Government affair hotline complaint classification and automatic order sending system and method based on large model

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a government affair hotline complaint classification and automatic order sending system and method based on a large model, and adopts a four-layer architecture to realize full-process intelligent management of government affair complaints from acceptance to disposal. The four-layer architecture comprises a multi-mode sensing layer, an intelligent decision-making layer, a dynamic single-layer dispatching layer and a closed-loop self-optimization layer; the method has the beneficial effects that an efficient, accurate and intelligent government affair hotline complaint classification and automatic order dispatching system is constructed by fusing a large model technology and an innovative data processing and order dispatching strategy, rapid and accurate classification and automatic order dispatching of complaint work orders are realized, the work order processing flow is remarkably optimized, the processing period is shortened, and the working efficiency is improved. The quality and efficiency of government affair services are improved, and the satisfaction degree of the public to the government affair services is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and specifically to a system and method for classifying and automatically dispatching government hotline complaints based on a large model. Background Art

[0002] The development of artificial intelligence technology, especially the powerful performance of large models in natural language processing, has provided a new approach to solving the problem of handling government hotline complaint tickets. In the government service system, government hotlines serve as a key channel for the government to listen to public demands and solve livelihood issues, and they receive a huge number of complaint tickets every day. Although some government hotlines have introduced simple text classification algorithms and rule engines to assist in dispatching complaints, these methods have difficulty accurately understanding the semantics of increasingly complex and diverse complaint content, and the accuracy and adaptability of classification and dispatching are insufficient:

[0003] 1) Knowledge update lag: Existing systems mostly use static knowledge bases, which make it difficult to respond to policy adjustments in a timely manner. They rely on manual intervention and have update delays.

[0004] 2) Single dispatch strategy: Existing dispatch strategies are mainly based on fixed rules and lack comprehensive consideration of multiple dimensions such as the department's real-time processing capabilities, geographical location, and historical performance.

[0005] 3) Lack of feedback mechanism: Most existing systems lack an effective closed-loop optimization mechanism and are unable to learn and improve from processing results. Summary of the Invention

[0006] The purpose of the present invention is to provide a government hotline complaint classification and automatic dispatching system and method based on a large model to solve the problems raised in the above background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solutions: a large-scale model-based government hotline complaint classification and automatic dispatching system, which uses a four-layer architecture to achieve intelligent management of the entire process of government complaints, from acceptance to disposal. The four-layer architecture includes a multimodal perception layer, an intelligent decision-making layer, a dynamic dispatching layer, and a closed-loop self-optimization layer.

[0008] Multimodal perception layer: This layer supports complaint input from multiple channels, including voice, text, and images. It uses multimodal data fusion technology for integrated processing, including voice processing, image processing, and text parsing modules. It also builds a database of sensitive government terms for data cleaning and employs a generative adversarial network to enhance model robustness.

[0009] Intelligent decision-making layer: Deeply integrates three technical modules: large-scale semantic understanding, dynamic knowledge graph, and multi-objective dispatch optimization. It adopts a multi-granularity semantic parsing framework to build an intelligent analysis and precise decision-making system for government complaints, and has an attention mechanism for policy timeliness.

[0010] Dynamic dispatching layer: Based on intelligent decision-making, it uses a multi-objective optimization algorithm to achieve accurate dispatch of work orders. A five-dimensional evaluation system is established, comprehensively considering factors such as functional matching, real-time department load, geographic proximity, and historical processing timeliness to generate the optimal dispatch plan. It also supports manual review and rapid response channels.

[0011] Closed-loop self-optimization layer: Design a complete closed-loop self-optimization mechanism, establish a dual-channel feedback system, including explicit feedback and implicit feedback, continuously collect and process feedback data, use curriculum learning strategies and adversarial sample generation technology to adjust model parameters, and verify the effect through A / B testing.

[0012] Preferably, the speech processing module in the multimodal perception layer achieves high-precision speech-to-text conversion based on an end-to-end model, has the ability to suppress environmental noise and recognize dialects, and automatically identifies user emotions and determines the urgency level of complaints through voiceprint feature analysis and sentiment computing models;

[0013] Image processing module: Integrates object detection and OCR technology to automatically identify illegal elements and extract structured text information from images;

[0014] Text parsing module: Relying on a large language model base, it achieves context-aware semantic understanding and accurately extracts key entities in complaint content.

