Catering merchant comment analysis system and method based on large language model

Abstract

The invention discloses a catering merchant comment analysis system and method based on a big language model, relates to the technical field of the cross field of big model and business data analysis, and aims to intelligently analyze online comment data of merchants based on the big language model and provide improvement suggestions for the merchants through a visualization method. The method is used for an online digital operation scene of the catering industry, and compared with the traditional information consultation industry, all historical information data can be effectively utilized, and all comments of merchants can be accurately and efficiently analyzed under the condition of large data volume; the problems of single industry, random sampling, incomplete sentiment analysis, low efficiency and the like in a traditional analysis model are solved, and the efficiency and the intelligent level of comment analysis are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of the intersection of big models and business data analysis, and in particular to a restaurant merchant review analysis system and method based on a big language model. Background Art

[0002] With the rapid development of the internet and the exponential growth of online data, catering businesses are increasingly shifting their operations online, including takeout, in-store dining, and various other channels. These online operations generate a significant amount of commentary on catering businesses. Brand-specific public opinion analysis and monitoring now play a crucial role in brand operations and management. This valuable insight not only helps businesses identify potential issues but also facilitates optimization and promotion.

[0003] Existing review analysis technologies for catering businesses primarily include manual analysis, word vector models, and keyword screening. These methods have the following characteristics: Manual analysis offers high accuracy and meets ultimate requirements. However, since review analysis is often conducted on large volumes of content, manual processing is inefficient, costly, and can only analyze a subset of samples. Word vector models perform poorly when dealing with complex sentiment or ambiguous semantics, failing to extract all required information from reviews with multiple issues and failing to capture effective semantics in shorter reviews. Keyword screening places high demands on the quality of review content; otherwise, appropriate keywords cannot be matched. Any discrepancies between keywords and actual review content can lead to discrepancies in the final analysis results, impacting the final conclusion. Furthermore, these traditional analysis methods can only reveal the strengths or problems of businesses within the review content, leaving manual analysis as the only viable method for obtaining final conclusions or insights.

[0004] With the rapid development of artificial intelligence, more and more large language models are appearing in our lives. They can replace humans in repetitive and mechanical tasks. The continuous evolution of these models, such as the release of Deepseek's R1 model and OPENAI's O1 model, means that large language models can implement some human-like thinking methods. This means that large language models not only have a deep understanding of the meaning and structure of language, but also have learned human logical and creative thinking. Such models can be applied to the analysis of restaurant reviews. Not only can large language models accurately identify the expression meaning, deep emotions, and multi-dimensional issues of review information, but they can also use models with thought chaining to gain in-depth insights into review information and trends. This effectively solves the low accuracy and inefficiency of other methods and improves the adaptability and efficiency of review analysis systems.

[0005] Traditional review analysis systems often have limited functionality and poor accuracy. They are also inefficient and can only perform sampling analysis. To address the problems with existing restaurant review analysis, we hope to explore the specific application of large models in the restaurant industry. By applying large language models in areas such as sentiment analysis of review content, problem insights, and in-depth analysis and suggestions, we can provide restaurant practitioners with highly accurate and effective analysis results when analyzing store problems. Summary of the Invention

[0006] The technical problem to be solved by the present invention is that the traditional review analysis system has relatively simple functions and poor accuracy, and also has the defects of low efficiency and only sampling analysis.

[0007] In order to solve the above technical problems, the technical solution of the present invention is: A restaurant merchant review analysis system based on a large language model, including: A data acquisition module, which is used to obtain online review information of a specified merchant through an API interface, including but not limited to takeaway reviews and in-store review information; A review information preprocessing module, which is used to standardize the data acquired by the data acquisition module and clean the data to exclude invalid information; A deep analysis module for comment sentiment and questions, which is used to analyze the data preprocessed by the comment information preprocessing module, perform basic classification of the overall sentiment, and call a large language model to conduct in-depth analysis of the reasons and complex emotions of the comment content; A review dimension information integration module, which is used to classify and organize the multi-dimensional information obtained by the review sentiment and question depth analysis module, and obtain the overall review trend; A review insight analysis module, which is used to analyze merchants from different perspectives; The analysis result generation module is used to generate comprehensive analysis results for the merchant, including but not limited to data display, insight analysis and improvement suggestions.

