A jackfruit wine tasting and review management system

Abstract

This invention discloses a jackfruit wine tasting and review management system. By collecting publicly available review data in real time, it first filters out non-professional content based on preset professional standards. Then, it performs deep clustering analysis of semantic and visual features on text and image information with professional value to identify consensus-based evaluation viewpoints. From this, it further selects high-quality human reviews with accompanying authentic verification information as high-weighted processing objects. Finally, it generates targeted professional-level responses including Q&A and improvement commitments, and can also improve the wine's production process or ingredient ratios based on professional reviews. This achieves the identification and hierarchical management of reviewers' professional levels, solving the technical problem of existing general review systems that use an equal weighting strategy for all users' reviews and lack the ability to discern reviewers' professional levels.

Description

Technical Field

[0001] This invention relates to the field of information management technology, specifically to a jackfruit wine tasting and review management system. Background Technology

[0002] With the development of e-commerce and social media, online tasting reviews of alcoholic beverages have become an important basis for consumer decision-making. Currently, common alcohol review systems, such as the review sections of e-commerce platforms and vertical alcohol communities, generally adopt a comprehensive rating model that combines free text reviews with star ratings.

[0003] For niche and distinctive wines like jackfruit wine, with their unique flavors and tasting dimensions differing from traditional wines or spirits, existing general review systems employ an equal weighting strategy for all user reviews, lacking the ability to discern the professional competence of reviewers. This results in ordinary consumers and experienced sommeliers, as well as jackfruit wine producers, receiving the same weighting coefficients. Consequently, professional and valuable reviews are easily drowned out by a massive influx of non-professional comments, reducing the overall reference value and credibility of the reviews. Summary of the Invention

[0004] The purpose of this invention is to provide a jackfruit wine tasting and review management system to solve the technical problem that existing general review systems adopt an equal weighting strategy for all users' reviews and lack the ability to distinguish the professional level of reviewers.

[0005] The technical solution of this invention is implemented as follows:

[0006] A jackfruit wine tasting and review management system includes a text acquisition module, a text screening module, a positive review analysis module, and a sentiment response module;

[0007] The text acquisition module is used to acquire and update comment data in real time, and the comment data includes text information and image information;

[0008] The text screening module is used to filter the comment data, removing general text information and irrelevant image information, and retaining professional text information and image information;

[0009] The evaluation analysis module is used to compare and analyze the professional text and image information, identify sentences with similar semantics and images with similar content, and select priority content that contains real addresses and is human comment.

[0010] The emotional response module is used to respond to the priority content, including answering questions and making improvement commitments based on suggestions.

[0011] A further technical solution is that the text acquisition module includes:

[0012] The data interface unit is used to access the comment data in real time from multiple heterogeneous data sources;

[0013] A data parsing unit is used to parse the comment data and separate the text information from the image information.

[0014] The real-time deduplication unit is used to perform real-time deduplication processing on the parsed text and image information;

[0015] The data storage and update unit is used to store the deduplicated comment data in the database and respond to update instructions to maintain the latest state of the data.

[0016] A further technical solution is that the real-time deduplication unit step includes:

[0017] Step S11: Receive text information stream and image information stream from the data parsing unit;

[0018] Step S12: Extract semantic feature vectors from the text information stream and extract visual feature vectors from the image information stream;

[0019] Step S13: Perform a real-time similarity comparison between the feature vector of the newly received data and the feature vector of the stored data, and determine whether it is duplicate content based on a preset threshold.

[0020] Step S14: If the content is determined to be duplicate, it is filtered out; if it is not duplicate, its feature vector is stored in the feature library and the data is output.

[0021] A further technical solution is that the text screening module includes:

[0022] An information preprocessing unit is used to receive the comment data and separate and preprocess the text information stream and image information stream therein;

[0023] The text professionalism determination unit analyzes the text information flow based on a preset keyword library for the tasting field and a syntactic complexity model;

[0024] The image relevance determination unit is used to identify the main content in the image information stream through image recognition technology, and determine whether it is an image related to wine based on a preset list of related items.

[0025] The compliance information integration unit is used to collect texts deemed professional and images deemed relevant, forming a filtered data set.

[0026] A further technical solution is that the text professionalism determination unit includes:

[0027] A text processing unit is used to receive the text information stream and perform word segmentation and stop word removal.