[0015] The dynamic government knowledge graph network in the intelligent decision-making layer preferably has three core advantages: real-time synchronization capability, intelligent weight adjustment, and full-dimensional knowledge coverage;

[0016] Real-time synchronization capability: Establish minute-level data channels with government data platforms, with an accuracy rate of up to 95% for detecting policy changes, enabling rapid synchronization of new policies within 24 hours;

[0017] Intelligent weight adjustment: Using the time-space decay algorithm, the influence coefficient is automatically adjusted according to the policy effectiveness;

[0018] Full-dimensional knowledge coverage: Integrate functional data from more than 200 government departments to build a complete government knowledge system.

[0019] Preferably, a scientific five-dimensional evaluation system is established for the dynamic dispatching layer, which comprehensively considers multiple factors such as functional matching, real-time department load, geographical proximity, and historical processing timeliness. The comprehensive score is obtained according to the calculation formula score = 0.4C + 0.3 (1-L) + 0.2S + 0.1G, where C is functional matching, L is department load rate, S is geographical proximity, and G is historical timeliness.

[0020] The system updates the work order backlog of each department every 5 minutes, calculates precise distances based on GIS road network data, and evaluates specialists' capabilities to generate the optimal dispatch plan.

[0021] It supports manual review mechanism, activates quick response channel for high-urgency complaints, and provides a complete dispatch basis chain and real-time visual monitoring interface.

[0022] Preferably, the closed-loop self-optimization layer has a well-designed dual-channel feedback system, including explicit feedback and implicit feedback mechanisms;

[0023] Display feedback: Automatically send a satisfaction questionnaire after a work order is completed, initiate a follow-up mechanism for low-scoring work orders, and open a text evaluation portal;

[0024] Implicit feedback: Continuously track processing time, secondary complaint rate, and work order reassignment rate indicators to form a comprehensive performance evaluation database;

[0025] The system automatically screens high-quality new data samples every week, uses a curriculum learning strategy to adjust model parameters in stages, and uses adversarial sample generation technology to enhance model adaptability;

[0026] Establish a strict effect verification system, compare the performance differences between the new and old models through A / B testing, and conduct significance tests.

[0027] A method for classifying and automatically dispatching complaints to a government hotline based on a large model, comprising the following steps:

[0028] Multimodal data fusion and cleaning: Multimodal data fusion technology is used to integrate and process multiple complaint data including voice, text, and images to ensure data integrity and accuracy. Speech processing uses an end-to-end model to achieve high-precision speech-to-text conversion, with environmental noise suppression and dialect recognition capabilities, and performs grammatical and semantic error correction through a post-processing module. Image processing integrates target detection and OCR technology to automatically identify illegal elements and extract structured text information from images. Text parsing relies on a large language model foundation to achieve context-aware semantic understanding and accurately extract key entities from complaint content. At the same time, a sensitive vocabulary library in the government sector is constructed for data cleaning, and a generative adversarial network is used to enhance model robustness.

[0029] Intelligent Decision-Making: Based on three technical modules: large-scale semantic understanding, dynamic knowledge graphs, and multi-objective dispatch optimization, this system builds an intelligent analysis and precise decision-making system for government complaints. It uses a multi-granularity semantic parsing framework to accurately identify key entities, analyze deep semantic relationships, and intelligently analyze the core demands of complainants. It also features a unique policy timeliness attention mechanism that dynamically adjusts reference weights based on policy release time, ensuring that decision-making is synchronized with the latest policies.

[0030] Dynamic dispatching: Based on intelligent decision-making, a multi-objective optimization algorithm is used to accurately dispatch work orders. A scientific five-dimensional evaluation system is established, comprehensively considering multiple factors such as functional matching, real-time department load, geographical proximity, and historical processing time. A comprehensive score is obtained based on a preset formula to generate the optimal dispatch plan. A manual review mechanism and rapid response channels are supported to ensure processing timeliness. A complete dispatch basis chain and a real-time visual monitoring interface are provided.

[0031] Closed-loop self-optimization: A comprehensive closed-loop self-optimization mechanism is designed, establishing a dual-channel feedback system, including explicit and implicit feedback. Explicit feedback automatically pushes a satisfaction questionnaire after a work order is completed, initiating a follow-up mechanism for low-scoring work orders. Implicit feedback continuously tracks processing timeliness and secondary complaint rates, forming a comprehensive performance evaluation database. High-quality new data samples are automatically screened weekly, and a curriculum learning strategy is used to adjust model parameters in stages. Adversarial sample generation technology is used to enhance model adaptability.

