that this system can make highly timely analysis and decisions.
[0071] The steps of sentiment analysis, topic extraction and triple extraction of social media data based on the big model include:
[0072] Perform sentiment analysis on the acquired social media data according to the big model to identify sentiment data in each social media data, wherein the sentiment data includes positive sentiment, negative sentiment and neutral sentiment;
[0073] Extract topics from the acquired social media data and determine the discussion topics of each social media data;
[0074] Extract the subject-predicate-object triple structure from each social media data;
[0075] The extracted sentiment data, discussion topics, and triple structures are correlated.
[0076] Among them, the main way to associate the extracted sentiment data, discussion topics and triple structures is time association, which associates sentiment, topics and triples with relevant data within the time window to ensure that the evolution and development of the topic can be accurately tracked.
[0077] The steps of sentiment analysis, topic extraction and triple extraction of social media data based on the big model also include:
[0078] Obtain the sentiment data from each social media data, sort it by time, and get the sentiment change trend;
[0079] Obtain discussion topics for each social media data as public opinion keywords;
[0080] The interaction information of each social media data is obtained, the interaction amount is calculated according to the number of likes and shares in the interaction information, and the social media data corresponding to the maximum interaction amount is selected as the most popular social media data.
[0081] On the basis of the sentiment analysis, topic extraction and triple extraction that have been carried out, the sentiment analysis results, topic extraction results and triple extraction results are further limited, including statistical analysis of sentiment data, understanding the trend of sentiment changes, analyzing the main viewpoints and their distribution on social media, understanding the overall tendency of public opinion, and counting the discussion volume and interaction volume of different topics, etc., to help identify which topics are most concerned.
[0082] The process of aggregating social media data based on sentiment analysis results, topic extraction results, and triple extraction results, inputting into a large model, and generating background analysis results on a specific topic is as follows:
[0083] Aggregate the cleaned and preprocessed data features (such as sentiment, theme, and triples) to form a comprehensive set of social media data features, and then input the aggregated data into a large model to generate background analysis on a specific topic. This background analysis includes information such as the cause of the event, the development process, and the main participants, helping to deeply understand the background of the topic.
[0084] The process of updating the iterative knowledge base based on the background analysis results, optimizing the background analysis results based on the knowledge base, and simultaneously predicting the causes of the background analysis results to obtain public opinion factors includes:
[0085] Combining the knowledge base, original social media data, and background content generated in the previous step, the big model reflects on the generated analysis results to refine or correct inaccurate or incomplete content. Based on the self-reflection of the big model, the system will further refine new knowledge points and add these knowledge points to the knowledge base.
[0086] In addition, this system combines the statistical characteristics of emotions, opinions and topics to perform attribution analysis on the background generated by the large model to determine which public opinion factors are related to the generation of the topic. At the same time, it introduces multi-dimensional label statistical characteristics to further enrich the content of the background analysis, making the attribution of topics more accurate and diversified.
[0087] Furthermore, the above contents also include:
[0088] Based on the self-learning ability of the large model, blind spots in the knowledge base are regularly identified, and the knowledge base is updated and iterated based on the blind spots; the blind spots are information not covered by the knowledge base.
[0089] Through the self-learning ability of the large model, the system will identify blind spots or uncovered information in the knowledge base, and automatically update and iterate the knowledge base to ensure the comprehensiveness and timeliness of the system knowledge.
[0090] Finally, the steps of generating marketing decision suggestions based on the background analysis results and public opinion factors with the help of the big model include:
[0091] Read the background analysis results and public opinion factors as context;
[0092] Identify context based on prior knowledge in the big model and generate marketing decision recommendations;
[0093] Open the intent receiving port, receive the intent uploaded by the user based on the intent receiving port, and filter and update the marketing decision suggestions according to the received intent.
[0094] Based on the analysis results of background and public opinion factors, the system generates specific marketing decision recommendations to help companies or users formulate more targeted market strategies; the system can also generate behavioral decision recommendations that are in line with user intentions or interests based on the analysis content, such as content promotion strategies, social interaction strategies, etc.
[0095] It is worth mentioning that this system has created a new intent receiving port to support users to pass in their personal intentions or interests, and further customize decision suggestions to make them more in line with users' actual needs and goals.
[0096] Figure 1 The flowchart of the method for extracting and generating social media topics based on large model retrieval is as follows:
[0097] (1) Data collection: By default, the system collects social media data from mainstream platforms, including Weibo, Xiaohongshu, Douyin, WeChat, and Baidu, through the (Qingbo) API and crawlers.
[0098] (2) Data cleaning and preprocessing: The large model extracts the sentiment, topics, and triples of social media data and associates the data within the time window for summarization.
[0099] (3) Statistical feature preprocessing: content statistics based on dimensions such as sentiment, opinion, and theme.
[0100] (4) Topic background analysis and generation: Based on the features obtained in (2), social media data is aggregated and integrated into the input large model to generate topic background analysis.
[0101] (5) Self-reflection of the big model: Combine the knowledge base, original social media, and the generated content of (4) to allow the big model to reflect on itself in order to refine or correct the results of the previous step.