[0008] Preferably, the data acquisition module includes: A data interface unit, which is used to communicate with external data services through a preset API interface. The API interface includes a store information acquisition interface and a review data acquisition interface. The data interface unit uses an API key to authenticate and obtain data download authorization; The data processing unit is used to process the acquired raw data, specifically including receiving and verifying the integrity and validity of API response data, decompressing compressed data files, and executing downloading, saving, and cleaning operations on data files.

[0009] Preferably, the comment information preprocessing module includes: A text data cleaning unit, which uses Python scripts to clean the comment text, remove redundant information, and unify the format encoding; The geographic information processing unit is used to process the geographic location information of different stores of the reviewed merchant brand, and to standardize the store address information by building and maintaining a city information database.

[0010] Preferably, the comment sentiment and question depth analysis module includes: A sentiment analysis unit, which is used to perform sentiment analysis on the review content and implement compound sentiment analysis by constructing a multi-level sentiment dictionary; The review summary analysis unit constructs a multi-dimensional summary word label dictionary, counts the review information and all summary word dimensions, and then inputs them into the large language model, and counts the good and bad reviews.

[0011] The theme negative review reason positioning unit, which constructs a negative review reason label library, improves the negative review dimension vocabulary, summarizes and locates the negative review reasons of catering businesses in different dimensions, and establishes two sets of multi-dimensional theme classification systems based on in-store reviews and takeaway reviews.

[0012] Preferably, the comment dimension information integration module includes: A summary word summary unit is used to count the top 50 most frequently appearing phenomena in all review information, and to analyze the main positive and negative review points of the restaurant business to assist in the optimization and improvement of the store; A popular dish summary unit is used to summarize and analyze popular dishes with high sales volume. By filtering and sorting the sales volume of dishes in the usage data, the reasons for the positive and negative reviews of high-selling dishes are analyzed to facilitate targeted dish improvements. A city feature summary unit, which is suitable for restaurant businesses with multiple chain stores. This unit can conduct comparative analysis across different cities and stores, and calculate the distribution of good and bad ratings for a single city and a single store, as well as the mention rate and praise rate of different dimensions, for comparative analysis; The review trend analysis unit is used to collect statistics on the development trends of review opinions in different time dimensions, including the ratings and the changing trends of good and bad reviews in weekly, monthly and quarterly units, and at the same time analyze the correlation between the ratings in different dimensions, and the correlation between the total review score and the taste score and the delivery evaluation and packaging score.

[0013] Preferably, the comment in-depth insight analysis module includes: A multi-dimensional analysis unit, which uses the different dimensional information obtained from the large model to analyze the ratio of good and bad reviews and existing problems in different dimensions; A city store multi-dimensional analysis unit is used to conduct separate in-depth analysis of different addresses or stores in the chain store, including the main reasons for negative reviews in each city and the differences in different dimensions.

[0014] Preferably, the analysis result generating module includes: Image generation module, which is used to visualize statistical data and is set in the front-end interface and report. A document report generation module, which is used to format and display the analysis results of statistical data, pictures and large models; The PPT report generation module is used to display analysis results and statistical data.

[0015] This application also discloses a method for analyzing restaurant reviews based on a large language model, which specifically includes the following steps: Step 1: Obtain online review information of the designated merchant through the API interface, including but not limited to takeaway reviews and in-store reviews; Step 2: Standardize the data obtained in step 1 and clean the data to eliminate invalid information; Step 3: Analyze the data preprocessed in step 2 and perform basic classification of overall sentiment. At the same time, use the large language model to conduct in-depth analysis of the reasons and complex emotions of the comments. Step 4: Classify and organize the multi-dimensional information obtained in step 3, and obtain the overall review trend; Step 5: Based on the data obtained in step 4, analyze the merchant from different perspectives; Step 6: Generate comprehensive analysis results for the merchant, including but not limited to data display, insight analysis and improvement suggestions.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The catering merchant review analysis system and method based on a large language model provided in this application is to perform intelligent analysis on the merchant's online review data based on the large language model, and provide improvement suggestions to the merchant through a visualization method. It is used in the online digital operation scenario of the catering industry. Compared with the traditional information consulting industry, it can effectively utilize all historical information data, and can accurately and efficiently realize the analysis of all merchant reviews under large data conditions. It solves the problems of single industry, random sampling, incomplete sentiment analysis, and low efficiency in traditional analysis models, and effectively improves the efficiency and intelligence level of review analysis.