[0032] Effect verification and continuous optimization: Establish a strict effect verification system, compare the performance differences between the new and old models through A / B testing, and conduct significance tests; continuously optimize model parameters and dispatch strategies based on verification results to achieve continuous improvement in processing efficiency.

[0033] Preferably, in the multimodal data fusion and cleaning steps, the voice processing module also automatically identifies user emotions and determines the urgency level of the complaint through voiceprint feature analysis and emotion calculation models.

[0034] Preferably, in the intelligent decision-making step, the dynamic government knowledge graph network has three core advantages: real-time synchronization capability, intelligent weight adjustment and full-dimensional knowledge coverage; among them, the real-time synchronization capability establishes a minute-level data channel with the government data platform, and the accuracy of policy change detection is as high as 95%, realizing rapid synchronization of new policies within 24 hours; intelligent weight adjustment adopts a time-space attenuation algorithm to automatically adjust the influence coefficient according to the policy timeliness; full-dimensional knowledge coverage integrates the functional data of more than 200 government departments to build a complete government knowledge system.

[0035] Preferably, in the dynamic dispatching step, the calculation formula of the five-dimensional evaluation system is score = 0.4C + 0.3 (1-L) + 0.2S + 0.1G, where score is the dispatching score, C is the functional matching degree, L is the department load rate, S is the geographical proximity, and G is the historical timeliness; the system updates the backlog of work orders in each department every 5 minutes, calculates the precise distance based on GIS road network data, and conducts evaluation based on the specialist's ability portrait.

[0036] Preferably, in the closed-loop self-optimization step, adversarial sample generation technology enhances the robustness of the model by simulating various edge cases, ensuring that the system can quickly adapt to changes in the policy environment and achieve a spiral increase in processing efficiency during long-term service.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] The big model-based government hotline complaint classification and automatic dispatching system and method proposed in the present invention, by integrating big model technology with innovative data processing and dispatching strategies, constructs an efficient, accurate and intelligent government hotline complaint classification and automatic dispatching system, realizes rapid and accurate classification and automatic dispatching of complaint work orders, significantly optimizes the work order processing flow, shortens the processing cycle, improves the quality and efficiency of government services, and enhances the public's satisfaction with government services. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0040] In order to clearly and completely describe the objectives and technical solutions of the present invention and make the advantages more clearly understood, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the present invention, not all of them, and are only used to explain the embodiments of the present invention, not to limit the embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0041] For example 1, please refer to Figure 1 The present invention provides a technical solution: a large-scale model-based government hotline complaint classification and automatic dispatching system, which uses a four-layer architecture to achieve intelligent management of the entire process of government complaints from acceptance to disposal. The four-layer architecture includes a multimodal perception layer, an intelligent decision-making layer, a dynamic dispatching layer, and a closed-loop self-optimization layer.

[0042] Multimodal perception layer: supports multi-channel complaint input such as voice, text, and images, and uses multimodal data fusion technology for integrated processing, including voice processing, image processing, and text parsing modules. It also builds a sensitive vocabulary library in the government field for data cleaning, and uses adversarial generative networks to enhance the robustness of the model. The voice processing module in the multimodal perception layer achieves high-precision voice-to-text conversion based on an end-to-end model, has the ability to suppress environmental noise and recognize dialects, and automatically identifies user emotions and determines the urgency level of complaints through voiceprint feature analysis and sentiment computing models. Image processing module: integrates target detection and OCR technology to automatically identify violation elements and extract structured text information from images. Text parsing module: relying on a large language model base, it achieves context-aware semantic understanding and accurately extracts key entities in the complaint content.

