Government affair complaint intelligent service platform based on big language model

Abstract

The invention discloses a government affair complaint intelligent service platform based on a large language model, and belongs to the field of government affair complaint processing. The preprocessing module is connected with the complaint receiving module; the large language model processing module is connected with the preprocessing module; the importance evaluation module is connected with the large language model processing module; the processing distribution module is connected with the importance evaluation module; the feedback module is connected with the processing distribution module; according to the invention, the complaint receiving module supports multi-mode input of characters, voices and pictures, and covers most common dialect scenes in combination with an OCR technology and a mainstream dialect voice recognition API; the preprocessing module effectively improves the voice or picture recognition accuracy through joint correction of a grammar error correction model and a large language model, and solves the problem that dialect expression cannot be understood.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of government affair complaint processing, and particularly relates to a government affair complaint intelligent service platform based on a large language model. BACKGROUND

[0002] With the promotion of digital transformation of government affairs, processing mass complaints through intelligent platforms has become an important way to improve the efficiency of government services. However, the existing government complaint processing system has the following shortcomings: 1. Insufficient complaint text processing capability: Traditional systems rely on rule engines or simple machine learning models, which have insufficient understanding of complex language forms (such as dialects and colloquial expressions). The existing system often misjudges the semantics due to the inability to accurately identify dialect expressions, affecting the subsequent processing flow.

[0003] 2. Lack of importance evaluation mechanism: There is a lack of multi-dimensional quantitative evaluation of complaint information, relying only on manual annotation or simple keyword matching to divide priorities, which cannot comprehensively consider the urgency, impact range and emotional tendency of complaints, and may cause high-priority complaint response delays and excessive resource occupation for low-priority matters. 3. Low efficiency of processing allocation: The complaint allocation strategy is fixed and does not dynamically adjust in combination with the real-time processing load of government departments and preset time limit requirements, which may result in contradictions such as "high importance complaint queuing" and "low load department resource idling". SUMMARY

[0004] In view of the problems in the prior art, the application provides a government affair complaint intelligent service platform based on a large language model. TECHNICAL SCHEME

[0005] To solve the above problems, the application adopts the following technical scheme.

[0006] A government affair complaint intelligent service platform based on a large language model, comprising: a complaint receiving module for receiving user input complaint text, the complaint text including complaint content submitted by the user through text input, voice input and picture text recognition; a preprocessing module connected with the complaint receiving module for preprocessing the received complaint text, including word segmentation, stop word removal, grammar correction of the complaint text, conversion of voice input to text content, and correction processing of picture text recognition content; The large language model processing module is connected with the preprocessing module, utilizes a large language model trained in advance to perform semantic understanding on the preprocessed complaint text, the large language model can process various language forms including dialects, identify dialect expressions in the complaint text and convert them into standard languages for understanding, extract key information in the complaint text, and the key information includes a complaint object, a complaint event, a complaint time, a related location, and a problem description. The importance evaluation module is connected with the large language model processing module, and evaluates the importance of the complaint information according to the semantic understanding result, the evaluation is based on preset rules or an evaluation model obtained through training, including the emergency degree of the complaint event, the number of people involved, the possible impact range and severity, and generates an importance level. The processing allocation module is connected with the importance evaluation module, and allocates the complaint information according to the importance level, for complaint information with a high importance level, preferentially allocates to an artificial customer service for processing or marks as a key processing matter, and for complaint information with a low importance level, preliminarily replies according to a preset reply template and rules by an intelligent customer service. The feedback module is connected with the processing allocation module, and is used for feeding back the processing result to the user and receiving feedback opinions of the user, and records and stores the processing process and result.

[0007] Preferably, the training data of the large language model includes historical government complaint texts, public government files, and common dialect corpora, and is trained through supervised learning, unsupervised learning, and reinforcement learning methods to improve the understanding ability of the government complaint text and dialect.

[0008] Further, the preprocessing module further includes a function of performing sentiment analysis on the complaint text, identifying the sentiment tendency of the user when complaining, and providing the sentiment tendency information to the large language model processing module and the importance evaluation module as a reference.

[0009] Preferably, the preset rules in the importance evaluation module include: setting high importance levels for events involving life and property safety, major public interests, and emergency situations; and setting low importance levels for events involving general service attitude and minor problems.

