Intelligent question and answer interaction method for district and county livelihood service

Abstract

The invention discloses an intelligent question and answer interaction method for district and county livelihood services, belongs to the technical field of multi-channel service deployment and user experience optimization, and solves the problems that the traditional hotline response efficiency is low, the offline service coverage is insufficient, and the dialect recognition and localization policy interpretation of an existing intelligent platform are weak. The method comprises three modules: 1, dialect adaptation and multi-channel deployment, training of a multi-accent voice model, and multi-terminal synchronization service; secondly, dynamic knowledge base and intelligent pre-auditing are carried out, 13,000 + data knowledge bases are built through a DeepSeek large model, and 60% of repeated auditing is reduced in a closed loop; and 3, scene customization and accurate portraying: constructing a user portraying and providing customization service according to three types of scenes. The consultation waiting time is compressed to be within 30 seconds, the service efficiency, the coverage rate and the accuracy are improved, and the method is suitable for district and county livelihood services.

Description

TECHNICAL FIELD

[0001] The technical scheme belongs to the technical field of multi-channel service deployment and user experience optimization, and particularly relates to an intelligent question and answer interaction method for county-level livelihood services. BACKGROUND

[0002] In the field of county-level livelihood services, improving service response efficiency and coverage is the key to achieving efficient governance. However, current county-level livelihood services generally face the problems of low service response efficiency and insufficient coverage. Although traditional hotline services have high connection rates, they are limited by the processing capacity of human agents and the time-consuming cross-departmental work order transfer, resulting in low service efficiency. At the same time, there is a digital divide in rural areas and among the elderly, making it difficult to recognize dialects and increasing the difficulty of intelligent service access, which limits service coverage. In addition, traditional platforms lack accuracy in local policy interpretation, which cannot meet the needs of specialized industries.

[0003] Existing solutions: Currently, county-level livelihood services mainly rely on traditional hotline services and offline service points, which have obvious limitations. For example, traditional hotline services are limited by the number of human agents and have limited processing capacity, making it difficult to handle a large number of concurrent livelihood demands. Offline service points have geographical limitations, and service coverage is low for rural areas and the elderly who have difficulty moving. In addition, existing intelligent service platforms have deficiencies in dialect recognition and local policy interpretation, resulting in low service access and accuracy.

[0004] Problems in the prior art: Although existing service methods provide livelihood services to some extent, there are still some problems and deficiencies. First, the processing capacity of human agents in traditional hotline services is limited, making it difficult to handle a large number of concurrent livelihood demands, resulting in low service response efficiency. Second, the geographical limitations of offline service points result in insufficient service coverage, especially for rural areas and the elderly. In addition, existing intelligent service platforms have deficiencies in dialect recognition and local policy interpretation, resulting in low service access and accuracy. Therefore, developing a new type of global interaction system to improve the response efficiency and coverage of county-level livelihood services has important practical significance and application value. SUMMARY

[0005] The present application aims to at least partially solve one of the above technical problems.

[0006] To achieve the above-mentioned purpose, the present application provides an intelligent question and answer interaction method for county-level livelihood services, comprising the following steps: S1: Constructing a dialect adaptation system and implementing multi-channel service deployment S11: Collecting characteristic dialect corpus in county-level administrative areas, and training a multi-accent voice interaction model based on the characteristic dialect corpus; S12: The voice interaction model is optimized for noise suppression using a microphone array technique to improve dialect recognition accuracy; S13: Intelligent question and answer services are deployed synchronously in WeChat mini programs, Alipay mini programs, and offline government hall terminals to realize the interactive function of "scanning code to ask - cross-end data synchronization" for users; S2: Build a dynamic knowledge base and establish a data governance mechanism for intelligent pre-examination S21: Take DeepSeek large model as the core, and build a dynamic knowledge base covering no less than 130,000 county-level livelihood service data, including policy documents, service guides, and characteristic industrial service information; S22: Realize real-time updating of the dynamic knowledge base through a three-level updating mechanism of "automatic grabbing - department verification - user feedback correction", and automatically grab policy documents into the database after publication; S23: For cross-department livelihood service demands, design a "intelligent pre-examination - manual review" closed loop process: the system automatically extracts the user's associated livelihood service data, and after preliminary verification, it is pushed to the corresponding administrative department for secondary review; S3: Build a scene customization module and realize precise user portrait service S31: According to the industrial characteristics of county-level administrative regions, divide three scene modules of government service, enterprise service, and travel service, and match exclusive intelligent question and answer interaction logic for each scene module; S32: Relying on the user behavior analysis system, collect user's livelihood service interaction data, and build a user's precise portrait including enterprise size, resident age, and historical demand type; S33: Based on the user's precise portrait, provide customized intelligent question and answer services for different groups: automatically adapt "voice speed + friend proxy permission opening" mode for elderly users, and push customized policy declaration list for small and micro enterprise users.