[0028] The keyword matching unit is used to match the preprocessed text with the keyword library of the tasting field and calculate the keyword hit rate and weight.

[0029] The syntactic structure analysis unit, based on the syntactic complexity model, analyzes the sentence structure, clause depth, and collocation of technical terms in the text, and generates a complexity score.

[0030] The professionalism determination unit is used to integrate the keyword hit rate, weight, and syntactic complexity score, and output the determination result of whether the text belongs to professional text according to preset rules.

[0031] A further technical solution is that the image correlation determination unit includes:

[0032] An image processing unit is used to receive the image information stream and perform size normalization, noise filtering, and region enhancement processing.

[0033] The recognition and extraction unit is used to recognize the preprocessed image and extract the visual feature vectors of the main objects in the image.

[0034] The associated item comparison unit is used to calculate the similarity between the extracted visual feature vector and the benchmark features of various categories in the preset associated item list, and output the matching score.

[0035] A relevance determination unit is used to receive the matching score and compare it with a preset threshold to determine whether the image is related to wine.

[0036] A further technical solution is that the performance evaluation analysis module includes:

[0037] The content clustering analysis unit is used to perform semantic clustering on the professional text information to identify similar semantic statements, and to perform visual feature clustering on the image information to identify images with homogeneous content.

[0038] The commenter authentication unit is used to determine whether the similar semantic statements and content-homogeneous images originate from real human comments by analyzing account behavior patterns and comment interaction characteristics.

[0039] A geographic information verification unit is used to parse and verify the existence of the real address from the semantic statements of the same type;

[0040] The priority content selection unit is used to select priority content from content that simultaneously meets the conditions of human comments and contains a valid real address, according to preset filtering rules.

[0041] A further technical solution is that the content clustering analysis unit includes:

[0042] The feature vectorization unit is used to convert the specialized text information into text semantic vectors and the image information into image visual feature vectors.

[0043] The model optimization unit uses the particle swarm optimization algorithm to search for the optimal cluster center or similarity threshold for the text semantic vector and the image visual feature vector;

[0044] The clustering execution unit performs clustering operations on the text semantic vector and the image visual feature vector, respectively, by the optimal cluster center or the partitioning threshold, to generate text semantic clusters and image visual clusters;

[0045] The clustering result mapping unit is used to map the text semantic clusters back to the original set of similar semantic statements, and to map the image visual clusters back to the original set of content-homogeneous images.

[0046] A further technical solution is that the model optimization unit performs the following steps:

[0047] Step S21: Initialize the particle swarm by randomly assigning each particle an initial solution representing the cluster center coordinates or similarity threshold, and initialize the particle velocity.

[0048] Step S22: For each solution represented by a particle, calculate the clustering of the text semantic vector and image visual feature vector based on the solution, and calculate the fitness of the particle according to the intra-class tightness and inter-class separation.

[0049] Step S23: Update the velocity and position of each particle based on its current fitness and its historical best solution, as well as the global historical best solution of the entire particle swarm, thereby updating the solution it represents.

[0050] Step S24: Repeat steps S22 and S23 until the preset number of iterations or fitness convergence condition is met. Output the solution corresponding to the global historical best solution of the particle swarm as the optimal cluster center or similarity partitioning threshold.

[0051] A further technical solution is that the emotion response module includes:

[0052] The response strategy generation unit is used to receive the priority content, parse the question sentences and improvement suggestions therein, and generate corresponding response scripts and improvement commitment frameworks respectively.

[0053] The response content synthesis unit matches or constructs specific response text from a preset knowledge base based on the response script;

[0054] The commitment tracking and management unit is used to register the improved commitment framework as a to-do item and associate it with the responsible party and the planned completion deadline;

[0055] A unified response output unit is used to integrate specific response texts and corresponding improvement commitments into a formatted final response, which is then published.