[0010] Further, the processing allocation module is further connected with a business system of a government department, and can push the complaint information to the corresponding government department for processing according to the complaint object and complaint content, and track the processing progress.

[0011] Further, the feedback module provides multiple feedback modes, including short message notification, platform message reminder, and electronic mail, to facilitate the user to obtain the processing result in time.

[0012] Preferably, the system also includes a data statistical analysis module, which is connected to the feedback module and the processing allocation module to perform statistical analysis on the handling of complaint information, generate statistical reports, and provide data support for government departments to optimize services. The statistical analysis includes the number of complaints, processing time, and user satisfaction indicators.

[0013] Furthermore, the complaint receiving module provides a user interface where users can view the complaint processing progress and submit supplementary information.

[0014] Furthermore, the importance assessment module also integrates a historical complaint processing database, which can dynamically adjust the weight parameters of the preset rules or assessment model based on the processing results of historical complaints and user feedback data, thereby achieving adaptive assessment of the importance of complaint information.

[0015] Furthermore, the processing allocation module incorporates a priority queue scheduling algorithm, which can dynamically adjust the allocation order of complaint information based on the importance level of the complaint information, the real-time processing load of government departments, and preset processing time requirements, ensuring that the processing response time of high-importance complaint information does not exceed a preset threshold.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention supports multimodal input of text, voice, and images through a complaint receiving module. It combines OCR technology and mainstream dialect speech recognition APIs to cover most common dialect scenarios. The preprocessing module effectively improves the accuracy of speech or image recognition through joint correction by a grammar error correction model and a large language model, solving the problem of incomprehensible dialect expressions.

[0017] This invention integrates preset rules and machine learning models through an importance assessment module, and dynamically adjusts the weights of the assessment model based on historical processing data to achieve accurate quantification of the importance of complaints; the processing allocation module adopts a dual-queue preemptive scheduling + load balancing algorithm to effectively shorten the processing waiting time.

[0018] This invention connects the processing and allocation module with the business systems of multiple government departments through API interfaces, automatically matching the complainant with the responsible department and tracking the processing progress; the data statistical analysis results can be shared across departments, facilitating collaborative solutions to cross-domain problems. Attached Figure Description

[0019] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of this application, but not all embodiments. Generally, the components of the embodiments of this application described and shown in the accompanying drawings can be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application. Example

[0022] like Figure 1 As shown, a government complaint intelligent service platform based on a large language model includes: Complaint receiving module: This module provides a user interface implemented using web development technologies, accessible via computers, mobile phones, and other terminal devices. The interface includes text input boxes, voice input buttons, and image upload buttons. For text input, it directly receives the user's text content. For voice input, it calls a speech recognition API to convert speech into initial text. For image text recognition, after the user uploads an image containing the complaint content, OCR technology is used to recognize the text in the image, obtaining the image text. Users can view the complaint processing progress on the interface, specifically by setting up a processing progress display area to show the current processing stage of the complaint in real time. Simultaneously, users can submit supplementary information on the interface, with a supplementary information input box and a submit button. After entering supplementary information and clicking submit, the supplementary information will be associated with the original complaint text.

[0023] Preprocessing module: Word segmentation and stop word removal: An open-source word segmentation tool was used to segment the complaint text, resulting in a word list. Then, stop words were removed from the word list based on a preset stop word list. Syntax correction: Using a syntax correction model to perform syntax correction on the text after word segmentation and stop word removal, correcting grammatical errors such as typos and incorrect word order. Speech conversion and image text correction: For the initial text converted from speech input, further correction is performed by combining context and language model to improve the accuracy of the text; for image text recognition content, the incorrectly recognized text is corrected by comparing it with the preliminary semantic understanding results of the large language model processing module. Sentiment Analysis: Sentiment analysis algorithms are used to analyze the complaint texts to identify the emotional tendencies of users when they complain, classifying them into positive, neutral, negative, etc. The sentiment information is then provided as labels to the large language model processing module and the importance assessment module for reference.