[0007] As an improvement, in step S12, the dialect recognition accuracy after optimization by the microphone array technique is ≥92%.

[0008] As an improvement, in step S22, the automatic grabbing of policy documents into the database takes ≤4 hours.

[0009] As an improvement, in step S11, the characteristic dialect corpus includes Wu dialect and Xiang dialect corpus, and the total duration of the corpus is ≥5000 hours.

[0010] As an improvement, in step S23, the "intelligent pre-examination - manual review" closed loop process reduces the workload of cross-department repeated audits by ≥60%.

[0011] As an improvement, in step S21, the accuracy rate of the dynamic knowledge base in matching the service guide is ≥98.6%.

[0012] As an improvement, in step S31, the functions of the scenario module include: agricultural machinery subsidy calculation and crop pest and disease prevention consultation functions for agricultural-led counties, and AR real-scene navigation and cultural tourism and food recommendation functions for tourism-led counties.

[0013] As an improvement, in step S33, the "voice speed adjustment" mode supports 0.7-1.5 times speech speed adjustment, and the "family and friends proxy permission" allows users to set 1-3 proxy persons and the scope of proxy services.

[0014] As an improvement, in step S33, the customized policy application list pushed to micro and small enterprises has a policy matching accuracy rate of ≥95%.

[0015] As an improvement, in step S13, the offline government service hall terminal supports both touch interaction and voice interaction operation modes, and the terminal response delay is ≤1.5 seconds.

[0016] Compared with existing technologies, the beneficial effects of this technical solution are as follows: Improve service response efficiency: By building a full-domain interactive system that combines dialect adaptation and multi-channel deployment, and by integrating dialect recognition technology with multi-channel service deployment, the service response speed has been significantly improved, reducing the average waiting time for the public to within 30 seconds. This effectively solves the problem of low service response efficiency caused by the limited processing capacity of traditional hotline human agents.

[0017] Expanding service coverage: By launching services simultaneously on WeChat, Alipay mini-programs, and offline government service hall terminals, the system has achieved "scan to ask questions and cross-platform synchronization," effectively covering rural areas and the elderly population, and solving the problem of insufficient service coverage caused by the geographical limitations of offline service points.

[0018] Improving the accuracy of dialect recognition and policy interpretation: By collecting county-specific dialect data to train a voice interaction model that supports multiple accents, the accuracy of dialect recognition has been improved to over 92%. At the same time, a dynamic knowledge base is built with the DeepSeek large model as the core to realize real-time data updates, which improves the accuracy of localized policy interpretation and effectively solves the problem of low service reach and accuracy caused by insufficient dialect recognition and localized policy interpretation in existing intelligent service platforms.

[0019] Achieving precise services: By dividing the county into three major scenario modules based on the county's industrial characteristics—government affairs, enterprise benefits, and culture and tourism—and relying on a user behavior analysis system to build precise profiles, customized services are provided for different groups. For example, the "voice speed adjustment + relatives and friends handling on behalf" mode is automatically adapted for the elderly, and customized policy application lists are pushed to micro and small enterprises, effectively improving the accuracy and satisfaction of services. Attached Figure Description

[0020] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a system framework diagram according to the present invention. Detailed Implementation

[0021] This implementation method is based on "a method for constructing a full-domain interactive system to improve the response efficiency and coverage of county-level public services". It focuses on three core modules: dialect adaptation and multi-channel deployment, dynamic knowledge base and intelligent pre-screening, and scenario customization and precise profiling. Combining the common needs of county-level public services, it elaborates on the implementation process of the technical solution and verifies the feasibility and effectiveness of the solution.