[0017] In this application, the comment sentiment and question in-depth analysis module uses a large model, which is more efficient than manual filtering. The large language model can achieve a comprehensive analysis of all comments. At the same time, the analysis of the large language model can replace manual analysis to achieve unified standards and accurate results.

[0018] The comment dimension information integration module in this application counts the information of all data and uses the analysis results for statistics more comprehensively. It has good performance and is easy to use.

[0019] The comment in-depth insight analysis module in this application solves the problem that traditional comment analysis systems can only count keywords and sentiments. This application can use the deep thinking chain of the current large language model to imitate experts to analyze the information and improvement methods in the comments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a data flow diagram of a restaurant merchant review analysis system based on a large language model according to the present invention; Figure 2 This is a flow chart of a method for analyzing restaurant reviews based on a large language model according to the present invention; Figure 3 A pie chart generated by an analysis result generation module in a restaurant merchant review analysis system based on a large language model according to the present invention; Figure 4 This is a histogram generated by an analysis result generation module in a restaurant merchant review analysis system based on a large language model according to the present invention; Figure 5 This is a trend graph generated by an analysis result generation module in a restaurant merchant review analysis system based on a large language model according to the present invention; Figure 6 This is a rating dimension correlation heat map generated by the analysis result generation module in the catering merchant review analysis system based on the large language model of the present invention. DETAILED DESCRIPTION

[0021] The following is a further description of specific embodiments of the present invention in conjunction with the accompanying drawings. It should be noted that the description of these embodiments is intended to facilitate understanding of the present invention and does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0022] The present invention discloses a restaurant merchant review analysis system based on a large language model. Figure 1 and 2 As shown, the system specifically includes the following modules: The data acquisition module is used to obtain online review information of designated merchants through the API interface, including but not limited to takeaway reviews and in-store reviews, for comprehensive analysis of merchant reviews. Specifically, the data acquisition module includes: The data interface unit is used to communicate with external data services through a preset API interface. The API interface includes a store information acquisition interface and a review data acquisition interface. The data interface unit authenticates and obtains data download authorization through the API key, extracting the review data of the corresponding store from the information database or manually uploading a table file; The data processing unit is used to process the acquired raw data, including receiving and verifying the integrity and validity of API response data, decompressing compressed data files, downloading, saving and cleaning data files, modifying errors and display content in the data, and aligning different data sources.

[0023] The review information preprocessing module is used to standardize the data obtained by the data acquisition module, such as review content, ratings, and geographic location information, and clean the data to exclude invalid information. Specifically, the review information preprocessing module includes: The text data cleaning unit uses Python scripts to clean review text, remove redundant information, and standardize the format encoding. Specifically, the text data cleaning unit cleans the stored information in the data. For example, it removes content other than dish information from order information, as well as content that is inconsistent with dish names and order storage information. It also converts JSON-encoded files into Chinese and a standard format for storage. It also removes useless information irrelevant to the overall public opinion analysis of food delivery, such as delivery driver comments, follow-up review time, and merchant response time.

[0024] The geographic information processing unit is used to process the geographic location information of different stores of the reviewed merchant brand. By building and maintaining a city information database, the store address information is standardized and saved in the database, which can be directly called and completed for subsequent analysis.

[0025] The deep analysis module for comment sentiment and issues is used to analyze the data preprocessed by the comment information preprocessing module and perform basic classification of the overall sentiment. For example, sentiment classification includes: positive, negative, neutral; complex sentiment: low score but positive sentiment, high score but negative sentiment. At the same time, it calls locally deployed open source large language models, such as qwen-32B, qwen2.5, deepseek-v3, deepseek-r1, and GLM-4-32B, to conduct in-depth analysis of the reasons and complex emotions of the comment content. Specifically, the deep analysis module for comment sentiment and issues includes: Sentiment analysis unit, the sentiment analysis unit is used to realize sentiment analysis of comment content. By constructing multi-level sentiment tags, it inputs them and comment content as prompt words into the large language model to realize the extraction of multiple complex emotions in the comment information. Sentiment analysis alternatives include rule-based sentiment judgment, deep learning-based sentiment recognition model and natural language processing model-based sentiment analysis method. Specifically, rule-based sentiment judgment: use predefined dictionaries and simple rules to judge the sentiment tendency of keywords; basic text processing: use Jieba word segmentation and regular expressions to process text.