[0024] Large Language Model Processing Module: Training data includes historical government complaint texts (obtained from historical complaint records of government departments, including complaint texts from different years, regions, and types), publicly available government documents (such as government-issued policies, regulations, and service guides), and a corpus of common dialects (collecting dialect words and sentences from various regions, including but not limited to common dialects such as Cantonese, Sichuanese, and Northeastern Mandarin). Training methods: The model is trained using supervised learning, unsupervised learning, and reinforcement learning. In supervised learning, labeled government complaint text data is used to train the model for tasks such as classification and named entity recognition, enabling the model to accurately extract key information such as the complaint target, complaint event, complaint time, location involved, and problem description. In unsupervised learning, a large amount of unlabeled text data is used to allow the model to learn the general features and rules of language. In reinforcement learning, a reward mechanism is set up to reward or punish the model based on its performance in processing actual complaint texts (such as the accuracy of key information extraction and the correctness of dialect conversion), thereby optimizing model parameters and improving its ability to understand government complaint texts and dialects. Dialect processing: When a complaint text containing dialect expressions is received, the large language model first identifies dialect words and sentence structure, and uses a pre-trained dialect conversion model (which can be based on statistical machine translation or neural network machine translation technology) to convert the dialect expressions into standard language, and then performs semantic understanding and key information extraction.

[0025] Importance assessment module: Preset rules: High importance level: involving life and property safety (such as complaints about buildings that are in danger of collapsing or food safety hazards that lead to poisoning), major public interests, and emergencies (such as sudden public safety incidents or emergencies caused by natural disasters). Low importance level: Complaints involving general service attitude and minor issues (such as complaints about minor damage to public facilities that do not affect normal use). Evaluation Model: A machine learning model (such as logistic regression, random forest, gradient boosting tree) is used as the evaluation model. It is trained using historical complaint data. The input includes features such as the urgency of the complaint, the number of people involved, the potential scope and severity of the impact, and the sentiment. The output is the importance level. Dynamic adjustment: Integrate a historical complaint handling database, which stores the handling results of historical complaints (such as handling methods, handling time, and handling effects) and user feedback data (such as user satisfaction ratings of the handling results); periodically adjust the weight parameters of the evaluation model based on historical data using adaptive algorithms (such as online learning algorithms).

[0026] Allocation processing module: Allocation Strategy: A built-in priority queue scheduling algorithm is used, such as a hybrid scheduling algorithm based on importance level, real-time processing load of government departments, and preset processing time requirements. Specifically, complaint information is placed into a priority queue from highest to lowest importance. For complaints with the same importance level, allocation is based on the real-time processing load of government departments, prioritizing departments with lower processing loads. Simultaneously, it ensures that the processing response time for high-importance complaints does not exceed a preset threshold (e.g., a preset response time of 1 hour for high-importance complaints). When the threshold is exceeded, an early warning mechanism is triggered to remind relevant departments to expedite processing. Connection with government department business systems: The system connects to the business systems of various government departments via API interfaces. Based on the complainant and the content of the complaint, the system categorizes and pushes complaint information to the appropriate government department for processing. For example, if the complainant is a company within the jurisdiction of the market supervision department, and the complaint concerns a product quality issue, the complaint information will be pushed to the market supervision department's business system. Simultaneously, the system tracks the processing progress in real time through the interface and obtains feedback on the processing status from the government departments.

[0027] Feedback module: Multiple feedback methods are available: SMS notification: The processing result will be sent to the mobile phone number provided by the user during registration via SMS through the SMS gateway interface. In-platform message notifications: The message center in the user interface displays message notifications of the processing results. Users can view the details after logging into the platform. Email: The processing results will be sent to the email address provided by the user during registration. At the same time, the system receives user feedback. Users can enter their opinions and suggestions on the processing and results in the feedback interface. The system records and stores the processing and results in the historical complaint processing database.

[0028] Data statistical analysis module: It connects to the feedback module and the processing and allocation module to perform statistical analysis on the processing of complaint information: Number of complaints: The number of complaints is counted by time, target of complaint, type of complaint, etc., and a complaint count report is generated. Processing time: Calculate the time interval from the time the complaint is received to the time the processing is completed, and generate a statistical report on the average processing time and the distribution of processing time by dimensions such as importance level and government department. User satisfaction metrics: Calculate user satisfaction scores based on user feedback and generate user satisfaction statistical reports.