[0043] Intelligent decision-making layer: deeply integrates the three technical modules of large-model semantic understanding, dynamic knowledge graph and multi-target dispatch optimization, adopts a multi-granularity semantic parsing framework, builds an intelligent analysis and precise decision-making system for government complaints, and has a policy timeliness attention mechanism; the dynamic government knowledge graph network in the intelligent decision-making layer has three core advantages: real-time synchronization capability, intelligent weight adjustment and full-dimensional knowledge coverage; real-time synchronization capability: establishes a minute-level data channel with the government data platform, and the policy change detection accuracy rate is as high as 95%, realizing rapid synchronization of new policies within 24 hours; intelligent weight adjustment: adopts the time-space attenuation algorithm to automatically adjust the influence coefficient according to the policy timeliness; full-dimensional knowledge coverage: integrates the functional data of more than 200 government departments to build a complete government knowledge system.

[0044] Dynamic dispatch layer: Based on intelligent decision-making, it uses a multi-objective optimization algorithm to achieve accurate dispatch of work orders, establishes a five-dimensional evaluation system, and comprehensively considers factors such as functional matching, department real-time load, geographical proximity, and historical processing time to generate the optimal dispatch plan, and supports manual review and rapid response channels. The dynamic dispatch layer establishes a scientific five-dimensional evaluation system, comprehensively considering multiple factors such as functional matching, department real-time load, geographical proximity, and historical processing time. The comprehensive score is obtained according to the calculation formula score = 0.4C + 0.3 (1-L) + 0.2S + 0.1G, where C is functional matching, L is department load rate, S is geographical proximity, and G is historical time. The system updates the backlog of work orders in each department every 5 minutes, calculates the precise distance based on GIS road network data, and evaluates based on the specialist's capability profile to generate the optimal dispatch plan. It supports a manual review mechanism, initiates a rapid response channel for high-urgency complaints, and provides a complete dispatch basis chain and a real-time visual monitoring interface.

[0045] Closed-loop self-optimization layer: A comprehensive closed-loop self-optimization mechanism is designed, and a dual-channel feedback system is established, including explicit feedback and implicit feedback. Feedback data is continuously collected and processed, and model parameters are adjusted using curriculum learning strategies and adversarial sample generation technology. The results are verified through A / B testing. The closed-loop self-optimization layer is designed with a comprehensive dual-channel feedback system, including explicit feedback and implicit feedback mechanisms. Explicit feedback: A satisfaction questionnaire is automatically pushed after a work order is completed, a follow-up mechanism is initiated for low-scoring work orders, and a text evaluation portal is opened. Implicit feedback: Processing time, secondary complaint rate, and work order duplicate rate are continuously tracked to form a comprehensive performance evaluation database. The system automatically screens high-quality new data samples every week, uses curriculum learning strategies to adjust model parameters in stages, and uses adversarial sample generation technology to enhance model adaptability. A rigorous effect verification system is established, and the performance differences between the new and old models are compared through A / B testing, and significance tests are performed.

[0046] Example 2, based on Example 1, proposes a method for classifying government hotline complaints and automatically dispatching orders based on a large model, including the following steps:

[0047] Step 1: Multimodal data fusion and cleaning

[0048] The system adopts multimodal data fusion technology and supports the integrated processing of multiple complaint data such as voice, text, and images. In terms of voice processing, it realizes high-precision speech-to-text conversion based on the end-to-end model, has the ability to suppress environmental noise and recognize dialects, and performs grammatical correction and semantic error correction through the post-processing module. At the same time, the system automatically identifies user emotions and determines the urgency level of complaints through voiceprint feature analysis and sentiment calculation models. The image processing module integrates target detection and OCR technology, which can automatically identify illegal elements (such as illegal construction signs, illegal advertisements, etc.) and extract structured text information from images. The text parsing module relies on the large language model base to achieve context-aware semantic understanding and accurately extract key entities in the complaint content (such as time, place, events, etc.). To ensure data quality, the system constructs a sensitive vocabulary in the government field for data cleaning, and uses adversarial generative networks to enhance the robustness of the model.

[0049] Step 2: Intelligent decision-making:

[0050] The intelligent decision-making layer is the core of this system. It deeply integrates the three major technical modules of large-scale model semantic understanding, dynamic knowledge graph, and multi-objective dispatch optimization to build a complete set of intelligent analysis and precise decision-making systems for government affairs complaints. This layer adopts an innovative multi-granularity semantic parsing framework: it realizes accurate identification of key entities at the basic level; at the semantic understanding level, it uses a multi-head attention mechanism to parse deep semantic relationships, effectively solving the problem of semantic ambiguity in government affairs scenarios; at the comprehensive analysis level, it uses context-based modeling to intelligently judge the core demands of the complainant. It is particularly worth mentioning that the system's original policy timeliness attention mechanism can dynamically adjust the reference weight according to the policy release time, ensuring that the decision-making basis is always synchronized with the latest policy.