[0056] The beneficial effects of this invention are as follows:

[0057] By collecting publicly available comment data in real time, the system first filters out non-professional content based on preset professional standards. Then, it performs deep clustering analysis of semantic and visual features on text and image information with professional value to identify consensus-based evaluation viewpoints. From this, high-quality human comments with accompanying verified information are further selected as high-weighted processing objects. Finally, it generates targeted professional-level responses including answers and improvement commitments, and can also improve the wine's production process or ingredient ratios based on professional comments. This achieves the identification and hierarchical management of commenters' professional levels, solving the technical problem of existing general comment systems that use an equal weighting strategy for all users' comments and lack the ability to discern the professional level of commenters. Attached Figure Description

[0058] Figure 1 This is a system block diagram of Embodiment 1;

[0059] Figure 2 This is a system block diagram of the text acquisition module in Embodiment 2;

[0060] Figure 3 This is a system block diagram of the text screening module in Example 2;

[0061] Figure 4 This is a system block diagram of the performance evaluation analysis module in Example 2;

[0062] Figure 5 This is a system block diagram of the emotion response module in Embodiment 2. Detailed Implementation

[0063] To better understand the technical content of this invention, specific embodiments are provided below, and the invention will be further described in conjunction with the accompanying drawings.

[0064] Example 1

[0065] See Figure 1This invention provides a jackfruit wine tasting and review management system, including a text acquisition module, a text screening module, a positive review analysis module, and a sentiment response module. The text acquisition module is used to acquire and update review data in real time, including text and image information. The text screening module is used to filter the review data, removing general text and irrelevant image information, and retaining professional text and image information. The positive review analysis module is used to compare and analyze professional text and image information, identify statements with similar semantics and images with similar content, and select priority content containing real addresses and being human reviews. The sentiment response module is used to respond to priority content, including answering questions and making improvement commitments based on suggestions.

[0066] Specifically, the text acquisition module automatically captures public comments from e-commerce platforms and social media through a network interface program and stores them in a distributed database in real time. Then, the text screening module is activated, using its built-in tasting keyword library and image recognition model to process the incoming data in parallel, automatically filtering out general texts such as "delicious" or "average," as well as images unrelated to the bottle, color, or brewing scene. Next, the performance evaluation module uses semantic vector clustering and image feature comparison algorithms to identify clusters of comments with similar professional descriptions such as "aroma layers" and "mellow taste" from the filtered text and image data. Then, through an account behavior analysis model and a geolocation verification interface, comments with real store visit information and genuine interaction characteristics are extracted from these clusters and prioritized. Finally, the sentiment response module generates professional answers based on the specific questions in the prioritized content by calling the knowledge base, while automatically generating work orders for process improvement suggestions and assigning them to the production management system.

[0067] In this embodiment of the invention, by collecting publicly available comment data in real time, non-professional content is first filtered out according to preset professional standards. Then, deep clustering analysis of semantic and visual features is performed on text and image information with professional value to identify consensus-based evaluation viewpoints. From these, high-quality human comments with accompanying authentic verification information are further selected as high-weight processing objects. Finally, a targeted professional-level response containing answers and improvement commitments is generated. Simultaneously, improvements can be made to the wine's production process or raw material ratio based on professional comments. This achieves the identification and hierarchical management of commenters' professional levels, solving the technical problem of existing general comment systems that use an equal weighting strategy for all users' comments and lack the ability to discern the professional level of commenters.

[0068] Example 2

[0069] Please refer to Figures 2 to 5As a further improvement to Embodiment 1, the text acquisition module includes: a data interface unit for real-time access to comment data from multiple heterogeneous data sources; a data parsing unit for parsing the comment data and separating the text information and image information therein; a real-time deduplication unit for real-time deduplication of the parsed text and image information; and a data storage and update unit for storing the deduplicated comment data in a database and responding to update commands to maintain the latest state of the data.

[0070] Specifically, through configured application programming interfaces (APIs) and web crawlers, comment data is continuously collected from multiple sources, including e-commerce platform product pages, social media topics, and offline tasting event application systems. The collected mixed-format data packets are parsed, separating them into independent text data and image files based on text encoding and image file header information. Then, by calculating text semantic fingerprints and image perceptual hash values, these are compared in real-time with corresponding features in existing data to identify and discard duplicate entries. After processing, the unique text and image data, along with their feature indexes, are stored in a time-series database. By monitoring data source changes or through periodic polling mechanisms, newly added or modified comment content is automatically acquired and integrated to update the database, ensuring the diversity of sources, the uniqueness of content, and the real-time status of the obtained comment data, providing a continuously updated data foundation for subsequent analysis.

