Analysis report generation method and system based on comment data

Abstract

The invention provides an analysis report generation method and system based on vehicle model comment data, and belongs to the technical field of vehicle data analysis. The method comprises the following steps: constructing a file-level thread pool, distributing to-be-processed vehicle model comment files to different threads of the thread pool, calling a large model in parallel based on the different threads, and performing multi-dimensional comment feature extraction on comment data in the vehicle model comment files by utilizing the large model to obtain initial tags; judging whether to call a large model to obtain a new label or not by combining confidence, and generating an analysis report based on the initial label and the new label. According to the embodiment of the invention, the large model is called in parallel through multiple threads, the multi-source data from the comment files of different vehicle types are effectively integrated, and the accuracy and processing efficiency of the analysis report are remarkably improved through multi-dimensional feature extraction, automatic comment data annotation and generation of the analysis report.

Description

Technical Field

[0001] This invention relates to the field of vehicle data analysis technology, and in particular to a method and system for generating analysis reports based on comment data. Background Technology

[0002] In an environment where the internet and mobile devices are ubiquitous, various platforms (such as e-commerce, social media, and forums) continuously generate large-scale, multi-source text comments and discussions, which contain rich information on user sentiment and trending topics. Existing comment data analysis systems typically rely on sentiment dictionaries or traditional machine learning models (such as SVM and Naive Bayes) to perform coarse-grained classification of data from a single source, or train multiple small models for different dimensions and scenarios, making it difficult to achieve cross-platform, integrated, and real-time sentiment aggregation and trend mining. At the same time, most solutions still require manual definition of rules and establishment of indicators, and rely on commercial analysis tools to complete visualization reports, resulting in a cumbersome overall process with high human involvement and poor scalability.

[0003] Therefore, how to accurately analyze review data and generate relevant analysis reports to accurately obtain vehicle evaluation information and improve user experience is an urgent problem to be solved. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art, and proposes a method and system for backing up vehicle mileage data.

[0005] In a first aspect, embodiments of the present invention provide a method for generating an analysis report based on vehicle model review data, the method comprising:

[0006] Obtain the review files for each vehicle model to be processed;

[0007] A file-level thread pool is constructed, and the review files of each car model to be processed are distributed to different threads in the thread pool. The large model is called in parallel based on the different threads. The large model extracts multi-dimensional review features from the review data in each car model review file. The review data is labeled based on the multi-dimensional review features to obtain the initial label corresponding to each review data.

[0008] For each initial label, a local fuzzy match is performed between the initial label and a predefined set of candidate labels to obtain the confidence score corresponding to the initial label. If the confidence score is higher than a preset threshold, the initial label is retained. Otherwise, the large model is called again to perform semantic mapping on the initial label to obtain a new label.

[0009] Based on all retained initial and new tags, data analysis is performed on each comment data, and analysis reports for each car model are generated based on the analysis results.

[0010] Optionally, obtaining the review files for each vehicle model to be processed includes:

[0011] Initial comment data was collected from various vehicle data platforms, and the collected initial comment data was converted into a unified format to obtain various initial comment files;

[0012] Load each initial comment file, deduplicate each comment in each initial comment file, generate a unique comment identifier for each comment, and record basic statistical indicators for each initial comment file; wherein, the basic statistical indicators include at least one of the following: total number of original comments, number of deduplicated comments, number of invalid short comments, and effective comment rate;

[0013] Each comment identifier and basic statistical indicator is associated with its corresponding initial comment file and stored to obtain the comment files for each vehicle model to be processed.

[0014] Optionally, the step of allocating the review files of each car model to be processed to different threads in the thread pool, and calling the large model in parallel based on the different threads, includes:

[0015] The review files for each vehicle model to be processed are assigned to different threads for parallel execution based on the vehicle model key; each review file contains a vehicle model key, which is used to identify the vehicle model.

[0016] During the execution of each thread, the thread independently initializes the annotation agent, which splits the car model review files received by the thread into different review processing tasks and assigns different processing tasks to different sub-threads according to task type;

[0017] The annotation agent calls the large model in parallel for different sub-threads to achieve a high degree of parallelism in the comment annotation process.

[0018] Optionally, the step of performing local fuzzy matching between the initial label and a predefined set of candidate labels to obtain the confidence level corresponding to the initial label includes:

[0019] For each candidate label in the predefined candidate label set, the edit distance between the initial label and the candidate label is calculated based on the Levenstein distance algorithm, and the string lengths of the initial label and the candidate label are calculated to determine the one with the larger string length; based on the larger string length and the edit distance, the confidence level between the initial label and the candidate label is obtained.

[0020] Based on the confidence level between the initial label and each candidate label, the confidence level corresponding to the initial label is determined.

[0021] Optionally, obtaining the confidence score between the initial tag and the candidate tag based on the larger of the string length and the edit distance includes:

[0022] The confidence level between the initial label and the candidate label is calculated using the following formula:

[0023]

[0024] in, This represents the confidence level between the initial label s and the candidate label c. This indicates the edit distance between the initial label and the candidate label; This indicates the length of the larger string.

[0025] Optionally, the method further includes:

[0026] After all threads have finished executing, for each new tag, the annotation agent classifies and outputs the new tag set for each car model according to the car model key contained in the comment data corresponding to the new tag.

[0027] Obtain the manual review results for the new tag set, and determine whether to update each new tag in the new tag set to the candidate tag set based on the manual review results.