[0029] These statistical reports provide data support for government departments to optimize services. For example, if a government department takes a long time to process a case, the reasons can be analyzed and optimization measures can be taken; if there are a large number of complaints of a certain type, management in that area can be strengthened in a targeted manner.

[0030] The above embodiments only illustrate preferred embodiments of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications, improvements, and substitutions without departing from the concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A government complaint intelligent service platform based on a large language model, characterized in that, include: The complaint receiving module is used to receive complaint text input by users, including complaint content submitted by users through text input, voice input, and image text recognition. The preprocessing module, connected to the complaint receiving module, is used to preprocess the received complaint text, including word segmentation, stop word removal, grammatical error correction, conversion of voice input into text content, and correction of image text recognition content. The large language model processing module is connected to the preprocessing module. It uses a pre-trained large language model to perform semantic understanding on the preprocessed complaint text. The large language model can handle multiple language forms, including dialects, identify dialect expressions in the complaint text and convert them into standard language for understanding. It also extracts key information from the complaint text, including the complaint object, complaint event, complaint time, location involved, and problem description. The importance assessment module is connected to the large language model processing module. It assesses the importance of the complaint information based on the semantic understanding results. The assessment criteria include preset rules such as the urgency of the complaint event, the number of people involved, the possible scope of impact and severity, or an assessment model obtained through training, and generates an importance level. The processing and allocation module is connected to the importance assessment module. It processes and allocates complaint information according to its importance level. For complaint information with a high importance level, it is given priority to be assigned to human customer service for processing or marked as a key matter. For complaint information with a low importance level, the intelligent customer service provides an initial response according to preset reply templates and rules. The feedback module, connected to the processing allocation module, is used to feed back the processing results to the user, receive user feedback, and record and store the processing process and results.

2. The intelligent government complaint service platform based on a large language model according to claim 1, characterized in that: The training data for the large language model includes historical government complaint texts, publicly available government documents, and common dialect corpora. It is trained using supervised learning, unsupervised learning, and reinforcement learning methods to improve the ability to understand government complaint texts and dialects.

3. The intelligent government complaint service platform based on a large language model according to claim 1, characterized in that: The preprocessing module also includes the function of performing sentiment analysis on complaint texts, identifying the emotional tendencies of users when making complaints, and providing the sentiment information to the large language model processing module and the importance assessment module as a reference.

4. The intelligent government complaint service platform based on a large language model according to claim 1, characterized in that: The preset rules in the importance assessment module include: matters involving life and property safety, major public interests, and emergency situations are set as high importance; matters involving general service attitude and minor complaints are set as low importance.

5. The intelligent government complaint service platform based on a large language model according to claim 1, characterized in that, The processing and allocation module is also connected to the business system of government departments, which can classify and push complaint information to the corresponding government departments for processing according to the complaint object and complaint content, and track the processing progress.

6. The intelligent government complaint service platform based on a large language model according to claim 1, characterized in that: The feedback module provides multiple feedback methods, including SMS notifications, in-platform message reminders, and emails, to facilitate users in obtaining processing results in a timely manner.

7. The intelligent government complaint service platform based on a large language model according to claim 1, characterized in that, It also includes a data statistical analysis module, which is connected to the feedback module and the processing and allocation module to perform statistical analysis on the handling of complaint information, generate statistical reports, and provide data support for government departments to optimize services. The statistical analysis includes the number of complaints, processing time, and user satisfaction indicators.

8. The intelligent government complaint service platform based on a large language model according to claim 7, characterized in that, The complaint receiving module provides a user interface where users can view the progress of complaint processing and submit supplementary information.

9. The intelligent government complaint service platform based on a large language model according to claim 1, characterized in that, The importance assessment module also integrates a historical complaint processing database, which can dynamically adjust the weight parameters of the preset rules or assessment model based on the processing results of historical complaints and user feedback data, so as to achieve adaptive assessment of the importance of complaint information.

10. A government complaint intelligent service platform based on a large language model according to claim 1, characterized in that, The processing and allocation module has a built-in priority queue scheduling algorithm that can dynamically adjust the allocation order of complaint information according to the importance level of the complaint information, the real-time processing load of government departments, and the preset processing time requirements, so as to ensure that the processing response time of high-importance complaint information does not exceed the preset threshold.

Images