[0051] The core of intelligent decision-making is the dynamic government knowledge graph network, which has three core advantages:

[0052] Real-time synchronization capability: Establish minute-level data channels with government data platforms, with an accuracy rate of up to 95% for detecting policy changes, enabling rapid synchronization of new policies within 24 hours;

[0053] Intelligent weight adjustment: The innovatively designed spatiotemporal attenuation algorithm can automatically adjust the influence coefficient according to the policy timeliness;

[0054] Full-dimensional knowledge coverage: Integrates functional data from more than 200 government departments, including 5,000+ responsibility relationships, 3,000+ policy and regulatory nodes, and 100,000+ historical cases, to build a complete government knowledge system.

[0055] At the multi-objective optimization level, based on this intelligent platform, the system adopts a three-level progressive processing mechanism:

[0056] -Simple matters (Level 1): Achieve millisecond-level automated response

[0057] - Complex matters (Level L2): Intelligent generation of cross-departmental collaborative solutions

[0058] -Difficult issues (Level L3): Deep similarity matching and recommendation based on case library

[0059] Target optimization: Simultaneously optimize processing timeliness (response speed), department load balancing (avoiding overload of individual departments), and public satisfaction (based on historical complaint closure rates), combined with policy compliance and responsibility boundaries, and finally obtain the classification results of complaint types, related departments, policy basis, and recommended processing solutions.

[0060] Step 3: Dynamic dispatch:

[0061] The dynamic dispatch layer, based on intelligent decision-making, utilizes a multi-objective optimization algorithm to precisely dispatch work orders. This layer establishes a scientific five-dimensional evaluation system that comprehensively considers multiple factors, including functional alignment, real-time departmental load, geographic proximity, and historical processing time, to generate a comprehensive score using the following formula.

[0062] score=0.4*C+0.3*(1-L)+0.2*S+0.1*G

[0063] Among them: score is the dispatch score; C is the functional matching degree, which is mainly calculated based on the dynamic knowledge graph; L is the department load rate, which monitors the number of pending work orders of each agency in real time; S is the geographical proximity, which calculates the optimal path based on the geographic information system (GIS) data; G is the historical timeliness, which is the value obtained by analyzing the historical processing efficiency of the department.

[0064] The system updates the backlog of work orders in each department every 5 minutes, calculates the precise distance based on GIS road network data, and conducts an assessment based on 200+ dimensions of specialist capability portraits to ultimately generate the optimal dispatch plan. For complex matters, the system supports a manual review mechanism; for high-urgency complaints, a rapid response channel is activated to ensure timely processing. The system provides a complete dispatch basis chain and a real-time visual monitoring interface to ensure that the decision-making process is transparent and explainable. More importantly, by continuously collecting and processing feedback data, the system automatically optimizes the dispatch strategy every week to achieve continuous improvement in processing efficiency, providing strong technical support for the digital transformation of government services.

[0065] Step 4: Closed-loop self-optimization:

[0066] The system has been designed with a comprehensive closed-loop self-optimization mechanism. It has established a dual-channel feedback system, which automatically pushes a satisfaction questionnaire with a rating after a work order is completed. A follow-up mechanism is initiated for low-scoring work orders, while a text-based review portal is available to collect specific suggestions. An implicit feedback mechanism continuously tracks metrics such as processing time, secondary complaint rate, and work order reassignment rate. It also records the efficiency of collaboration across departments and the spatial and temporal distribution of complaint hotspots, forming a comprehensive performance evaluation database.

[0067] The system automatically screens approximately 5,000 high-quality new data samples each week and uses a curriculum learning strategy to adjust model parameters in stages. The system also employs adversarial sample generation technology to enhance model adaptability and strengthen model robustness by simulating various edge cases.

[0068] To ensure the quality of each optimization iteration, the system has established a rigorous effectiveness verification system. A / B testing compares the performance differences between the new and old models and conducts significance tests. This continuous self-optimization design not only enables the system to quickly adapt to changes in the policy environment but also achieves a spiral increase in processing efficiency over the long term, providing a solid technical foundation for improving the quality of government services.

