An intelligent government customer service application system

Abstract

The present application belongs to the field of data processing, and in particular, relates to an intelligent government affairs customer service application system, in which a task establishment module responds to a government affairs question request initiated by a user, and establishes a pending customer service task that matches the government affairs question request; a demand identification module calls an intelligent government affairs customer service model in the pending customer service task, and uses the intelligent government affairs customer service model to perform an interactive dialogue with the user, so as to obtain user demand data corresponding to the pending customer service task; a demand analysis module uses a government affairs diversion model to perform a government affairs demand analysis on user demand data, and obtains a target government affairs processing flow that matches the user demand data. Thus, through the target government affairs processing flow, it is possible to intelligently understand user needs, quickly match appropriate government affairs processing flows, and provide users with personalized and efficient government affairs services, thereby assisting in improving the level of government affairs services.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to an intelligent government affairs customer service application system. Background Art

[0002] Government customer service refers to a customer service system used to provide communication and services between government agencies or organizations and the public. Government customer service aims to interact with the public through various channels (such as telephone, website, mobile application, etc.), answer their questions about government policies, public services, administrative processes, etc., and provide necessary help and support.

[0003] The goal of government customer service is to provide convenient, efficient and accurate services to facilitate the public to obtain the government information and services they need, while also helping government agencies understand the needs and feedback of the public, thereby improving policies and services. Government customer service usually uses technical means, such as artificial intelligence and automatic answering systems, to provide quick responses and answer common questions.

[0004] In related technologies, communication technology customer service that uses traditional communication technology, such as city 12345 and administrative department telephone hotlines, cannot support high-concurrency access, resulting in poor user experience such as busy lines and disconnections. At the same time, the quality and knowledge of customer service personnel limit the effectiveness of problem solving. At present, although some related technologies also design automated question-and-answer scenarios based on simple standardized processing procedures, these scenario dialogues cannot achieve effective interaction and cannot adapt to complex business scenarios such as policy changes. In addition, in order to achieve capabilities such as high concurrency, good user experience, and timely response, it is necessary to increase manual customer service, which leads to increasing costs.

[0005] Therefore, in order to improve user experience, it is urgent to propose a new intelligent government customer service application solution to overcome the above technical problems. Summary of the invention

[0006] This application provides an intelligent government customer service application system to intelligently understand user needs, quickly match appropriate government processing procedures, and provide users with personalized and efficient government services, assist in improving the level of government service, improve government customer service efficiency and government processing efficiency, and assist in improving user experience.

[0007] In a first aspect, the present application provides an intelligent government customer service application system, including:

[0008] The task establishment module is configured to respond to the government affairs question request initiated by the user and establish a pending customer service task matching the government affairs question request; the pending customer service task is used to guide the user to further feedback his own government affairs needs;

[0009] The demand identification module is configured to call the intelligent government customer service model in the pending customer service task, and use the intelligent government customer service model to perform an interactive dialogue with the user to obtain user demand data corresponding to the pending customer service task;

[0010] The demand analysis module is configured to adopt the government affairs diversion model to perform government affairs demand analysis on the user demand data and obtain a target government affairs processing flow that matches the user demand data.

[0011] In a second aspect, an embodiment of the present application provides an intelligent government customer service application method, the method comprising:

[0012] In response to a government affairs question request initiated by a user, a pending customer service task matching the government affairs question request is established; the pending customer service task is used to guide the user to further feedback his or her own government affairs needs;

[0013] In the pending customer service task, the intelligent government affairs customer service model is called, and the intelligent government affairs customer service model is used to perform an interactive dialogue with the user to obtain user demand data corresponding to the pending customer service task;

[0014] A government affairs diversion model is adopted to perform government affairs demand analysis on the user demand data to obtain a target government affairs processing flow that matches the user demand data.

[0015] In a third aspect, an embodiment of the present application provides a computing device, the computing device comprising:

[0016] at least one processor, memory, and input-output unit;

[0017] Among them, the memory is used to store computer programs, and the processor is used to call the computer programs stored in the memory to execute the intelligent government customer service application system of the first aspect.

[0018] In a fourth aspect, a computer-readable storage medium is provided, which includes instructions, which, when executed on a computer, enable the computer to execute the intelligent government affairs customer service application system of the first aspect.