[0102] (6) Knowledge base update and iteration: Combine existing knowledge and historical content, use the big model to extract knowledge, and inspire the big model to find information not covered by the knowledge base.
[0103] (7) Background attribution analysis: Based on the background generated by (5), the statistical characteristics of the label dimension of (3) are introduced to determine the public opinion factors that generated the topic.
[0104] (8) Decision recommendations: Using the background and public opinion factors generated by (3), (5), and (7) as the context, and utilizing the rich prior knowledge of the large model, marketing decision recommendations or behavioral decision recommendations are generated, which can accept the personal intentions or interests passed in by users.
[0105] As a preferred embodiment of the technical solution of the present invention, the present invention further provides a system for extracting and generating social media topics based on large model retrieval, the system comprising:
[0106] A social media data acquisition module, used to periodically acquire social media data within the authorized platform; the social media data includes but is not limited to the text, comments and interactive information posted by users, and the interactive information includes the number of likes and shares;
[0107] The social media data recognition module is used to perform sentiment analysis, topic extraction, and triple extraction on social media data based on a large model. There are three types of sentiment analysis results: positive, negative, and neutral. The result of topic extraction is a keyword, and the result of triple extraction is a subject-predicate-object triple structure.
[0108] The social media data aggregation module is used to aggregate social media data based on sentiment analysis results, topic extraction results, and triple extraction results, input into the big model, and generate background analysis results on specific topics;
[0109] A cause prediction module is used to update the iterative knowledge base based on the background analysis results, optimize the background analysis results based on the knowledge base, and simultaneously predict the causes of the background analysis results to obtain public opinion factors;
[0110] The suggestion generation module is used to generate marketing decision suggestions based on background analysis results and public opinion factors with the help of a large model.
[0111] Furthermore, the social media data identification module includes:
[0112] An emotion recognition unit is used to perform emotion analysis on the acquired social media data according to the large model, and identify the emotion data in each social media data, wherein the emotion data includes positive emotion, negative emotion and neutral emotion;
[0113] A topic identification unit is used to extract topics from the acquired social media data and determine the discussion topic of each social media data;
[0114] A triple extraction unit, used to extract the subject-predicate-object triple structure from each social media data;
[0115] The information association unit is used to associate the extracted sentiment data, discussion topics and triple structures.
[0116] Specifically, the social media data identification module also includes:
[0117] The situation analysis unit is used to obtain the sentiment data in each social media data, sort them according to time, and obtain the sentiment change trend;
[0118] The public opinion analysis unit is used to obtain the discussion topics of each social media data as public opinion keywords;
[0119] The social media data selection unit is used to obtain the interactive information of each social media data, calculate the interactive amount according to the number of likes and shares in the interactive information, and select the social media data corresponding to the maximum interactive amount as the most popular social media data.
[0120] Furthermore, the working contents of the system also include:
[0121] Based on the self-learning ability of the large model, blind spots in the knowledge base are regularly identified, and the knowledge base is updated and iterated based on the blind spots; the blind spots are information not covered by the knowledge base.
[0122] In addition, the suggestion generation module includes:
[0123] A context reading unit, used for reading background analysis results and public opinion factors as contexts;
[0124] Prior recognition unit, used to identify context based on prior knowledge in the large model and generate marketing decision suggestions;
[0125] The intent receiving unit is used to open an intent receiving port, receive the intent uploaded by the user based on the intent receiving port, and filter and update the marketing decision suggestions according to the received intent.
[0126] The overall description of the above system is as follows:
[0127] The big model serves as the underlying engine: The big model engine can simultaneously complete multiple tasks such as extraction (2)(6), analysis and generation (4)(5), and judgment and decision-making (7)(8). Its performance is better than that of traditional dedicated models, and it replaces the existing method of stacking multiple dedicated models, greatly simplifying the complexity of system maintenance.
[0128] Topic background generation: Use a large model to extract and summarize topic background (2)(4), replacing the traditional private chat that focuses on feature engineering but not content summary. This can improve the effect and interpretability of feature extraction. Introducing the self-reflection step of the large model can also further improve the background generation effect (5).
[0129] Knowledge base: The knowledge base effectively improves the professionalism and timeliness of the large model when performing extraction (2) and generation (4). At the same time, using the large model's extraction capabilities to iterate the knowledge base (6) can maintain the overall self-optimization of the system.
[0130] Background attribution analysis: Combining the statistical characteristics of social media data (3) and topic background for attribution analysis (7) can reflect the characteristics of topics at the social network level, enhance the diversity of topic background analysis, and bring more inspiration to users.
[0131] Decision suggestions: Utilize the inherent knowledge of the big model and generate decision suggestions that are closer to users based on social media topic data, bringing more direct value to users.