[0069] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A government hotline complaint classification and automatic dispatching system based on a large model, characterized by: A four-layer architecture is used to achieve intelligent management of the entire process of government complaints, from acceptance to disposal. The four-layer architecture includes a multimodal perception layer, an intelligent decision-making layer, a dynamic dispatching layer, and a closed-loop self-optimization layer. Multimodal perception layer: This layer supports complaint input from multiple channels, including voice, text, and images. It uses multimodal data fusion technology for integrated processing, including voice processing, image processing, and text parsing modules. It also builds a database of sensitive government terms for data cleaning and employs a generative adversarial network to enhance model robustness. Intelligent decision-making layer: Deeply integrates three technical modules: large-scale semantic understanding, dynamic knowledge graph, and multi-objective dispatch optimization. It adopts a multi-granularity semantic parsing framework to build an intelligent analysis and precise decision-making system for government complaints, and has an attention mechanism for policy timeliness. Dynamic dispatching layer: Based on intelligent decision-making, it uses a multi-objective optimization algorithm to achieve accurate dispatch of work orders. A five-dimensional evaluation system is established, comprehensively considering factors such as functional matching, real-time department load, geographic proximity, and historical processing timeliness to generate the optimal dispatch plan. It also supports manual review and rapid response channels. Closed-loop self-optimization layer: Design a complete closed-loop self-optimization mechanism, establish a dual-channel feedback system, including explicit feedback and implicit feedback, continuously collect and process feedback data, use curriculum learning strategies and adversarial sample generation technology to adjust model parameters, and verify the effect through A / B testing.

2. A large-scale model-based government hotline complaint classification and automatic dispatching system according to claim 1, characterized in that: The speech processing module in the multimodal perception layer achieves high-precision speech-to-text conversion based on an end-to-end model. It has the ability to suppress environmental noise and recognize dialects. It also automatically identifies user emotions and determines the urgency level of complaints through voiceprint feature analysis and sentiment computing models. Image processing module: Integrates object detection and OCR technology to automatically identify illegal elements and extract structured text information from images; Text parsing module: Relying on a large language model base, it achieves context-aware semantic understanding and accurately extracts key entities in complaint content.

3. The system for classifying and automatically dispatching government hotline complaints based on a large model according to claim 2 is characterized by: The dynamic government knowledge graph network in the intelligent decision-making layer has three core advantages: real-time synchronization capability, intelligent weight adjustment, and full-dimensional knowledge coverage; Real-time synchronization capability: Establish minute-level data channels with government data platforms, with an accuracy rate of up to 95% for detecting policy changes, enabling rapid synchronization of new policies within 24 hours; Intelligent weight adjustment: Using the time-space decay algorithm, the influence coefficient is automatically adjusted according to the policy effectiveness; Full-dimensional knowledge coverage: Integrate functional data from more than 200 government departments to build a complete government knowledge system.

4. The system for classifying and automatically dispatching government hotline complaints based on a large model according to claim 1 is characterized by: Dynamic dispatch establishes a scientific five-dimensional evaluation system, which comprehensively considers multiple factors such as functional matching, real-time department load, geographical proximity, and historical processing timeliness. The comprehensive score is calculated according to the formula score = 0.4C + 0.3 (1-L) + 0.2S + 0.1G, where C is functional matching, L is department load rate, S is geographical proximity, and G is historical timeliness. The system updates the work order backlog of each department every 5 minutes, calculates precise distances based on GIS road network data, and evaluates specialists' capabilities to generate the optimal dispatch plan. It supports manual review mechanism, activates quick response channel for high-urgency complaints, and provides a complete dispatch basis chain and real-time visual monitoring interface.

5. The system for classifying and automatically dispatching government hotline complaints based on a large model according to claim 1 is characterized by: The closed-loop self-optimization layer has a well-designed dual-channel feedback system, including explicit feedback and implicit feedback mechanisms; Display feedback: Automatically send a satisfaction questionnaire after a work order is completed, initiate a follow-up mechanism for low-scoring work orders, and open a text evaluation portal; Implicit feedback: Continuously track processing time, secondary complaint rate, and work order reassignment rate indicators to form a comprehensive performance evaluation database; The system automatically screens high-quality new data samples every week, uses a curriculum learning strategy to adjust model parameters in stages, and uses adversarial sample generation technology to enhance model adaptability; Establish a strict effect verification system, compare the performance differences between the new and old models through A / B testing, and conduct significance tests.