[0028] Optionally, based on all retained initial tags and new tags, data analysis is performed on each comment data, and an analysis report for each vehicle model is generated based on the analysis results, including:

[0029] Based on all the retained initial tags and new tags, construct a summary table of comment information for each comment data;

[0030] Based on the aforementioned comment information summary table, a drawing agent is invoked to generate a label map for each comment data, and a data analysis agent is invoked to perform data analysis on each comment data to obtain the analysis results;

[0031] An analysis report for each vehicle model is generated based on the label map and the analysis results.

[0032] Secondly, embodiments of the present invention provide an analysis report generation system based on vehicle model review data, the system comprising:

[0033] The acquisition module is used to acquire review files for each car model to be processed;

[0034] The annotation module is used to build a file-level thread pool, which distributes the review files of each car model to be processed to different threads in the thread pool. Based on the different threads, the large model is called in parallel. The large model extracts multi-dimensional review features from the review data in each car model review file, and annotates each review data based on the multi-dimensional review features to obtain the initial label corresponding to each review data.

[0035] The matching module is used to perform local fuzzy matching between each initial label and a predefined set of candidate labels to obtain the confidence score corresponding to the initial label. If the confidence score is higher than a preset threshold, the initial label is retained; otherwise, the large model is called again to perform semantic mapping on the initial label to obtain a new label.

[0036] The generation module is used to perform data analysis on each comment data based on all retained initial tags and new tags, and generate analysis reports for each car model based on the analysis results.

[0037] Thirdly, embodiments of the present invention provide an electronic device, including:

[0038] One or more processors;

[0039] Memory, used to store one or more programs;

[0040] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the methods described above.

[0041] Fourthly, embodiments of the present invention provide a computer-readable medium storing a computer program, which, when executed by a processor, implements the steps of any of the methods described above.

[0042] The method and system for generating analysis reports based on vehicle model review data provided in this invention constructs a file-level thread pool, distributing the vehicle model review files to be processed to different threads within the pool. A large model is invoked in parallel across these threads. This large model extracts multi-dimensional review features from the review data in each vehicle model review file to obtain initial labels. Confidence levels are then used to determine whether to invoke the large model again to obtain new labels. Finally, an analysis report is generated based on the initial and new labels. In summary, the analysis report generation method provided in this invention effectively integrates multi-source data from different vehicle model review files by using multi-threaded parallel invocation of a large model. Through multi-dimensional feature extraction, local fuzzy matching, and large model invocation, it automatically labels review data and generates analysis reports, significantly improving the accuracy and processing efficiency of the analysis reports. Attached Figure Description

[0043] Figure 1A flowchart illustrating a method for generating an analysis report based on vehicle model review data, provided in an embodiment of the present invention;

[0044] Figure 2 This is a flowchart illustrating an optional specific implementation method of step S11 in an embodiment of the present invention;

[0045] Figure 3 This is a flowchart illustrating an optional specific implementation method of step S12 in an embodiment of the present invention;

[0046] Figure 4 A structural block diagram of an analysis report generation system based on vehicle model review data provided in an embodiment of the present invention;

[0047] Figure 5 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0048] To enable those skilled in the art to better understand the technical solutions of the present invention, exemplary embodiments of the present invention are described below in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0049] Where there is no conflict, the various embodiments of the present invention and the features thereof may be combined with each other.

[0050] As used herein, the term “and / or” includes any and all combinations of one or more related enumerated entries.

[0051] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used herein, the singular forms “a” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “made of” are used in this specification, the presence of the stated feature, integral, step, operation, element, and / or component is specified, but the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded. Terms such as “connected” or “linked” are not limited to physical or mechanical connections but can include electrical connections, whether direct or indirect.

[0052] Unless otherwise specified, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having the meaning consistent with their meaning in the context of the relevant art and the invention, and will not be interpreted as having an idealized or overly formal meaning unless expressly so defined herein.

[0053] In the technical solution of this invention, the collection, storage, use, processing, transmission, provision, and disclosure of user personal information all comply with relevant laws and regulations and do not violate public order and good morals. The use of user data in this technical solution follows relevant national laws and regulations (e.g., the "Information Security Technology - Personal Information Security Specification"). For example: appropriate measures are taken for personal information access control; restrictions are imposed on the display of personal information; the purpose of using personal information does not exceed the scope of direct or reasonable association; and explicit identity targeting is eliminated when using personal information to avoid precisely locating a specific individual.

[0054] In related technologies, comment data analysis is a crucial step in generating accurate analysis reports. For example, prior art one, patent application CN115292674A, discloses a method and system for detecting fraudulent applications based on user comment data. This method primarily identifies comment information describing fraudulent behavior from massive amounts of user comments based on multi-keyword classification rules. Prior art two, patent application CN111858944A, mainly employs an entity-level sentiment analysis method based on an attention mechanism for comment data. It extracts attribute words through a historical perception attention mechanism and combines BERT and hierarchical attention mechanisms for sentiment analysis to achieve accuracy in comment data analysis.