[0132] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for extracting and generating social media topics based on large model retrieval, characterized in that: The method comprises: Regularly obtain social media data within the authorized platform; the social media data includes but is not limited to the text, comments and interactive information posted by users, and the interactive information includes the number of likes and shares; Based on the big model, sentiment analysis, topic extraction and triple extraction are performed on social media data. There are three types of sentiment analysis results: positive, negative and neutral. The results of topic extraction are keywords, and the results of triple extraction are subject-predicate-object triple structures. Aggregate social media data based on sentiment analysis results, topic extraction results, and triple extraction results, input into a large model, and generate background analysis results on specific topics; Based on the background analysis results, the knowledge base is updated and iterated, the background analysis results are optimized based on the knowledge base, and the causes of the background analysis results are predicted simultaneously to obtain public opinion factors; With the help of big models, marketing decision recommendations are generated based on background analysis results and public opinion factors.
2. The method for extracting and generating social media topics based on large model retrieval according to claim 1 is characterized in that: The steps of performing sentiment analysis, topic extraction and triple extraction on social media data based on the big model include: Perform sentiment analysis on the acquired social media data according to the big model to identify sentiment data in each social media data, wherein the sentiment data includes positive sentiment, negative sentiment and neutral sentiment; Extract topics from the acquired social media data and determine the discussion topics of each social media data; Extract the subject-predicate-object triple structure from each social media data; The extracted sentiment data, discussion topics, and triple structures are correlated.
3. The method for extracting and generating social media topics based on large model retrieval according to claim 2 is characterized in that: The steps of performing sentiment analysis, topic extraction and triple extraction on social media data based on the big model also include: Obtain the sentiment data from each social media data, sort it by time, and get the sentiment change trend; Obtain discussion topics for each social media data as public opinion keywords; The interaction information of each social media data is obtained, the interaction amount is calculated according to the number of likes and shares in the interaction information, and the social media data corresponding to the maximum interaction amount is selected as the most popular social media data.
4. The method for extracting and generating social media topics based on large model retrieval according to claim 1 is characterized in that: The method further comprises: Based on the self-learning ability of the large model, blind spots in the knowledge base are regularly identified, and the knowledge base is updated and iterated based on the blind spots; the blind spots are information not covered by the knowledge base.
5. The method for extracting and generating social media topics based on large model retrieval according to claim 1 is characterized in that: The step of generating marketing decision suggestions based on background analysis results and public opinion factors by means of the big model includes: Read the background analysis results and public opinion factors as context; Identify context based on prior knowledge in the big model and generate marketing decision recommendations; Open the intent receiving port, receive the intent uploaded by the user based on the intent receiving port, and filter and update the marketing decision suggestions according to the received intent.
6. A system for extracting and generating social media topics based on large model retrieval, characterized in that: The system comprises: A social media data acquisition module, used to periodically acquire social media data within the authorized platform; the social media data includes but is not limited to the text, comments and interactive information posted by users, and the interactive information includes the number of likes and shares; The social media data recognition module is used to perform sentiment analysis, topic extraction, and triple extraction on social media data based on a large model. There are three types of sentiment analysis results: positive, negative, and neutral. The result of topic extraction is a keyword, and the result of triple extraction is a subject-predicate-object triple structure. The social media data aggregation module is used to aggregate social media data based on sentiment analysis results, topic extraction results, and triple extraction results, input into the big model, and generate background analysis results on specific topics; A cause prediction module is used to update the iterative knowledge base based on the background analysis results, optimize the background analysis results based on the knowledge base, and simultaneously predict the causes of the background analysis results to obtain public opinion factors; The suggestion generation module is used to generate marketing decision suggestions based on background analysis results and public opinion factors with the help of a large model.
7. The system for extracting and generating social media topics based on large model retrieval according to claim 6, characterized in that: The social media data identification module includes: An emotion recognition unit is used to perform emotion analysis on the acquired social media data according to the large model, and identify the emotion data in each social media data, wherein the emotion data includes positive emotion, negative emotion and neutral emotion; A topic identification unit is used to extract topics from the acquired social media data and determine the discussion topic of each social media data; A triple extraction unit, used to extract a subject-predicate-object triple structure from each social media data; The information association unit is used to associate the extracted sentiment data, discussion topics and triple structures.
8. The system for extracting and generating social media topics based on large model retrieval according to claim 7, characterized in that: The social media data identification module also includes: The situation analysis unit is used to obtain the sentiment data in each social media data, sort them according to time, and obtain the sentiment change trend; The public opinion analysis unit is used to obtain the discussion topics of each social media data as public opinion keywords; The social media data selection unit is used to obtain the interactive information of each social media data, calculate the interactive amount according to the number of likes and shares in the interactive information, and select the social media data corresponding to the maximum interactive amount as the most popular social media data.
9. The system for extracting and generating social media topics based on large model retrieval according to claim 6, characterized in that: The system also includes: Based on the self-learning ability of the large model, blind spots in the knowledge base are regularly identified, and the knowledge base is updated and iterated based on the blind spots; the blind spots are information not covered by the knowledge base.
10. The system for extracting and generating social media topics based on large model retrieval according to claim 6, characterized in that: The suggestion generation module comprises: A context reading unit, used for reading background analysis results and public opinion factors as contexts; Prior recognition unit, used to identify context based on prior knowledge in the large model and generate marketing decision suggestions; The intent receiving unit is used to open an intent receiving port, receive the intent uploaded by the user based on the intent receiving port, and filter and update the marketing decision suggestions according to the received intent.