[0071] Furthermore, the real-time deduplication unit step includes:

[0072] Step S11: Receive text information stream and image information stream from the data parsing unit;

[0073] Step S12: Extract semantic feature vectors from the text information stream and extract visual feature vectors from the image information stream;

[0074] Step S13: Perform a real-time similarity comparison between the feature vector of the newly received data and the feature vector of the stored data, and determine whether it is duplicate content based on a preset threshold.

[0075] Step S14: If the content is determined to be duplicate, it is filtered out; if it is not duplicate, its feature vector is stored in the feature library and the data is output.

[0076] Specifically, upon receiving the text and image data stream, the system immediately converts each text comment into a semantic vector reflecting its core meaning using a natural language model. Simultaneously, it converts each image into a feature vector representing its visual content using an image feature extraction algorithm. These newly generated feature vectors are then rapidly compared to the feature vectors of all historical data in the feature library. If the similarity to any historical vector exceeds a preset deduplication threshold, the new data is considered duplicate content and discarded. For data deemed non-duplicate, the system outputs its original content to downstream modules while simultaneously storing its newly generated feature vector as a new comparison benchmark in the feature library. This process identifies and filters identical or highly similar comments repeatedly published at different times or from different sources, ensuring that every text and image entering the subsequent analysis process is unique and maintaining the freshness of the dataset.

[0077] In a preferred embodiment, the text screening module includes: an information preprocessing unit for receiving comment data and separating and preprocessing the text information stream from the image information stream; a text professionalism determination unit for analyzing the text information stream based on a preset keyword library for the tasting field and a syntactic complexity model; an image relevance determination unit for identifying the main content in the image information stream using image recognition technology and determining whether it is an image related to wine based on a preset list of related items; and a compliance information integration unit for aggregating the text determined to be professional and the images determined to be relevant to form a filtered data set.

[0078] Specifically, upon receiving mixed text and image review data, the system automatically separates it into independent text and image streams, and performs standardized formatting for subsequent analysis. For the text stream, it compares it against a built-in tasting terminology database, analyzing sentence structure complexity and terminology usage to determine if it contains specific technical details. For the image stream, a computer vision model identifies key objects and matches them against a predefined list of wine-related items such as bottles, glasses, and ingredients to determine if the image content is relevant to the jackfruit wine theme. Finally, the system automatically merges the professionally judged text and relevant images, outputting a purified, high-quality text and image dataset. This allows the system to effectively remove vague descriptions and irrelevant images from the original reviews, retaining only professional content with substantial analytical value, providing a data foundation for subsequent in-depth analysis and prioritization.

[0079] Furthermore, the text professionalism determination unit includes: a text processing unit for receiving a text information stream and performing word segmentation and stop word removal processing; a keyword matching unit for matching the preprocessed text with a keyword library in the tasting field and calculating the keyword hit rate and weight; a syntactic structure analysis unit for analyzing the sentence structure, clause depth, and professional term collocations of the text based on a syntactic complexity model to generate a complexity score; and a professionalism determination unit for integrating the keyword hit rate, weight, and syntactic complexity score and outputting a determination result on whether the text belongs to a professional text according to a preset rule.

[0080] Specifically, the input review text is automatically segmented, and common words without actual meaning such as "de" and "le" are filtered to obtain a core vocabulary sequence. Subsequently, this sequence is compared with a specially constructed vocabulary library containing professional terms such as the flavor, process, and taste of jackfruit wine, the occurrence frequency of professional vocabulary is counted, and the importance differences of different terms are considered; at the same time, the grammatical structure of the sentence is analyzed to evaluate whether it contains complex sentence patterns and whether the collocations between professional terms conform to tasting logic, and a score reflecting the rigor of expression is given accordingly; finally, the above vocabulary matching result and the sentence pattern analysis score are integrated according to a preset algorithm to automatically determine whether this paragraph of text constitutes a professional tasting review, thereby distinguishing simple subjective evaluations from discussions containing specific technical details and accurately screening out professional texts with analytical value.

[0081] Furthermore, the image relevance determination unit includes: an image processing unit for receiving an image information stream and performing size normalization, noise filtering, and region enhancement processing; an identification and extraction unit for identifying the preprocessed image and extracting the visual feature vectors of the main objects in the image; an associated item comparison unit for calculating the similarity between the extracted visual feature vectors and the benchmark features of various items in a preset associated item list and outputting a matching degree score; and a relevance determination unit for receiving the matching degree score and comparing it with a preset threshold to determine whether the image is an image related to the wine product.