[0055] The main drawbacks of the existing technology are: 1. Excessive human involvement in database construction. Current knowledge base construction heavily relies on manual annotation and feature extraction. Each addition or update requires experts to analyze each comment and formulate corresponding labeling rules. This manual screening and annotation process is not only time-consuming and labor-intensive, but also struggles to ensure consistency and comprehensive coverage. Furthermore, maintenance costs increase exponentially with the expansion of data volume. 2. Classification rules primarily rely on TF-IDF keyword extraction combined with enumeration-based classification rules, failing to capture deep semantics, contextual relationships, and sentiment in comments. Simple keyword statistics are easily affected by noise words, synonyms, and implicit expressions, leading to unstable classification accuracy and recall when facing diverse fraud methods. 3. Delayed knowledge base updates and slow response speed. The current system only feeds new features back to the knowledge base after detection, lacking real-time performance and online learning capabilities. Faced with rapidly evolving fraud methods, delayed knowledge base updates significantly reduce the detection model's ability to identify new techniques, and the response efficiency fails to meet the actual business needs for rapid fraud detection.

[0056] Existing technology two primarily relies on BERT for sentiment analysis of comment data, but it suffers from the following shortcomings: 1. Limited deep semantic understanding: While BERT possesses good language representation capabilities, it still has limitations in cross-sentence and cross-paragraph context modeling, making it difficult to fully capture sentiment tendencies in complex contexts. 2. Insufficient fine-grained sentiment recognition: For multi-dimensional, fine-grained sentiment expressions (such as sarcasm, double negation, degree adverbs, etc.), the BERT model's ability to distinguish and quantify is weak, resulting in inaccurate and unstable scoring. 3. High dependence on human intervention: Traditional dictionary matching and statistical feature extraction require extensive manual design and maintenance, leading to low efficiency and difficulty in adapting to new domains and vocabulary, making it difficult to maintain the model's rapid response to new scenarios. 4. Difficulty in integrating the prior advantages of multiple models: A single BERT model cannot fully integrate the advantages of traditional multi-model prior knowledge (such as dictionaries and rules) and lightweight feature engineering, resulting in insufficient information collaborative processing capabilities.

[0057] After discovering the above-mentioned shortcomings, the inventors of this application conducted thorough research and found that Large Language Model (LLM, hereinafter referred to as Large Model) has significant advantages in the field of natural language understanding and generation. However, there is still a lack of mature frameworks for how to integrate its capabilities into large-scale comment data analysis and generate analysis reports based on the analysis results to achieve multi-source sentiment and hot topic analysis for user reviews.

[0058] Based on this, this invention proposes a method for generating analysis reports based on review data. It extracts multi-dimensional review features from review data in various vehicle model review files using a large model, labels each review data point based on these features, and combines this with local fuzzy matching to obtain initial and new labels for each review data point. Finally, it analyzes the review data based on these initial and new labels to generate analysis reports for each vehicle model. This invention effectively integrates multi-source data from different vehicle model review files by using multi-threaded parallel calls to the large model. Through multi-dimensional analysis, it automatically generates high-quality hotspot analysis reports, significantly improving the method's flexibility and processing efficiency.

[0059] To address at least one of the technical problems existing in the aforementioned related technologies, this invention provides a method and system for generating analysis reports based on comment data. By using multi-threaded parallel calls to a large model, it effectively integrates multi-source data from comment files of different car models. Through multi-dimensional feature extraction, it automatically labels the comment data and generates analysis reports, significantly improving the accuracy and processing efficiency of the analysis reports.

[0060] Figure 1 This is a flowchart illustrating a method for generating an analysis report based on vehicle model review data, provided as an embodiment of the present invention. Figure 1 As shown, the analysis report generation method provided in this embodiment of the invention includes the following steps:

[0061] S11, retrieve the review files for each vehicle model to be processed;

[0062] Vehicle review files can include review data, which may include user experiences, expectations, and more for various vehicle models. Review data can be obtained from major vehicle data platforms or from keywords related to vehicle discussions on public platforms such as social media. Each vehicle review file can contain a vehicle model key to uniquely identify the vehicle model corresponding to that review file. Different vehicle models can include all models from the same brand or all models from different brands.

[0063] User reviews of various car models can be collected through a vehicle data platform, and then the review data of each car model can be integrated to form a review file for each car model.

[0064] The initial collection of comment data may contain a lot of redundant data. To improve processing efficiency, the collected comment data can be deduplicated before being integrated to form comment files for each car model.

[0065] In one embodiment of the present invention, such as Figure 2 As shown, step S11 specifically includes the following steps:

[0066] S111: Collect initial comment data from various vehicle data platforms, and convert the collected initial comment data into a unified format to obtain various initial comment files;

[0067] S112, Load each initial comment file, deduplicate each comment data in each initial comment file, generate a unique comment identifier for each comment data, and record basic statistical indicators for each initial comment file;

[0068] The basic statistical indicators include at least one of the following: total number of original comments, number of duplicate comments, number of invalid short comments, and effective comment rate;

[0069] S113. Associate and store each comment identifier, basic statistical indicator and its corresponding initial comment file to obtain the comment files for each vehicle model to be processed.

[0070] User review data for different car models can be obtained from various vehicle data platforms (such as Autohome and Car Enthusiast Forum) using legitimate web scraping tools. This initial review data can then be standardized and converted to a CSV (Comma-Separated Values) file format for easier data retrieval, and stored in a designated folder. Subsequently, the car model review data collected from various platforms can be retrieved in batches via a specified folder path.