6. A method for a large-scale model-based government hotline complaint classification and automatic dispatching system according to claim 5, characterized in that: The following steps are involved: Multimodal data fusion and cleaning: Multimodal data fusion technology is used to integrate and process multiple complaint data including voice, text, and images to ensure data integrity and accuracy. Speech processing uses an end-to-end model to achieve high-precision speech-to-text conversion, with environmental noise suppression and dialect recognition capabilities, and performs grammatical and semantic error correction through a post-processing module. Image processing integrates target detection and OCR technology to automatically identify illegal elements and extract structured text information from images. Text parsing relies on a large language model foundation to achieve context-aware semantic understanding and accurately extract key entities from complaint content. At the same time, a sensitive vocabulary library in the government sector is constructed for data cleaning, and a generative adversarial network is used to enhance model robustness. Intelligent Decision-Making: Based on three technical modules: large-scale semantic understanding, dynamic knowledge graphs, and multi-objective dispatch optimization, this system builds an intelligent analysis and precise decision-making system for government complaints. It uses a multi-granularity semantic parsing framework to accurately identify key entities, analyze deep semantic relationships, and intelligently analyze the core demands of complainants. It also features a unique policy timeliness attention mechanism that dynamically adjusts reference weights based on policy release time, ensuring that decision-making is synchronized with the latest policies. Dynamic dispatching: Based on intelligent decision-making, a multi-objective optimization algorithm is used to accurately dispatch work orders. A scientific five-dimensional evaluation system is established, comprehensively considering multiple factors such as functional matching, real-time department load, geographical proximity, and historical processing time. A comprehensive score is obtained based on a preset formula to generate the optimal dispatch plan. A manual review mechanism and rapid response channels are supported to ensure processing timeliness. A complete dispatch basis chain and a real-time visual monitoring interface are provided. Closed-loop self-optimization: A comprehensive closed-loop self-optimization mechanism is designed, establishing a dual-channel feedback system, including explicit and implicit feedback. Explicit feedback automatically pushes a satisfaction questionnaire after a work order is completed, initiating a follow-up mechanism for low-scoring work orders. Implicit feedback continuously tracks processing timeliness and secondary complaint rates, forming a comprehensive performance evaluation database. High-quality new data samples are automatically screened weekly, and a curriculum learning strategy is used to adjust model parameters in stages. Adversarial sample generation technology is used to enhance model adaptability. Effect verification and continuous optimization: Establish a strict effect verification system, compare the performance differences between the new and old models through A / B testing, and conduct significance tests; continuously optimize model parameters and dispatch strategies based on verification results to achieve continuous improvement in processing efficiency.

7. A method according to claim 6, characterized in that: During the multimodal data fusion and cleaning steps, the voice processing module also automatically identifies user emotions and determines the urgency level of complaints through voiceprint feature analysis and emotion calculation models.

8. A method according to claim 6, characterized in that: In the intelligent decision-making step, the dynamic government knowledge graph network has three core advantages: real-time synchronization capability, intelligent weight adjustment and full-dimensional knowledge coverage. Among them, the real-time synchronization capability establishes a minute-level data channel with the government data platform, and the policy change detection accuracy rate is as high as 95%, realizing rapid synchronization of new policies within 24 hours; intelligent weight adjustment adopts a time-space attenuation algorithm to automatically adjust the influence coefficient according to the policy timeliness; full-dimensional knowledge coverage integrates the functional data of more than 200 government departments to build a complete government knowledge system.

9. A method according to claim 6, characterized in that: In the dynamic dispatching step, the calculation formula of the five-dimensional evaluation system is score = 0.4C + 0.3 (1-L) + 0.2S + 0.1G, where score is the dispatching score, C is the functional matching degree, L is the department load rate, S is the geographical proximity, and G is the historical timeliness; the system updates the backlog of work orders in each department every 5 minutes, calculates the precise distance based on GIS road network data, and conducts evaluation based on the specialist's ability profile.

10. A method according to claim 6, characterized in that: In the closed-loop self-optimization step, adversarial sample generation technology enhances the robustness of the model by simulating various edge cases, ensuring that the system can quickly adapt to changes in the policy environment and achieve a spiral increase in processing efficiency during long-term service.

Images