[0082] Specifically, upon receiving image data from reviews, the system automatically resizes the images and removes noise, while enhancing key areas that may contain bottles, labels, or wine. Next, a pre-trained deep learning model identifies the main objects in the images and converts them into digital vectors representing their visual characteristics. These vectors are then compared one by one with a pre-built standard database containing features of various bottles, glasses, ingredients, and brewing scenes, calculating the similarity score for each standard category. Finally, the highest score among these scores is compared to a set threshold. If the score exceeds this threshold, the image is considered relevant to the wine; otherwise, it is excluded. This automatic and accurate classification of images in reviews as relevant or irrelevant to the wine ensures that only images with practical reference value are retained.

[0083] In a preferred embodiment, the evaluation analysis module includes: a content clustering analysis unit, used to perform semantic clustering on professional text information to identify similar semantic statements, and to perform visual feature clustering on image information to identify images with homogeneous content; a commenter authentication unit, used to determine whether similar semantic statements and images with homogeneous content originate from real human comments by analyzing account behavior patterns and comment interaction characteristics; a geographic information verification unit, used to parse and verify the existence of real addresses from similar semantic statements; and a priority content selection unit, used to select priority content from content that simultaneously meets the conditions of human comments and contains valid real addresses, according to preset filtering rules.

[0084] Specifically, the system automatically clusters selected professional texts and images based on semantic and visual features, grouping comments expressing similar viewpoints or presenting similar content into the same topic group. Simultaneously, authenticity verification is performed: on one hand, by analyzing account login history, interaction patterns, and content generation characteristics, it automatically determines whether comments within each topic group originate from genuine user actions, rather than automated scripts or fake accounts; on the other hand, it automatically extracts geographical location information mentioned in the comment text and calls external trusted data sources for validity verification. Based on preset integration rules, only content from topic groups that pass both human comment verification and real address verification is prioritized as having high credibility and reference value. This automatically identifies high-quality content with consistent themes, authentic sources, and verifiable backgrounds, providing a reliable basis for subsequent responses and improvements.

[0085] Furthermore, the content clustering analysis unit includes: a feature vectorization unit, used to convert professional text information into text semantic vectors and image information into image visual feature vectors; a model optimization unit, which uses particle swarm optimization to search for optimal cluster centers or similarity thresholds for text semantic vectors and image visual feature vectors; a clustering execution unit, which performs clustering operations on text semantic vectors and image visual feature vectors respectively using the optimal cluster centers or thresholds to generate text semantic clusters and image visual clusters; and a clustering result mapping unit, used to map text semantic clusters back to the original set of similar semantic statements and image visual clusters back to the original set of content-homogeneous images.

[0086] Specifically, using natural language processing models and convolutional neural networks, comment text and images are converted into high-dimensional numerical vectors, forming a unified mathematical representation. A particle swarm optimization algorithm is used to search in parallel within the vector space, dynamically adjusting and evaluating the clustering effect of data under different cluster centers or similarity thresholds. Ultimately, the algorithm autonomously determines the parameter combination that maximizes the density of similar data and minimizes the separation of dissimilar data. Based on these optimized parameters, all text and image vectors are partitioned, automatically generating text and image clusters with similar internal features. Finally, each numerical cluster is accurately reconstructed into its corresponding original comment statement and image group through index association. This allows for the automatic identification and summarization of reliable datasets with common themes or content features from professional comments and images without relying on manually preset rules.

[0087] Furthermore, the model optimization unit execution steps include:

[0088] Step S21: Initialize the particle swarm by randomly assigning each particle an initial solution representing the cluster center coordinates or similarity threshold, and initialize the particle velocity.

[0089] Step S22: For each solution represented by a particle, calculate the text semantic vector and image visual feature vector based on the clustering of the solution, and calculate the fitness of the particle according to the intra-class tightness and inter-class separation.

[0090] Step S23: Update the velocity and position of each particle based on its current fitness and its historical best solution, as well as the global historical best solution of the entire particle swarm, thereby updating the solution it represents.