[0071] During the data reading process, each comment can be deduplicated and a unique five-digit ID (identifier) ​​can be generated for subsequent tracking and association. Simultaneously, at least one or a combination of the following can be recorded: the original total number of comments, the number of deduplicated comments, the number of invalid short comments, and the effective comment rate. The original total number of comments represents the total number of comment data in the comment file before deduplication, and the number of deduplicated comments represents the total number of comment data in the comment file after deduplication. The number of invalid short comments represents the total number of comment data with fewer than a preset comment value, which can be adjusted according to actual needs, such as 0, 10, or 20 characters. The effective comment rate represents the proportion of comment data that can be used for effective analysis; it can be the ratio of the total number of comment data with at least a preset comment value to the number of deduplicated comments, or the ratio of the total number of comment data with at least a preset comment value to the original total number of comments. This invention does not limit this ratio.

[0072] For each initial comment file, after obtaining basic statistical indicators such as the total number of original comments, the number of duplicate comments, the number of invalid short comments, and the effective comment rate, as well as the comment identifier of each comment in the file, these data can be associated and stored with the file. For example, they can be recorded in a table and stored in a one-to-one correspondence.

[0073] S12, construct a file-level thread pool, allocate the review files of each car model to be processed to different threads in the thread pool, call the large model in parallel based on the different threads, extract multi-dimensional review features from the review data in each car model review file through the large model, label each review data based on the multi-dimensional review features, and obtain the initial label corresponding to each review data;

[0074] To fully utilize computing resources and achieve file-level parallel scheduling, this invention constructs a file-level thread pool, which distributes multiple review files for each vehicle model to be processed into different threads for parallel execution. This allows each thread to call the large model in parallel, thus accelerating data processing efficiency.

[0075] The large model can simultaneously capture features from multiple dimensions of comment data based on a multi-head attention mechanism, resulting in multi-dimensional comment features. These features can include multiple dimensions such as emotional intensity, question type, intent, and urgency. Furthermore, each comment data point can be labeled based on these multi-dimensional comment features to obtain the initial labels corresponding to each comment data point. Therefore, it is not necessary to model each dimension separately, and a complete multi-dimensional analysis result can be output in a single processing step.

[0076] In one specific implementation, such as Figure 3 As shown, step S12, which involves allocating the review files for each car model to be processed to different threads in the thread pool and then calling the large model in parallel based on these different threads, may include:

[0077] S121, Assign the unprocessed car model review files to different threads according to the car model key; whereby each car model review file contains a car model key; the car model key is used to identify the car model;

[0078] S122, During the execution of each thread, the thread independently initializes the annotation agent. The annotation agent splits the car model review file received by the thread into different review processing tasks, and assigns different processing tasks to different sub-threads according to the task type.

[0079] S123, the annotation agent calls the large model in parallel for different sub-threads to achieve a high degree of parallelism in the comment annotation process.

[0080] Each vehicle model review file can be in CSV format, allowing tasks to be assigned to different threads for parallel execution based on the vehicle model key. The vehicle model key (e.g., `model_key`) uniquely identifies the vehicle model. Each processing task corresponds to a `model_key` (e.g., "Toyota_Camry_2023" or "BYD_Han") and all associated CSV files. After submitting each `model_key` task to the thread pool, each thread can receive a vehicle model task, thus independently and completely executing all data processing steps for that vehicle model.

[0081] Each thread can independently initialize an annotation agent, and when the annotation agent is started, a sub-thread pool at the comment level can be automatically allocated, that is, each comment is split into different processing tasks to achieve a high degree of parallelism in the comment annotation process.

[0082] Specifically, within each file-level thread, when processing a car model begins, a comment-level sub-thread pool is automatically initialized. In practice, the size of this sub-pool can be optimized based on the amount of data for a single car model. All valid comments for that car model (from all its CSV files) are broken down into smaller task units and submitted to this sub-thread pool for parallel execution of the annotation agent, thus achieving parallel comment-level annotation.

[0083] Annotation agents can perform complex natural language processing operations on individual comments, such as: sentiment analysis (positive / negative / neutral); entity recognition (did the comments mention "fuel consumption", "interior", or "power"?); opinion extraction (does the user think "fuel consumption" is "high" or "low"?), and so on.

[0084] By applying the embodiments of this invention, two-level parallel processing is achieved through file-level and comment-level parallel processing, which improves processing efficiency. Moreover, the data volume of different models may vary greatly. The file-level thread pool can ensure that when one thread is processing a large number of models, other threads can process multiple models with smaller data volumes simultaneously, thus achieving load balancing. The processing of each model_key is independent, and the processing process between different models is independent. The annotation Agent is initialized in the comment-level thread, avoiding thread safety issues that may occur when multiple threads share a complex model, thus achieving modularity and isolation. In addition, the size of the two-level thread pool can be easily adjusted to adapt to different hardware and task requirements, providing high flexibility and ensuring that the system can process large-scale data with industrial-grade throughput and reliability.

[0085] S13, for each initial label, perform local fuzzy matching between the initial label and the predefined candidate label set to obtain the confidence level corresponding to the initial label; if the confidence level is higher than a preset threshold, retain the initial label; otherwise, call the large model again to perform semantic mapping on the initial label to obtain a new label;

[0086] The candidate tag set can be a pre-established local high-confidence tag library. This library can come from manually labeled tag datasets, or from a tag library accumulated after LLM and high-confidence threshold filtering in the early stages of processing. The text in the tag library can be stored in the form of text pairs, such as (comment data, tag), or other forms; this invention does not limit this. The preset threshold can be set according to actual needs; this invention does not limit this, for example, it can be 80, 85, 90, etc.