[0019] In the technical solution provided by the embodiment of the present application, in the intelligent government affairs customer service application system, the task establishment module responds to the government affairs question request initiated by the user and establishes a pending customer service task that matches the government affairs question request. Here, the task establishment module can quickly match the relevant pending customer service tasks according to the government affairs question request initiated by the user, thereby improving the efficiency and accuracy of task allocation. In addition, the module guides the user to further feedback his own government affairs needs through the pending customer service tasks, promotes the user to express the needs more clearly, and is conducive to the accuracy of subsequent demand identification and analysis. Furthermore, the demand identification module calls the intelligent government affairs customer service model in the pending customer service tasks, and uses the intelligent government affairs customer service model to perform interactive dialogues with the user to obtain user demand data corresponding to the pending customer service tasks. In this way, the module can achieve a more intelligent and targeted user interactive experience and improve the intelligent dialogue experience by calling the intelligent government affairs customer service model for interactive dialogues. In addition, the module obtains user demand data through interactive dialogues with the user, which helps to accurately understand the user's intentions and needs, further improves the accuracy of user demand extraction, and provides data support for subsequent demand analysis. Finally, the demand analysis module uses the government affairs diversion model to perform government affairs demand analysis on the user demand data, and obtains the target government affairs processing flow that matches the user demand data. In this way, the use of the government affairs diversion model to analyze user demand data can quickly and accurately match the target government affairs processing process and improve government affairs processing efficiency. Therefore, the target government affairs processing process can meet the government affairs needs of users, which helps to complete intelligent government affairs customer service, meet the government affairs needs of users, and improve service quality and efficiency.

[0020] In the technical solution of this application, through the synergy of the above modules, the intelligent government customer service system can automatically and intelligently understand user needs, quickly match appropriate government processing procedures, and provide users with personalized and efficient government services, assisting in improving the level of government services. In addition, the synergy of the above modules also greatly improves the matching accuracy of the target government processing procedures, ensures the accuracy of intelligent government customer service, avoids irrelevant answers or overly vague answers, and further improves the user's government customer service experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0022] Figure 1 It is a structural diagram of an intelligent government affairs customer service application system according to an embodiment of the present application;

[0023] Figure 2 It is a schematic diagram of the principle of an intelligent government affairs customer service model in an embodiment of the present application;

[0024] Figure 3 It is a schematic diagram of the principle of a government affairs diversion model in an embodiment of the present application;

[0025] Figure 4 It is a schematic diagram of the principle of a method for constructing a government data lake in an embodiment of the present application;

[0026] Figure 5 It is a flowchart of an intelligent government affairs customer service application method according to an embodiment of the present application;

[0027] Figure 6 It is a structural schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0030] In related technologies, communication technology customer service that uses traditional communication technology, such as city 12345 and administrative department telephone hotlines, cannot support high-concurrency access, resulting in poor user experience such as busy lines and disconnections. At the same time, the quality and knowledge of customer service personnel limit the effectiveness of problem solving. At present, although some related technologies also design automated question-and-answer scenarios based on simple standardized processing procedures, these scenario dialogues cannot achieve effective interaction and cannot adapt to complex business scenarios such as policy changes. In addition, in order to achieve capabilities such as high concurrency, good user experience, and timely response, it is necessary to increase manual customer service, which leads to increasing costs.

[0031] Therefore, in order to improve user experience, it is urgent to propose a new intelligent government customer service application solution to overcome at least one of the above technical problems.

[0032] In order to solve at least one of the above-mentioned technical problems, an embodiment of the present application provides an intelligent government affairs customer service application system.

[0033] Specifically, in the intelligent government affairs customer service application system, the task establishment module responds to the government affairs question request initiated by the user and establishes the pending customer service task that matches the government affairs question request. Here, the task establishment module can quickly match the relevant pending customer service tasks according to the government affairs question request initiated by the user, thereby improving the efficiency and accuracy of task allocation. Moreover, the module guides users to further feedback their own government affairs needs through the pending customer service tasks, promotes users to express their needs more clearly, and is conducive to the accuracy of subsequent demand identification and analysis. Furthermore, the demand identification module calls the intelligent government affairs customer service model in the pending customer service tasks, and uses the intelligent government affairs customer service model to perform interactive dialogues with users to obtain user demand data corresponding to the pending customer service tasks. In this way, the module can achieve a more intelligent and targeted user interaction experience and enhance the intelligent dialogue experience by calling the intelligent government affairs customer service model for interactive dialogues. In addition, the module obtains user demand data through interactive dialogues with users, which helps to accurately understand user intentions and needs, further improves the accuracy of user demand extraction, and provides data support for subsequent demand analysis. Finally, the demand analysis module uses the government affairs diversion model to perform government affairs demand analysis on user demand data and obtains the target government affairs processing flow matched with user demand data. In this way, the use of the government affairs diversion model to analyze user demand data can quickly and accurately match the target government affairs processing process and improve government affairs processing efficiency. Therefore, the target government affairs processing process can meet the government affairs needs of users, which helps to complete intelligent government affairs customer service, meet the government affairs needs of users, and improve service quality and efficiency.

[0034] In the technical solution of this application, through the synergy of the above modules, the intelligent government customer service system can automatically and intelligently understand user needs, quickly match appropriate government processing procedures, and provide users with personalized and efficient government services, assisting in improving the level of government services. In addition, the synergy of the above modules also greatly improves the matching accuracy of the target government processing procedures, ensures the accuracy of intelligent government customer service, avoids irrelevant answers or overly vague answers, and further improves the user's government customer service experience.