[0091] Step S24: Repeat steps S22 and S23 until the preset number of iterations or fitness convergence condition is met. Output the solution corresponding to the global historical best solution of the particle swarm as the optimal cluster center or similarity partitioning threshold.

[0092] In one example, within the possible parameter range, multiple sets of cluster center locations or similarity thresholds are randomly generated as initial candidate solutions. Each solution is assigned a random adjustment direction and step size, forming an initial search particle swarm. Each candidate solution is used to perform a complete clustering of all text and image feature vectors. Mathematical methods are used to calculate whether the average distance between data points of the same class is sufficiently close and whether the average distance between data points of different classes is sufficiently far apart, thus quantitatively evaluating the clustering quality of the solution set as its fitness score. After each round of evaluation, each particle dynamically adjusts its movement direction and distance by referring to its historical best solution and the best solution currently found by the entire particle swarm, thereby exploring better regions in the parameter space. This evaluation and movement process is repeated until the set maximum number of iterations is reached or the solutions of all particles no longer significantly improve. Finally, the best parameter combination recorded throughout the search process is determined as the optimal solution, autonomously finding and determining the most suitable division criteria for the professional comments and image data to be clustered, thereby more accurately and reasonably grouping comments with similar meanings and images with similar content together automatically.

[0093] In a preferred embodiment, the emotional response module includes: a response strategy generation unit, used to receive priority content and parse the interrogative sentences and improvement suggestions therein, and generate corresponding response scripts and improvement commitment frameworks respectively; a response content synthesis unit, used to match or construct specific response texts from a preset knowledge base based on the response scripts; a commitment tracking and management unit, used to register the improvement commitment frameworks as to-do items and associate them with the responsible parties and planned completion deadlines; and a unified response output unit, used to integrate the specific response texts and corresponding improvement commitments into a formatted final response and publish it.

[0094] Specifically, upon receiving priority comments filtered by the aforementioned modules, the system automatically analyzes the text structure, identifies specific questions and improvement suggestions, and generates preliminary response directions and commitment points. Based on the generated response directions, it searches and adapts the internally stored standard script library and product knowledge base, automatically combining them to generate a detailed response text. Simultaneously, the improvement commitment points are automatically converted into tracking task items with clearly defined responsible parties and completion deadlines. The generated detailed response text and the formatted commitment task descriptions are automatically merged to form a complete response, which is then pushed to the appropriate public or designated channels for publication. This ensures that user questions are addressed specifically and that improvement suggestions are transformed into trackable and implementable tasks, thus completing the management of user feedback.

[0095] 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, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A jackfruit wine tasting and review management system, characterized in that, It includes a text acquisition module, a text screening module, a merit analysis module, and a sentiment response module; The text acquisition module is used to acquire and update comment data in real time, and the comment data includes text information and image information; The text screening module is used to filter the comment data, removing general text information and irrelevant image information, and retaining professional text information and image information; The evaluation analysis module is used to compare and analyze the professional text and image information, identify sentences with similar semantics and images with similar content, and select priority content that contains real addresses and is human comment. The emotional response module is used to respond to the priority content, including answering questions and making improvement commitments based on suggestions.

2. The jackfruit wine tasting and review management system according to claim 1, characterized in that, The text acquisition module includes: The data interface unit is used to access the comment data in real time from multiple heterogeneous data sources; A data parsing unit is used to parse the comment data and separate the text information from the image information. The real-time deduplication unit is used to perform real-time deduplication processing on the parsed text and image information; The data storage and update unit is used to store the deduplicated comment data in the database and respond to update instructions to maintain the latest state of the data.

3. The jackfruit wine tasting and review management system according to claim 2, characterized in that, The real-time deduplication unit steps include: Step S11: Receive text information stream and image information stream from the data parsing unit; Step S12: Extract semantic feature vectors from the text information stream and extract visual feature vectors from the image information stream; Step S13: Perform a real-time similarity comparison between the feature vector of the newly received data and the feature vector of the stored data, and determine whether it is duplicate content based on a preset threshold. Step S14: If the content is determined to be duplicate, it is filtered out; if it is not duplicate, its feature vector is stored in the feature library and the data is output.