[0087] When processing a single car model review file in each thread, the platform identifier and car model key corresponding to the file can be extracted first. Then, all review texts are subjected to length filtering and local fuzzy permutation for deduplication. Invalid short reviews are removed by length filtering, thereby clearing noisy data, reducing unnecessary subsequent calculations, and improving overall processing efficiency. Local fuzzy permutation for deduplication identifies and removes highly repetitive reviews, further purifying the dataset and preventing duplicate viewpoints from having excessive weight in subsequent analysis. Then, the annotation agent is used to call the Large Model (LLM) in parallel to complete the initial annotation and obtain the initial labels. Then, local fuzzy matching is performed using databases such as RapidFuzz or difflib to quickly fuzz match the initial label of the current review with the candidate labels in the candidate label library. By calculating the confidence between the two, the matching is further verified. If the match is successful (the confidence exceeds a high threshold, such as 80%), the initial label is considered to have high confidence and can be retained without calling LLM again. Since the speed of local matching is much faster than calling LLM over the network, high-efficiency processing is achieved.

[0088] If the initial label does not match in the local matching or has insufficient confidence (e.g., confidence does not exceed 80%), LLM is called again for semantic mapping to obtain a new label; the annotation agent is designed with more accurate prompt words, and semantic mapping judgment is performed through LLM instead of generative analysis, thereby improving processing efficiency.

[0089] Finally, each comment and its initial or new tags can be written to JSON and output as a log, with processing time recorded in the log. Specifically, the final result of each comment (including the original text, comment_id, model_key, platform_id, and all generated tags) can be appended to the output file in JSON format to obtain the processing log. Furthermore, key performance indicators can be recorded in the processing log, such as the total time spent processing the CSV file, the total number of comments, the number of valid comments, the number of comments skipped by local matching, and the number of LLM calls, for monitoring system performance, analyzing cost composition, and troubleshooting.

[0090] Specifically, annotation agents initialized in a comment-level sub-thread pool can be used to call large models (such as GPT series, Qwen, etc.) in parallel to perform preliminary analysis on each comment. The annotation agents input prompt words into the large model to prompt the LLM to output structured labels and annotation content, such as:

[0091] Emotional tendency: positive, negative, neutral.

[0092] Discussion dimensions: power, interior, space, fuel consumption, intelligent features, price, etc.

[0093] Problem type: Information retrieval problem.

[0094] In another specific implementation, the step of performing local fuzzy matching between the initial label and a predefined set of candidate labels to obtain the confidence level corresponding to the initial label includes...

[0095] For each candidate label in the predefined candidate label set, the edit distance between the initial label and the candidate label is calculated based on the Levenstein distance algorithm, and the string lengths of the initial label and the candidate label are calculated to determine the one with the larger string length; based on the larger string length and the edit distance, the confidence level between the initial label and the candidate label is obtained.

[0096] Based on the confidence level between the initial label and each candidate label, the confidence level corresponding to the initial label is determined.

[0097] The Levenshtein distance algorithm can calculate the edit distance between the initial label and the candidate label, and further obtain the confidence between the initial label and the candidate label based on the larger string length and the edit distance.

[0098] The confidence level of the initial label is determined based on the confidence level between the initial label and each candidate label. This can be achieved by taking the largest confidence level between the initial label and all candidate labels as the confidence level of the initial label; or by other methods, which are not limited in this invention.

[0099] Specifically, obtaining the confidence score between the initial tag and the candidate tag based on the larger of the string length and the edit distance includes:

[0100] The confidence level between the initial label and the candidate label is calculated using the following formula:

[0101]

[0102] in, This represents the confidence level between the initial label s and the candidate label c. This indicates the edit distance between the initial label and the candidate label; The length of the larger string can be represented as: .

[0103] when (For example, in this invention, T=80) It is believed that It can be merged into .

[0104] S14: Based on all retained initial tags and new tags, perform data analysis on each comment data, and generate analysis reports for each car model based on the analysis results.

[0105] This invention leverages the rich performance of a large model to achieve unified extraction of multi-dimensional comment information, overcoming the redundancy, rigidity, and fragmentation inherent in traditional small models. Furthermore, this invention utilizes a large model based on a multi-head attention mechanism to simultaneously capture features across multiple dimensions of comment data, such as sentiment intensity, question type, intent, and urgency, eliminating the need for separate modeling for each dimension and outputting complete multi-dimensional analysis results in a single processing step. Compared to traditional methods, this invention's large model also supports flexible expansion of analysis dimensions (such as adding new tag types) through cue engineering or fine-tuning, thereby avoiding the costs of model reconstruction.

[0106] In summary, the analysis report generation method provided by this invention effectively integrates multi-source data from different vehicle model review files by using multi-threaded parallel calls to a large model, automatically labels review data through multi-dimensional feature extraction, and generates analysis reports, significantly improving the accuracy and processing efficiency of the analysis reports.

[0107] As another embodiment of the present invention, the method further includes:

[0108] After all threads have finished executing, for each new tag, the annotation agent classifies and outputs the new tag set for each car model according to the car model key contained in the comment data corresponding to the new tag.

[0109] Obtain the manual review results for the new tag set, and determine whether to update each new tag in the new tag set to the candidate tag set based on the manual review results.