[0026] The review summary analysis unit constructs a multi-dimensional summary word label dictionary. It first counts the review information and all summary word dimensions and then inputs them into the large language model to let it determine what dimensions of evaluation exist. Then, it inputs the large language description of the review content with the dimension combined with the label of the dimension to let it determine what words can be used to summarize the review. This unit counts as many words as possible mentioned in the good and bad reviews.

[0027] The topic-based negative review reason location unit, by constructing a negative review reason label library and an improveable negative review dimension vocabulary, is used to summarize and locate the reasons for negative reviews of catering businesses in different dimensions. It has established two sets of multi-dimensional topic classification systems based on in-store reviews and takeaway reviews, and combined with large models to realize the identification, extraction and classification of problems in different application scenarios and reviews; specific large models: qwen-32B, qwen2.5, deepseek-v3, deepseek-r1 and GLM-4-32B; the main general dimensions of analysis: takeaway scenarios: taste satisfaction, packaging presentation, merchant service, ingredient presentation, delivery efficiency, sense of security, cost-effectiveness, etc.; in-store review dimensions: taste satisfaction, merchant service, dining environment, cost-effectiveness, dish presentation, store efficiency, sense of security, etc.

[0028] The review dimension information integration module is used to classify and organize the multi-dimensional information obtained by the review sentiment and problem in-depth analysis module. The specific general dimensions include: takeaway scenarios: taste satisfaction, packaging presentation, merchant service, ingredient presentation, delivery efficiency, security, cost-effectiveness, etc.; in-store review dimensions: taste satisfaction, merchant service, dining environment, cost-effectiveness, dish presentation, store efficiency, security, etc., and obtain the overall review trend. The trend is divided into quarterly, monthly, weekly and daily trends in the ratio of good and bad reviews, and seasonal or cyclical problems in different dimensions are explored. The specific review dimension information integration module includes: The summary word unit is used to count the top 50 most frequently appearing phenomena in all review information, and to analyze the main positive and negative review points of the restaurant business to assist in store optimization and improvement; The popular dish summary unit is used to summarize and analyze popular dishes with high sales. By filtering and sorting the sales of dishes in the usage data, the reasons for positive and negative reviews of high-selling dishes are analyzed to facilitate targeted dish improvements. The city feature summary unit is suitable for catering businesses with multiple chain stores. It can conduct comparative analysis across different cities and stores, and compile statistics on the distribution of good and bad ratings for a single city and a single store, as well as the mention rate and praise rate of different dimensions for comparative analysis. The review trend analysis unit is used to collect statistics on the development trends of review opinions in different time dimensions, including weekly, monthly, and quarterly ratings and the trend of positive and negative reviews. It also analyzes the correlation between ratings in different dimensions, and the correlation between the total review score and the taste score, delivery evaluation and packaging score. In this application, Pearson correlation analysis is used to analyze the correlation between different ratings. The Pearson correlation coefficient is a statistic that measures the degree of linear correlation between two continuous variables. Its calculation formula is as follows: The Review Insight Analysis module is used to analyze merchants from different perspectives, such as positive reviews, negative reviews, cities, and dishes. The insights can be used to assist merchants in future improvements and optimization methods. The Review Insight Analysis module includes: The multi-dimensional analysis unit obtains information of different dimensions through the big model, and conducts insight analysis on the ratio of good and bad reviews and existing problems in different dimensions. Specifically, the statistical data on the causes and ratios of bad reviews in different dimensions obtained by the big model analysis are input into the big language model of the deep thinking chain for analysis and insights, and provides guiding suggestions and in-depth insight analysis on the ratio of good and bad reviews and existing problems in different dimensions.