[0022] To address the unique dialect usage scenarios within district and county-level administrative regions, and focusing on frequently asked questions in public services (social security inquiries, subsidy applications, medical insurance reimbursements, and service guides), a dialect data collection project will be conducted over a period of 2-3 months. Corpus Source: Residents of different age groups (20-75 years old, with the elderly population over 60 years old accounting for 30%-40%) from various townships within the county were selected as corpus contributors. Daily public service consultation dialogues were collected through two methods: on-site recording at the government service hall and telephone interview recording. The corpus covers more than 200 typical consultation scenarios, including "subsidy application conditions", "list of application materials", "policy validity period" and "processing time limit". Corpus processing: The collected speech data is denoised, segmented and labeled, and invalid data with background noise exceeding 60 decibels is removed, ultimately forming an effective corpus of no less than 5,000 hours to ensure coverage of the main dialect types used in the county. Model Training: Drawing on the dialect recognition technology framework of the HarmonyOS Next system, the CNN-BiLSTM-CTC deep learning model is adopted. The collected dialect data is used as the training set and iteratively trained for 15-20 rounds. After each round of training, the recognition weights of high-frequency dialect words related to people's livelihood such as "subsidy", "enrollment in insurance" and "reimbursement" are optimized through confusion matrix. Ultimately, the dialect recognition accuracy is improved to over 92%, meeting the recognition needs of daily livelihood consultation scenarios.

[0023] For the noisy environment of district and county-level government service halls (daily traffic of 500+ people, background noise of 45-70 decibels) and the outdoor consultation scenarios of rural markets and township convenience service points (susceptible to noise interference from wind, conversations, etc.), the voice acquisition module of the intelligent question-and-answer terminal has been optimized: Hardware configuration: The offline terminals in government service halls and mobile service vehicles in rural areas are all integrated with 4-6 microphone arrays, with the microphone spacing set to 8-10cm. Beamforming technology is used to focus on the user's voice direction and suppress environmental noise from non-target directions. Algorithm optimization: Through an adaptive noise suppression algorithm, steady-state noise (such as the sound of air conditioning running in the lobby and the sound of heat dissipation of terminal equipment) and non-steady-state noise (such as the sound of people talking, vehicle horns, and market vendors) are identified and filtered in real time. In a noise intensity of 70 dB, the signal-to-noise ratio of the speech signal is improved to more than 25 dB, ensuring the stability and accuracy of dialect recognition.

[0024] Referring to Alibaba Cloud's multi-channel deployment solution, we can achieve full coverage of "online + offline" services within district and county-level administrative regions, adapting to the usage habits of different groups: Online channels: The intelligent Q&A service is launched simultaneously on WeChat Mini Program and Alipay Mini Program, supporting a dual mode of "text input + voice interaction". Users can quickly access the service page by scanning the QR codes posted in government service halls, community bulletin boards, and township service points. Cross-platform data is synchronized in real time (for example, consultations that are not completed on WeChat can be continued on Alipay, and historical dialogues and answers are fully retained). Offline channels: Deploy offline smart terminals in all government service halls, township service centers, and key rural communities within the county. The terminal screen size is set to 15.6-19 inches (adapting to the visual needs of the elderly), supporting touch operation (response speed ≤0.5 seconds) and voice operation (customizable wake-up words, such as "People's Livelihood Assistant"). The interface font is enlarged to 16-18 points by default. At the same time, a "senior mode" is provided—hiding complex function entrances and retaining only the four core modules of "social security consultation", "medical insurance inquiry", "service guide" and "subsidy application" to reduce the complexity of operation. Results Verification: After the module was launched, the average waiting time for public service consultations within the county was reduced from 10-15 minutes to within 30 seconds, the proportion of online consultations increased from 10%-15% to over 60%, and the offline terminal usage rate among the elderly population over 60 years old reached over 35%, effectively breaking down the digital divide, geographical limitations, and dialect barriers between rural areas and the elderly population.

[0025] Combining the industrial characteristics (such as agriculture-led, industry-led, culture and tourism-led, etc.) and public service needs of district and county-level administrative regions, a dynamic knowledge base covering no less than 130,000 data points is constructed with the DeepSeek big data model as the core. Data Classification and Sources: Knowledge base data is divided into three core categories: Policy documents (30,000-40,000 items): These are local policies issued by county government websites and various administrative departments, including subsidy details, support measures, and management regulations. Service guides (50,000-60,000 items): These are derived from the service manuals at the government service hall windows and departmental business specifications, covering the procedures, material lists, and processing time limits for matters such as social security enrollment, medical insurance reimbursement, household registration, business registration, and real estate registration. Specialized service information (30,000-50,000 items): These are derived from on-site surveys and reports from enterprises / institutions, including the addresses and contact information of agricultural machinery repair shops, opening hours and ticket information of cultural and tourist attractions, the scope of medical services provided by township health centers, and the distribution of rural e-commerce service stations. Data structuring: The knowledge base data is structured and labeled using the DeepSeek large model. Each data entry is tagged with keywords (e.g., 'rice subsidy', 'scenic spot reservation', 'social security payment'), applicable population / target (e.g., 'grain farmers', 'micro and small enterprises', 'elderly residents'), validity period (e.g., '2024.1.1-2024.12.31'), and department, enabling precise retrieval of user inquiries. The model's accuracy in recognizing user inquiry intent exceeds 95%.