[0110] The annotation agent can automatically collect new tags during the annotation process that cannot be matched with the predefined tag system, such as unknown intents, unknown question types, and unknown modules. After all threads are completed, the output can be uniformly sorted by vehicle model key, and the output format can be...<model_key> The `_unknown_summary.json` file facilitates subsequent manual review and expansion of the tagging system. It distinguishes data by vehicle model key (`model_key`), ensuring data independence for different vehicle models and facilitating targeted analysis. The output file becomes a high-quality data pool. Tagging experts or product managers can use these real, uncovered user cases to assess whether new tags need to be created. If so, the new tags are added to the candidate tag set, thereby iterating and optimizing the entire tagging system.

[0111] This invention is based on a large-scale pre-trained model, inheriting and integrating the advantages of traditional multi-small models in dictionary matching and statistical feature extraction. Simultaneously, it leverages the powerful capabilities of the large model in deep semantic understanding and contextual modeling to achieve higher-precision sentiment scoring. Specifically, this invention designs an annotation agent framework on top of the large model, combining the prior knowledge of each small model with the adaptive learning capabilities of the large model. This not only retains the efficient processing capabilities of traditional models and utilizes file-level and comment-level thread pools for parallel processing, but also significantly improves the ability to identify and quantify the sentiment tendencies of complex, multi-dimensional user comments, providing more accurate and stable scoring results for fine-grained sentiment expressions.

[0112] As another embodiment of the present invention, the step of performing data analysis on each comment data based on all retained initial tags and new tags, and generating analysis reports for each vehicle model based on the analysis results, includes:

[0113] Based on all the retained initial tags and new tags, construct a summary table of comment information for each comment data;

[0114] Based on the aforementioned comment information summary table, a drawing agent is invoked to generate a label map for each comment data, and a data analysis agent is invoked to perform data analysis on each comment data to obtain the analysis results;

[0115] An analysis report for each vehicle model is generated based on the label map and the analysis results.

[0116] The analysis mode is enabled by calling a data analysis agent. A summary table is built based on statistical data from all models and platforms, for example, named stats_summary.csv. An external plotting module, which can be a plotting agent, is called to generate a label chart of the comment data, such as an overall stacked bar chart. Finally, the data analysis results are combined to form a structured hotspot analysis report, which enables decision-makers to intuitively compare the core indicators such as the comment quality, deduplication, and effectiveness of each model on different channels.

[0117] By calling Agent (intelligent agent) architecture with different capabilities, each functional module (sentiment classification, topic extraction, statistical aggregation, report generation) is encapsulated into a reusable intelligent agent. Through unified scheduling and parallel execution, it effectively integrates multi-source data, multi-dimensional analysis and automatically generates high-quality hot spot analysis reports, significantly improving system flexibility and processing efficiency.

[0118] For example, the present invention will be described below in conjunction with specific application scenarios.

[0119] Step 1: Obtain the review files for each car model to be processed;

[0120] Step 2: Perform deduplication on the review files for each vehicle model to remove duplicate review content;

[0121] Step 3: Determine if there are any deduplicated car model review files containing fewer than 20 characters of text. If so, proceed to Step 4; otherwise, proceed to Step 5.

[0122] Step 4: Extract fields from the comment data and output them according to a preset fixed template. The fixed template can include file identifier (id), comment content (comment), and annotation. The annotation can specifically include intent, urgency, sentiment, summary, question type, and the module involved in the question. For example, the fixed output module can be: id:00016; comment: Supports automatic parking; annotation: (intent: consultation; urgency: low; sentiment: neutral; summary: supports automatic parking; question type: information search question; the module involved in the question: other).

[0123] Step 5: Use the large model to perform initial annotations and generate initial labels; the annotation results can be output by vehicle type.

[0124] Step 6: Perform local fuzzy matching using a fuzzy matching discriminator; if the confidence level of the initial label is higher than a preset threshold, retain the initial label; otherwise, call the large model again to perform semantic mapping on the initial label to obtain a new label.

[0125] Step 7: Based on all retained initial and new tags, perform data analysis on each comment data, and generate analysis reports for each car model based on the analysis results; for example, it could be: id:00034; comment: Dear car owners, when you drive, does the steering wheel veer to the right? I test drove the 2024 top-of-the-line four-wheel drive yesterday, and it kept veering to the right. The test driver said it was because it hadn't had a four-wheel alignment done for a long time; annotation: (Image search: Inquiry; Urgency: Medium; Mood: Neutral; Summary: User inquired about the steering wheel veering to the right during the test drive of the 2024 top-of-the-line four-wheel drive, and the test driver explained it as a four-wheel alignment issue; Problem type: Other; Problem-related modules: Other).

[0126] Step 8: Update the tag system based on the new tags.

[0127] The labels provided by the large model may not be label values ​​in the label system. For these "new" labels, fuzzy matching will be performed according to the label definition. If the match is successful, the label value defined in the label definition will be output. Otherwise, it will be regarded as a new label value, and the label system can be updated according to manual review.

[0128] Based on the same inventive concept, embodiments of the present invention also provide a system for generating analysis reports based on vehicle model review data. For example... Figure 4 As shown, it includes:

[0129] Module 41 is used to acquire review files for each car model to be processed;

[0130] The annotation module 42 is used to construct a file-level thread pool, allocate the review files of each car model to be processed to different threads in the thread pool, call the large model in parallel based on the different threads, extract multi-dimensional review features from the review data in each car model review file through the large model, and annotate each review data based on the multi-dimensional review features to obtain the initial label corresponding to each review data.