[0029] The city store multi-dimensional analysis unit is used to conduct separate in-depth analysis of different addresses or stores of chain stores, including the main reasons for negative reviews in each city and the differences in different dimensions. It mainly focuses on the concerns of consumers in different cities and the problems existing in stores, and is used to assist merchants in making targeted improvements to different stores.

[0030] The final suggestions are to provide targeted improvement methods for the reasons for negative reviews in different dimensions, or to analyze the problems in different areas or stores of chain merchants and provide which dimensions of the stores need to be focused on for improvement.

[0031] The analysis result generation module is used to generate comprehensive analysis results for the merchant, including but not limited to data display, insight analysis and improvement suggestions.

[0032] The analysis result generation module includes: Image generation module,The image generation module is used to visualize statistical data, such as Figure 3 、 Figure 4 、 Figure 5 and Figure 6 As shown, the image generation module is set in the front-end interface and report, including word cloud charts, trend charts, pie charts, and bar charts.

[0033] The document report generation module is used to display the analysis results of statistical data, pictures and large models in an aesthetically pleasing format, and provide merchants with detailed text analysis to analyze existing problems and make improvements; The PPT report generation module is used to present analysis results and statistical data more clearly, and to display analysis results through more charts and structured data.

[0034] like Figure 1 As shown, this application also discloses a restaurant merchant review analysis method based on a large language model, which specifically includes the following steps: Step 1: Obtain online review information of designated merchants through the API interface, including but not limited to takeaway reviews and in-store review information. The entire process is automated and compatible with manually uploaded data. Specifically, the preset API interface communicates with the external data service. The API interface mainly includes a store information acquisition interface and a review data acquisition interface. The data interface unit authenticates and obtains data download authorization through the API key, extracts the review data of the corresponding store from the information library or manually uploads a table file; and processes the obtained raw data, specifically including receiving and verifying the integrity and validity of the API response data, decompressing the compressed data file, downloading, saving and cleaning the data file, modifying errors and display content in the data, and aligning different data sources.

[0035] Step 2: Standardize the data obtained in Step 1 and clean the data to remove invalid information, replacing manual information deletion and filtering to improve processing efficiency and data consistency across different data sources. Specifically, clean the order information except for the dishes, as well as any inconsistencies between dish names and order storage information. Convert the JSON-encoded files to Chinese and a standard format for storage. Remove useless information irrelevant to the overall public opinion analysis of food delivery, such as delivery driver comments, follow-up review times, and merchant response times.

[0036] Step 3: Analyze the data preprocessed in Step 2 and perform a basic classification of overall sentiment. Simultaneously, a large language model is used to further analyze the reasons behind and complex emotions in the comments. Specifically, basic classification is performed on overall sentiment, such as positive, negative, and neutral; complex emotions are classified as low-rated but positive, and high-rated but negative. Simultaneously, a large language model, such as qwen-32B, qwen2.5, deepseek-v3, deepseek-r1, and GLM-4-32B, is used to further analyze the reasons behind and complex emotions in the comments.

[0037] By constructing a multi-dimensional summary word label dictionary, the review information and all summary word dimensions are first counted and then input into the large language model to let it determine what dimensions of evaluation exist. Then, the review content with existing dimensions is combined with the label of the dimension and input into the large language description to let it determine what vocabulary can be used to judge the review. This unit counts the widest possible range of words mentioned in good and bad reviews.

[0038] By constructing a negative review reason label library and an improveable negative review dimension vocabulary, we summarize and locate the reasons for negative reviews of catering businesses in different dimensions, establish two multi-dimensional topic classification systems based on in-store reviews and takeaway reviews, and combine them with large models to identify, extract and classify problems in different application scenarios and reviews; specific large models: qwen-32B, qwen2.5, deepseek-v3, deepseek-r1 and GLM-4-32B; main general dimensions for analysis: takeaway scenarios: taste satisfaction, packaging presentation, merchant service, ingredient presentation, delivery efficiency, sense of security, cost-effectiveness, etc.; in-store review dimensions: taste satisfaction, merchant service, dining environment, cost-effectiveness, dish presentation, store efficiency, sense of security, etc.