[0026] A three-tiered workflow of "automatic data retrieval + departmental verification + user feedback correction" is adopted to ensure the timeliness and accuracy of the knowledge base content. Automated data crawling: A dedicated data crawler program was developed to connect with the county government's official website, various departmental business systems, and government information disclosure platform. It is set to automatically perform a check every 2 hours. When a new policy document is released, a service guide is updated, or special service information is changed, the core content such as the document title, body, attachments, and release time is automatically extracted and the data is entered into the database within 4 hours to avoid information lag. Departmental Verification: Establish a "Departmental Verification Responsibility List" to allocate knowledge base data to corresponding administrative departments according to their respective fields. Each department should designate 1-2 dedicated personnel to be responsible for verification. Within 24 hours of data entry, the accuracy of the content should be reviewed (such as checking whether the subsidy amount, application conditions, and material list are consistent with the actual policy). Once the verification is passed, it will be marked as "valid data". If it fails, specific modification suggestions will be provided, and the system will complete the adjustment and resubmit the data for verification within 12 hours. User feedback correction: A "Content Correction" button is set in a prominent position on the intelligent Q&A service interface. When users find that the knowledge base content is incorrect (such as "the list of application materials is missing the requirement for a copy of the ID card"), outdated (such as "the policy is still showing 'valid' even though the validity period has expired"), or incomplete (such as "key steps in the process are missing"), they can submit feedback and select the problem type (error, outdated, incomplete). The system will respond within 12 hours, and after the relevant department reviews and confirms the data, it will update the data and push an "update notification" to the user who submitted the feedback. Results Verification: After the mechanism was implemented, the delay in updating the knowledge base data within the county was shortened from 7 days under traditional manual maintenance to within 4 hours, the accuracy rate of matching service guides reached over 98.6%, and the rate of secondary inquiries by users due to "outdated / incorrect / incomplete information" decreased from 25%-30% to below 5%.

[0027] To address cross-departmental public service requests at the district and county levels (such as social security reimbursement requiring linking medical insurance and tax data, enterprise policy applications requiring linking market supervision and finance data, and household registration transfers requiring linking relevant department and community data), a closed-loop process of "intelligent pre-review - manual review" has been designed to reduce redundant review workload. Data Integration: Connect with the county-level government data sharing platform to obtain core public service data such as users' social security participation records, medical insurance payment records, enterprise registration information, tax registration information, and household registration information, ensuring that the system can automatically extract users' related information without requiring users to fill in the information repeatedly; Intelligent pre-screening: Taking "social security reimbursement consultation" as an example, when a user initiates "inpatient expense reimbursement ratio inquiry", the system automatically extracts data such as the user's medical insurance participation type (employee medical insurance / resident medical insurance), payment period, hospital level, and inpatient expense amount. Combined with local medical insurance reimbursement policies in the knowledge base, the system preliminarily determines the user's reimbursement ratio (e.g., if an employee has paid medical insurance for more than 2 years and is hospitalized in a secondary hospital, the reimbursement ratio is 85%), and generates an "Intelligent Pre-screening Report" containing "user information, extracted data, policy basis, and pre-screening results". Manual review: The system automatically pushes the "Intelligent Pre-review Report" to the relevant business department of the administrative department. Staff do not need to repeatedly query user data. They only need to check the consistency between the pre-review results and the policy provisions. After confirming that there are no errors, the results are fed back to the user to complete the consultation. If there are any questions about the pre-review results, staff can retrieve relevant materials through the system for further verification and feedback. Results Verification: After the implementation of this closed loop, the review time for cross-departmental inquiries within the county was shortened from 2-3 days for traditional manual inquiries to within 1 day. The workload of repeated cross-departmental reviews was reduced by more than 60%, and the average daily number of inquiries handled by the corresponding department staff increased from 30-40 to 80-100, significantly improving service efficiency.