[0131] The matching module 43 is used to perform local fuzzy matching between each initial label and a predefined set of candidate labels to obtain the confidence level corresponding to the initial label; if the confidence level is higher than a preset threshold, the initial label is retained; otherwise, the large model is called again to perform semantic mapping on the initial label to obtain a new label.

[0132] The generation module 44 is used to perform data analysis on each comment data based on all the retained initial tags and new tags, and generate analysis reports for each car model based on the analysis results.

[0133] Optionally, the acquisition module is specifically used for: collecting initial comment data from each vehicle data platform, converting the collected initial comment data into a unified format to obtain each initial comment file; loading each initial comment file, deduplicating each comment data in each initial comment file, generating a unique comment identifier for each comment data, and recording basic statistical indicators for each initial comment file; wherein, the basic statistical indicators include at least one of the following: total number of original comments, number of deduplicated comments, number of invalid short comments, and effective comment rate; associating and storing each comment identifier, basic statistical indicators, and their corresponding initial comment files to obtain the comment files for each vehicle model to be processed.

[0134] Optionally, the annotation module distributes the review files of each car model to be processed to different threads in the thread pool, and calls the large model in parallel based on the different threads, specifically for:

[0135] The review files for each vehicle model to be processed are assigned to different threads for parallel execution based on the vehicle model key; each review file contains a vehicle model key, which is used to identify the vehicle model.

[0136] During the execution of each thread, the thread independently initializes the annotation agent, which splits the car model review files received by the thread into different review processing tasks and assigns different processing tasks to different sub-threads according to task type;

[0137] The annotation agent calls the large model in parallel for different sub-threads to achieve a high degree of parallelism in the comment annotation process.

[0138] Optionally, the matching module performs local fuzzy matching between the initial label and a predefined set of candidate labels to obtain the confidence level corresponding to the initial label. Specifically, it is used to: for each candidate label in the predefined set of candidate labels, calculate the edit distance between the initial label and the candidate label based on the Levinstein distance algorithm, and calculate the string lengths of the initial label and the candidate label to determine the one with the larger string length; obtain the confidence level between the initial label and the candidate label based on the larger string length and the edit distance; and determine the confidence level corresponding to the initial label based on the confidence level between the initial label and each candidate label.

[0139] Optionally, the matching module system obtains the confidence score between the initial tag and the candidate tag based on the larger string length and the edit distance, specifically for:

[0140] The confidence level between the initial label and the candidate label is calculated using the following formula:

[0141]

[0142] in, This represents the confidence level between the initial label s and the candidate label c. This indicates the edit distance between the initial label and the candidate label; This indicates the length of the larger string.

[0143] Optionally, the system further includes:

[0144] The classification output module is used to classify and output each new tag according to the vehicle model key contained in the comment data corresponding to the new tag after all threads have finished executing, so as to obtain a new tag set for each vehicle model.

[0145] The update module is used to obtain the manual review results for the new tag set, and determine whether to update each new tag in the new tag set to the candidate tag set based on the manual review results.

[0146] Optionally, the generation module performs data analysis on each comment data based on all retained initial tags and new tags, and generates an analysis report for each car model based on the analysis results, specifically used for:

[0147] Based on all the retained initial tags and new tags, construct a summary table of comment information for each comment data;

[0148] Based on the aforementioned comment information summary table, a drawing agent is invoked to generate a label map for each comment data, and a data analysis agent is invoked to perform data analysis on each comment data to obtain the analysis results;

[0149] An analysis report for each vehicle model is generated based on the label map and the analysis results.

[0150] Based on the same inventive concept, embodiments of the present invention also provide an electronic device. Figure 5 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. Figure 5 As shown, an embodiment of the present invention provides an electronic device including: one or more processors 101, a memory 102, and one or more I / O interfaces 103. The memory 102 stores one or more programs, which, when executed by the one or more processors, enable the one or more processors to implement the analysis report generation method based on vehicle model review data as described in any of the above embodiments; the one or more I / O interfaces 103 are connected between the processor and the memory, configured to enable information interaction between the processor and the memory.

[0151] The processor 101 is a device with data processing capabilities, including but not limited to a central processing unit (CPU); the memory 102 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and flash memory (FLASH); the I / O interface (read / write interface) 103 is connected between the processor 101 and the memory 102, and can realize information interaction between the processor 101 and the memory 102, including but not limited to a data bus (Bus).

[0152] In some embodiments, the processor 101, memory 102, and I / O interface 103 are interconnected via bus 104, and thus connected to other components of the computing device.

[0153] In some embodiments, the one or more processors 101 include a field-programmable gate array.

[0154] This invention also provides a computer-readable medium. The computer-readable medium stores a computer program, which, when executed by a processor, implements the steps in the analysis report generation method based on vehicle model review data as described in any of the above embodiments. The computer-readable storage medium can be volatile or non-volatile.

[0155] This invention also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code. When the computer-readable code is run in the processor of an electronic device, the processor in the electronic device executes the above-described method for generating analysis reports based on vehicle model review data.

[0156] Those skilled in the art will understand that all or some of the steps, systems, and apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software can be distributed on a computer-readable storage medium, which may include computer storage media (or non-transitory media) and communication media (or transient media).