[0039] Step 4: Classify and organize the multi-dimensional information obtained in step 3, obtain the overall review trend, count all the data information, and use the analysis results for more comprehensive statistics. Specific general multi-dimensional dimensions include: takeout scenarios: taste satisfaction, packaging presentation, merchant service, ingredient presentation, delivery efficiency, safety, and value for money; in-store review dimensions: taste satisfaction, merchant service, dining environment, value for money, dish presentation, store efficiency, and safety. This also includes statistics on the top 50 most frequently appearing reviews, and application analysis of the restaurant's main positive and negative review points to assist in store optimization and improvement. This can be used to summarize and analyze popular dishes with high sales volume. By filtering and sorting the sales volume of dishes using usage data, the reasons for positive and negative reviews for high-selling dishes can be analyzed to facilitate targeted dish improvements. For restaurants with multiple chain stores, comparative analysis can be conducted across different cities and stores, with statistics on the distribution of positive and negative reviews in a single city and store, as well as the mention rate and positive review rate of different dimensions for comparative analysis. Statistics are also collected on review and public opinion trends across different time dimensions, including weekly, monthly, and quarterly ratings and positive and negative review trends. The correlation between ratings across different dimensions is also analyzed, as is the correlation between the total review score and the taste, delivery, and packaging ratings.

[0040] Step 5: Based on the data obtained in Step 4, analyze the merchant from various perspectives. This addresses the limitation of traditional review analysis systems, which are limited to keyword and sentiment statistics. Using the Deep Thinking Chain within the current large language model, the merchant is analyzed from different perspectives, mimicking expert analysis of review information and improvement strategies. Specifically, the merchant is analyzed from various perspectives, such as positive and negative reviews, city, and dish. The insights gained can be used to inform future improvements and optimization strategies. Using the diverse information obtained from the large model, insights are analyzed into the ratio of positive and negative reviews and existing issues across different dimensions. The statistical data on the reasons and ratios for negative reviews across different dimensions analyzed by the large model is fed into the Deep Thinking Chain large language model for analysis. This analysis provides guiding recommendations and in-depth insights into the ratios and issues across different dimensions. The City Store Multi-Dimensional Analysis Unit provides in-depth analysis of different locations or stores within the chain, including the primary reasons for negative reviews and differences across different dimensions within each city. This analysis focuses on consumer concerns and store issues across different cities, assisting merchants in implementing targeted improvements across different stores. The final suggestions are to provide targeted improvement methods for the reasons for negative reviews in different dimensions, or to analyze the problems in different areas or stores of chain merchants and provide which dimensions of the stores need to be focused on for improvement.

[0041] Step 6: Generate comprehensive analysis results for the merchant, including but not limited to data display, insight analysis, and improvement suggestions. This changes the traditional visualization-only analysis method. By generating result reports and PPTs, the review analysis results can be more clearly viewed, and the merchant's improvements can be compared to see if they have actually worked. Specifically, this includes generating word clouds, trend charts, pie charts, and bar charts from images; displaying statistical data, images, and large model analysis results in an attractive layout; and providing detailed text analysis to merchants to analyze existing problems and improve them. Analytical statistics are presented through more charts and structured data.

[0042] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It is apparent to those skilled in the art that various changes, modifications, substitutions, and variations to these embodiments may be made without departing from the principles and spirit of the present invention, and these changes and modifications still fall within the scope of protection of the present invention.

Claims

1. A restaurant review analysis system based on a large language model, characterized by: include, A data acquisition module, which is used to obtain online review information of a specified merchant through an API interface, including but not limited to takeaway reviews and in-store review information; A review information preprocessing module, which is used to standardize the data acquired by the data acquisition module and clean the data to exclude invalid information; A deep analysis module for comment sentiment and questions, which is used to analyze the data preprocessed by the comment information preprocessing module, perform basic classification of the overall sentiment, and call a large language model to conduct in-depth analysis of the reasons and complex emotions of the comment content; A review dimension information integration module, which is used to classify and organize the multi-dimensional information obtained by the review sentiment and question depth analysis module, and obtain the overall review trend; A review insight analysis module, which is used to analyze merchants from different perspectives; The analysis result generation module is used to generate comprehensive analysis results for the merchant, including but not limited to data display, insight analysis and improvement suggestions.