[0028] Based on the industrial type (agriculture-led, industry-led, culture and tourism-led, etc.) and the focus of public services in the district / county-level administrative region, three major scenario modules—"government services, enterprise-benefiting services, and culture and tourism services"—have been defined, and dedicated functions have been developed accordingly. The government services module focuses on residents' frequent daily needs and has developed functions such as "Social Security Inquiry (payment records, account balance, qualification certification status)", "Medical Insurance Reimbursement Calculation (enter hospitalization expenses, hospital level, and medical insurance type to automatically calculate reimbursement amount)", "Household Registration Guide (divided into sub-scenarios such as 'newborn household registration', 'household registration transfer', and 'ID card replacement', clearly defining the process and materials)" and "Subsidy Application Inquiry (inquire about the application progress and disbursement status of various livelihood subsidies)". The interface design follows the principles of "simplification and step-by-step" with no more than 3 core operations. Business Support Module: Features are tailored to the specific needs of different types of businesses. For agriculture-led counties: Added features include "Agricultural Machinery Subsidy Calculation (enter agricultural machinery model, quantity, and purchase amount to automatically calculate subsidy amount)", "Pest and Disease Control Consultation (supports uploading crop disease photos, and the system matches local pest and disease control solutions and expert contact information)", and "Agricultural Product Sales Matching (recommends local e-commerce platforms and buyer information)". Industry-led counties: Added features include "Policy Application Reminder (pushing suitable industry support policies based on enterprise industry and size)", "Tax Consultation (answering common questions about enterprise value-added tax, income tax, etc.)" and "Financing Matching Portal (connecting with local banks and guarantee institutions)". For counties with tourism-led development: Add "Tourism Subsidy Application" and "Tourist Traffic Inquiry (provides real-time visitor traffic data for various scenic spots within the county to assist tourism enterprises in adjusting their operational strategies)"; Cultural and tourism service module: Integrates county-level cultural and tourism resources, and develops functions such as "AR real-view navigation (scanning scenic spot maps to generate AR navigation routes to guide tourists)", "cultural and tourism food recommendations (recommending characteristic restaurants within 3 kilometers based on the user's location, indicating average cost per person, business hours, and signature dishes)", "cultural activity registration (online registration and ticket booking for activities such as folk festivals, harvest festivals, and cultural exhibitions)" and "attraction reservation (time-slot reservations for popular attractions to avoid crowds)".

[0029] Leveraging a user behavior analysis system, we collect user interaction data and build multi-dimensional, precise user profiles to provide customized intelligent question-and-answer services for different groups. Data Collection and User Profile Building: Collect user consultation types (e.g., "social security", "subsidies", "culture and tourism"), historical request records (contact content, frequency, and satisfaction rating for the past 3 months), devices used (mobile / offline), regions (urban / township / rural), and identity attributes (age, occupation, company size, industry type, etc. obtained through data integration) to build two types of user profiles: Resident user profile: covering age (e.g., "elderly users aged 65 and above" and "young users aged 18-35"), frequently consulted areas (e.g., "social security-related inquiries account for 70%" and "cultural and tourism-related inquiries account for 60%)), and usage habits (e.g., "prefers voice interaction" and "prefers text interaction"). Enterprise user profile: covering enterprise size (micro and small enterprises / medium-sized enterprises / large enterprises), industry (agriculture / industry / service industry / culture and tourism), and type of policy needs (subsidies / financing / taxation). Customized service implementation: Senior Citizen Service: When the system recognizes a user's age as 60 or above, it automatically activates "Senior Mode," with the voice interaction speed adjusted to 0.8x by default (manual adjustment from 0.7x to 1.5x is supported to meet the needs of different hearing conditions). It also provides a "Friends / Relatives Proxy Service" function—users can enter the mobile phone numbers of 1-3 friends / relatives and set their proxy permissions (e.g., only "Social Security Inquiry" and "Application Progress Inquiry" are enabled, while "Subsidy Application" is not enabled). After logging in with a verification code, friends / relatives can assist in initiating inquiries and checking application progress, solving the operational difficulties faced by elderly users. Services for micro and small enterprises: For micro and small enterprise users, the system automatically pushes a customized policy application list based on their industry (such as micro and small enterprises in agricultural planting) and historical consultation records (such as consultation on "agricultural machinery subsidies"). The list clearly marks "application conditions, required materials, application time and online access", reducing the time cost for enterprises to search for policies on their own. Rural Resident Services: To address the frequent inquiries from rural residents regarding "agricultural subsidies," "agricultural machinery services," and "agricultural product sales," a dedicated entry point has been set up on the homepage of the intelligent Q&A service to simplify the search process. At the same time, a "voice consultation priority" mode is provided to adapt to the usage habits of users in rural areas. Results Verification: After the application of this module, the usage rate of smart services among elderly users in the county increased from 10%-15% to over 40%, and the satisfaction rate of policy application consultation for micro and small enterprises reached over 90%.