[0157] As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable program instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), flash memory or other memory technologies, portable compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is known to those skilled in the art that communication media typically contain computer-readable program instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0158] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0159] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.

[0160] The computer program product described herein can be implemented specifically through hardware, software, or a combination thereof. In one alternative embodiment, the computer program product is specifically embodied in a computer storage medium; in another alternative embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0161] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0162] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0163] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0164] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0165] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in conjunction with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in conjunction with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of the invention as set forth in the appended claims.

Claims

1. A method for generating an analysis report based on vehicle model review data, characterized by, The method comprises: acquiring each vehicle model review file to be processed; constructing a file-level thread pool, distributing each vehicle model review file to be processed to different threads of the thread pool, calling a large model in parallel based on the different threads, performing multi-dimensional review feature extraction on review data in each vehicle model review file by the large model, labeling each review data based on the multi-dimensional review feature to obtain an initial label corresponding to each review data; for each initial label, performing local fuzzy matching between the initial label and a predefined candidate label set to obtain a confidence degree corresponding to the initial label; if the confidence degree is higher than a preset threshold, the initial label is retained; otherwise, the large model is called again to perform semantic mapping on the initial label to obtain a new label; based on all retained initial labels and new labels, performing data analysis on each review data, and generating an analysis report of each vehicle model based on the analysis result.

2. The method of claim 1, wherein, The acquisition of each vehicle model review file to be processed comprises: collecting initial review data on each vehicle data platform, and performing unified format conversion on the collected initial review data to obtain each initial review file; loading each initial review file, performing deduplication processing on each review data in each initial review file, generating a unique review identifier for each review data, and recording basic statistical indicators for each initial review file; wherein the basic statistical indicators comprise at least one of the following: total number of original reviews, number of reviews after deduplication, number of short review invalid reviews, and effective review rate; storing each review identifier, basic statistical indicators, and the initial review file corresponding thereto in association to obtain each vehicle model review file to be processed.

3. The method of claim 1, wherein, The distribution of each vehicle model review file to be processed to different threads of the thread pool and the parallel calling of the large model based on the different threads comprise: distributing each vehicle model review file to be processed to different threads according to vehicle model keys; wherein each vehicle model review file contains a vehicle model key, which is used to identify a vehicle model; in the execution process of each thread, the thread independently initializes an annotation intelligent agent, which splits the vehicle model review file received by the thread into different review processing tasks and distributes different processing tasks to different sub-threads according to task types; The annotation intelligent agent calls the large model in parallel for different sub-threads to realize high parallelization of the review labeling process.

4. The method of claim 1, wherein, The local fuzzy matching of the initial label with the predefined candidate label set to obtain the confidence degree corresponding to the initial label comprises. For each candidate label in the predefined candidate label set, based on the Levenshtein distance algorithm, the edit distance between the initial label and the candidate label is calculated, and the string length of the initial label and the candidate label is calculated to determine the larger one; based on the larger one and the edit distance, the confidence degree between the initial label and the candidate label is obtained; based on the confidence degree between the initial label and each candidate label, the confidence degree corresponding to the initial label is determined.

5. The method of claim 4, wherein, The confidence degree between the initial label and the candidate label based on the larger one and the edit distance comprises: The confidence between the initial label and the candidate label is calculated based on the following formula: wherein, represents a confidence between the initial label s and the candidate label c, represents an edit distance between the initial label and the candidate label; represents a string length of a string of which length is greater.

6. The method of claim 3, wherein, The method further includes: After all thread executions are completed, for each new label, the annotation agent classifies and outputs the vehicle model key contained in the comment data corresponding to the new label, to obtain a new label set of each vehicle model; Obtain the manual review result of the new label set, and determine whether to update each new label in the new label set to the candidate label set based on the manual review result.

7. The method of claim 1, wherein, Based on the remaining all initial labels and new labels, data analysis is performed on each comment data, and an analysis report of each vehicle model is generated based on the analysis result, including: Based on the remaining all initial labels and new labels, a comment information summary table of each comment data is constructed; Based on the comment information summary table, a drawing agent is called to generate a label graph of each comment data, and a data analysis agent is called to perform data analysis on each comment data to obtain an analysis result; Based on the label graph and the analysis result, an analysis report of each vehicle model is generated.

8. An analysis report generation system based on vehicle model review data, characterized by, The system includes: An acquisition module is configured to acquire vehicle model comment files to be processed; An annotation module is configured to construct a file-level thread pool, distribute the vehicle model comment files to be processed to different threads of the thread pool, call a large model in parallel based on the different threads, perform multi-dimensional comment feature extraction on comment data in each vehicle model comment file by using the large model, annotate each comment data based on the multi-dimensional comment features, and obtain initial labels corresponding to each comment data; A matching module is configured to, for each initial label, perform local fuzzy matching between the initial label and a predefined candidate label set to obtain a confidence of the initial label; if the confidence is higher than a preset threshold, the initial label is retained; otherwise, the large model is called again to perform semantic mapping on the initial label to obtain a new label; A generation module is configured to, based on the remaining all initial labels and new labels, perform data analysis on each comment data, and generate an analysis report of each vehicle model based on the analysis result.

9. An electronic device, comprising: It includes: One or more processors; Memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1-7.

10. A computer readable medium having stored thereon a computer program, characterized in that The computer program, when executed by a processor, implements the steps in the method of any one of claims 1-7.

Images