2. A restaurant merchant review analysis system based on a large language model according to claim 1, characterized in that: The data acquisition module includes: A data interface unit, which is used to communicate with external data services through a preset API interface. The API interface includes a store information acquisition interface and a review data acquisition interface. The data interface unit uses an API key to authenticate and obtain data download authorization; The data processing unit is used to process the acquired raw data, specifically including receiving and verifying the integrity and validity of API response data, decompressing compressed data files, and executing downloading, saving, and cleaning operations on data files.

3. The restaurant merchant review analysis system based on a large language model according to claim 2, characterized in that: The comment information preprocessing module includes: A text data cleaning unit, which uses Python scripts to clean the comment text, remove redundant information, and unify the format encoding; The geographic information processing unit is used to process the geographic location information of different stores of the reviewed merchant brand, and to standardize the store address information by building and maintaining a city information database.

4. The restaurant merchant review analysis system based on a large language model according to claim 3, characterized in that: The comment sentiment and question depth analysis module includes: A sentiment analysis unit, which is used to perform sentiment analysis on the review content and implement compound sentiment analysis by constructing a multi-level sentiment dictionary; The review summary analysis unit constructs a multi-dimensional summary word label dictionary, counts the review information and all summary word dimensions, and then inputs them into the large language model, and counts the good and bad reviews.

5. Theme negative review reason positioning unit, which constructs a negative review reason label library, improves the negative review dimension vocabulary, summarizes and locates the negative review reasons of catering merchants in different dimensions, and establishes two sets of multi-dimensional theme classification systems based on in-store reviews and takeaway reviews.

6. The restaurant merchant review analysis system based on a large language model according to claim 4, characterized in that: The comment dimension information integration module includes: A summary word summary unit is used to count the top 50 most frequently appearing phenomena in all review information, and to analyze the main positive and negative review points of the restaurant business to assist in the optimization and improvement of the store; A popular dish summary unit is used to summarize and analyze popular dishes with high sales volume. By filtering and sorting the sales volume of dishes in the usage data, the reasons for the positive and negative reviews of high-selling dishes are analyzed to facilitate targeted dish improvements. A city feature summary unit, which is suitable for restaurant businesses with multiple chain stores. This unit can conduct comparative analysis across different cities and stores, and calculate the distribution of good and bad ratings for a single city and a single store, as well as the mention rate and praise rate of different dimensions, for comparative analysis; The review trend analysis unit is used to collect statistics on the development trends of review opinions in different time dimensions, including the ratings and the changing trends of good and bad reviews in weekly, monthly and quarterly units, and at the same time analyze the correlation between the ratings in different dimensions, and the correlation between the total review score and the taste score and the delivery evaluation and packaging score.

7. The restaurant merchant review analysis system based on a large language model according to claim 5, characterized in that: The review insight analysis module includes: A multi-dimensional analysis unit, which uses the different dimensional information obtained from the large model to analyze the ratio of good and bad reviews and existing problems in different dimensions; A city store multi-dimensional analysis unit is used to conduct separate in-depth analysis of different addresses or stores in the chain store, including the main reasons for negative reviews in each city and the differences in different dimensions.

8. The restaurant merchant review analysis system based on a large language model according to claim 6, characterized in that: The analysis result generation module includes: Image generation module, which is used to visualize statistical data and is set in the front-end interface and report. A document report generation module, which is used to format and display the analysis results of statistical data, pictures and large models; The PPT report generation module is used to display analysis results and statistical data.

9. A method for analyzing restaurant reviews based on a large language model, characterized by: The specific steps include: Step 1: Obtain online review information of the designated merchant through the API interface, including but not limited to takeaway reviews and in-store reviews; Step 2: Standardize the data obtained in step 1 and clean the data to eliminate invalid information; Step 3: Analyze the data preprocessed in step 2 and perform basic classification of overall sentiment. At the same time, use the large language model to conduct in-depth analysis of the reasons and complex emotions of the comments. Step 4: Classify and organize the multi-dimensional information obtained in step 3, and obtain the overall review trend; Step 5: Based on the data obtained in step 4, analyze the merchant from different perspectives; Step 6: Generate comprehensive analysis results for the merchant, including but not limited to data display, insight analysis and improvement suggestions.

Images