[0030] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0031] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0032] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A smart question-and-answer interaction method for county-level public services, characterized in that, Includes the following steps: S1: Construct a dialect adaptation system and implement multi-channel service deployment. S11: Collect characteristic dialect data within the district / county-level administrative region, and train a voice interaction model that supports multiple accents based on the characteristic dialect data; S12: Microphone array technology is used to optimize the noise suppression of the voice interaction model in order to improve the accuracy of dialect recognition; S13: Deploy intelligent Q&A services simultaneously on WeChat mini-programs, Alipay mini-programs, and offline government service hall terminals to enable users to "scan the code to ask questions - cross-platform data synchronization" interactive functions; S2: Construct a dynamic knowledge base and establish an intelligent pre-screening data governance mechanism. S21: Using the DeepSeek big model as the core, construct a dynamic knowledge base covering no less than 130,000 pieces of county-level public service data, including policy documents, service guides and information on characteristic industries; S22: The dynamic knowledge base is updated in real time through a three-level update mechanism of "automatic capture - departmental verification - user feedback correction", wherein policy documents are automatically captured and stored in the database after they are published; S23: To address cross-departmental public service requests, a closed-loop process of "intelligent pre-review - manual review" is designed: the system automatically extracts the public service data associated with the user, and after preliminary verification, it is pushed to the corresponding administrative department for secondary review; S3: Build a scenario customization module and implement precise user profiling services S31: Based on the industrial characteristics of district and county-level administrative regions, the system is divided into three major scenario modules: government services, business support services, and cultural and tourism services. Each scenario module is matched with a unique intelligent question-and-answer interaction logic. S32: Relying on the user behavior analysis system, collect users' interaction data on public services and build accurate user profiles that include enterprise size, resident age, and types of historical demands; S33: Based on the precise user profile, provide customized intelligent Q&A services for different groups: automatically adapt the "voice speed adjustment + permission to handle matters on behalf of relatives and friends" mode for elderly users, and push customized policy application lists to micro and small enterprise users.

2. The intelligent question-and-answer interaction method for county-level public services according to claim 1, characterized in that, In step S12, the dialect recognition accuracy after optimization by microphone array technology is ≥92%.

3. The intelligent question-and-answer interaction method for county-level public services according to claim 1, characterized in that, In step S22, the automatic capture and storage time of policy documents is ≤4 hours.

4. The intelligent question-and-answer interaction method for county-level public services according to claim 1, characterized in that, In step S11, the characteristic dialect corpus includes Wu dialect and Xiang dialect corpus, and the total duration of the corpus is ≥5000 hours.

5. The intelligent question-and-answer interaction method for county-level public services according to claim 1, characterized in that, In step S23, the closed-loop process of "intelligent pre-review - manual review" reduces the workload of cross-departmental repetitive review by ≥60%.

6. The intelligent question-and-answer interaction method for county-level public services according to claim 1, characterized in that, In step S21, the dynamic knowledge base achieves a matching accuracy of ≥98.6% for the service guide.

7. The intelligent question-and-answer interaction method for county-level public services according to claim 1, characterized in that, In step S31, the functions of the scenario module include: agricultural machinery subsidy calculation and crop pest and disease prevention consultation for agricultural-led counties, and AR real-scene navigation and cultural tourism and food recommendation for tourism-led counties.

8. The intelligent question-and-answer interaction method for county-level public services according to claim 1, characterized in that, In step S33, the "voice speed adjustment" mode supports adjusting the speech speed from 0.7 to 1.5 times, and the "family and friends proxy permission" allows users to set 1 to 3 proxy persons and the scope of proxy services.

9. The intelligent question-and-answer interaction method for county-level public services according to claim 1, characterized in that, In step S33, a customized policy application list is pushed to micro and small enterprises, with a policy matching accuracy rate of ≥95%.

10. The intelligent question-and-answer interaction method for county-level public services according to claim 1, characterized in that, In step S13, the offline government service hall terminal supports two operation modes: touch interaction and voice interaction, and the terminal response delay is ≤1.5 seconds.

Images