4. The jackfruit wine tasting and review management system according to claim 1, characterized in that, The text screening module includes: An information preprocessing unit is used to receive the comment data and separate and preprocess the text information stream and image information stream therein; The text professionalism determination unit analyzes the text information flow based on a preset keyword library for the tasting field and a syntactic complexity model; The image relevance determination unit is used to identify the main content in the image information stream through image recognition technology, and determine whether it is an image related to wine based on a preset list of related items. The compliance information integration unit is used to collect texts deemed professional and images deemed relevant, forming a filtered data set.

5. The jackfruit wine tasting and review management system according to claim 4, characterized in that, The text professionalism determination unit includes: A text processing unit is used to receive the text information stream and perform word segmentation and stop word removal. The keyword matching unit is used to match the preprocessed text with the keyword library of the tasting field and calculate the keyword hit rate and weight. The syntactic structure analysis unit, based on the syntactic complexity model, analyzes the sentence structure, clause depth, and collocation of technical terms in the text, and generates a complexity score. The professionalism determination unit is used to integrate the keyword hit rate, weight, and syntactic complexity score, and output the determination result of whether the text belongs to professional text according to preset rules.

6. The jackfruit wine tasting and review management system according to claim 4, characterized in that, The image correlation determination unit includes: An image processing unit is used to receive the image information stream and perform size normalization, noise filtering, and region enhancement processing. The recognition and extraction unit is used to recognize the preprocessed image and extract the visual feature vectors of the main objects in the image. The associated item comparison unit is used to calculate the similarity between the extracted visual feature vector and the benchmark features of various categories in the preset associated item list, and output the matching score. A relevance determination unit is used to receive the matching score and compare it with a preset threshold to determine whether the image is related to wine.

7. The jackfruit wine tasting and review management system according to claim 1, characterized in that, The performance evaluation analysis module includes: The content clustering analysis unit is used to perform semantic clustering on the professional text information to identify similar semantic statements, and to perform visual feature clustering on the image information to identify images with homogeneous content. The commenter authentication unit is used to determine whether the similar semantic statements and content-homogeneous images originate from real human comments by analyzing account behavior patterns and comment interaction characteristics. A geographic information verification unit is used to parse and verify the existence of the real address from the semantic statements of the same type; The priority content selection unit is used to select priority content from content that simultaneously meets the conditions of human comments and contains a valid real address, according to preset filtering rules.

8. The jackfruit wine tasting and review management system according to claim 7, characterized in that, The content clustering analysis unit includes: The feature vectorization unit is used to convert the specialized text information into text semantic vectors and the image information into image visual feature vectors. The model optimization unit uses the particle swarm optimization algorithm to search for the optimal cluster center or similarity threshold for the text semantic vector and the image visual feature vector; The clustering execution unit performs clustering operations on the text semantic vector and the image visual feature vector, respectively, by the optimal cluster center or the partitioning threshold, to generate text semantic clusters and image visual clusters; The clustering result mapping unit is used to map the text semantic clusters back to the original set of similar semantic statements, and to map the image visual clusters back to the original set of content-homogeneous images.

9. A jackfruit wine tasting and review management system according to claim 8, characterized in that, The model optimization unit execution steps include: Step S21: Initialize the particle swarm by randomly assigning each particle an initial solution representing the cluster center coordinates or similarity threshold, and initialize the particle velocity. Step S22: For each solution represented by a particle, calculate the clustering of the text semantic vector and image visual feature vector based on the solution, and calculate the fitness of the particle according to the intra-class tightness and inter-class separation. Step S23: Update the velocity and position of each particle based on its current fitness and its historical best solution, as well as the global historical best solution of the entire particle swarm, thereby updating the solution it represents. Step S24: Repeat steps S22 and S23 until the preset number of iterations or fitness convergence condition is met. Output the solution corresponding to the global historical best solution of the particle swarm as the optimal cluster center or similarity partitioning threshold.

10. A jackfruit wine tasting and review management system according to claim 1, characterized in that, The emotional response module includes: The response strategy generation unit is used to receive the priority content, parse the question sentences and improvement suggestions therein, and generate corresponding response scripts and improvement commitment frameworks respectively. The response content synthesis unit matches or constructs specific response text from a preset knowledge base based on the response script; The commitment tracking and management unit is used to register the improved commitment framework as a to-do item and associate it with the responsible party and the planned completion deadline; A unified response output unit is used to integrate specific response texts and corresponding improvement commitments into a formatted final response, which is then published.

Images