[0035] The intelligent government customer service application solution provided in the embodiment of the present application can be executed by an electronic device, which can be a server, a server cluster, or a cloud server. The electronic device can also be a terminal device such as a mobile phone, a computer, a tablet computer, a wearable device, or a dedicated device (such as a dedicated terminal device with an intelligent government customer service application system). In an optional embodiment, a service program for executing the intelligent government customer service application solution can be installed on the electronic device.

[0036] Figure 1 A schematic diagram of an intelligent government affairs customer service application system provided in an embodiment of the present application is shown in FIG. Figure 1As shown, the system includes the following functions:

[0037] The task establishment module 101 is configured to respond to the government affairs question request initiated by the user and establish a pending customer service task matching the government affairs question request; the pending customer service task is used to guide the user to further feedback his own government affairs needs;

[0038] The demand identification module 102 is configured to call the intelligent government customer service model in the pending customer service task, and use the intelligent government customer service model to perform an interactive dialogue with the user to obtain user demand data corresponding to the pending customer service task;

[0039] The demand analysis module 103 is configured to adopt the government affairs diversion model to perform government affairs demand analysis on the user demand data to obtain a target government affairs processing flow that matches the user demand data.

[0040] Through the synergy of the above modules 101 to 103, the intelligent government customer service system realizes automated and intelligent understanding of user needs, quickly matches appropriate government processing procedures, and provides users with personalized and efficient government services, effectively assisting in improving the level of government services. These modules work together to greatly improve the matching accuracy of the target government processing procedures, ensure the accuracy of the intelligent government customer service system, avoid irrelevant or ambiguous answers, and further improve the user's government customer service experience. Users can obtain the required information and services more quickly, while government departments can also provide more accurate and personalized support to promote the modernization and efficiency of government services.

[0041] This intelligent government customer service method will provide users with a more convenient and thoughtful service experience, help enhance mutual trust between the government and citizens, and promote the development of social science and technology. In addition, this intelligent government customer service method has significantly improved the response speed of government customer service, improved the problems of low efficiency of traditional manual customer service and long waiting time for users, and thus greatly optimized the user experience. The intelligent government customer service system can quickly and accurately provide users with the information and services they need, allowing users to experience efficient and convenient service experience.

[0042] In practical applications, the intelligent government customer service application system can be implemented based on deep learning and natural language processing (NLP) technology.

[0043] First, in the task establishment module 101, in response to the government affairs question request initiated by the user, a pending customer service task matching the government affairs question request is established. Here, the pending customer service task is used to guide the user to further feedback his own government affairs needs.

[0044] For example, when a user initiates a government affairs question request, creating a pending customer service task can guide the user to further feedback their government affairs needs, so as to better understand the user's problem and provide accurate help. The following is an example:

[0045] Assume that a user initiates a government affairs inquiry request: "I want to know the process of applying for a business license, can you help me?" Then, the task creation module 101 creates a pending customer service task and replies as follows: "Thank you for your question! In order to help you better apply for a business license, I need to further understand your specific situation. In which region are you going to apply for a business license? Have you prepared the required materials?" Through such a reply, the task creation module 101 guides the user to provide more relevant information, such as regional information and material preparation, so that subsequent modules can more accurately understand the user's needs and provide the user with corresponding application procedures and guidance.

[0046] In another example, when a user initiates a government affairs question request, the task establishment module 101 can establish a matching pending customer service task based on the content of the user's question, and guide the user to further feedback his or her government affairs needs. The following is an example:

[0047] The user initiates a government affairs inquiry request: "I want to inquire about public transportation routes. Can you help me check?" Then, the task establishment module 101 establishes a pending customer service task and replies: "Thank you for your inquiry! In order to better help you query public transportation route information, I need to further understand your travel needs. Please tell me the specific starting point and destination you want to query? Do you choose to take the bus, subway or other means of transportation?" Through such a reply, the task establishment module 101 guides the user to provide more detailed information, such as the starting point, destination and travel options, so that the system can accurately query and provide relevant public transportation route information. This pending customer service task can help the system better understand the user's needs and provide targeted help, thereby improving user experience and service quality.

[0048] This pending customer service task can prompt users to describe their needs more clearly, and also provide more information to the system, which helps to improve the understanding and solution capabilities of user needs. By guiding users to further feedback their own government needs, the intelligent government customer service system can better provide users with personalized and accurate services and enhance user experience.

[0049] Exemplarily, the task establishment module 101 can use the intent recognition model to establish the pending customer service task that matches the government affairs question request. Here, the intent recognition model is introduced to preliminarily judge the user's specific intention based on the text content of the user's question, such as querying information, handling business, making complaints and suggestions, etc. Therefore, the task establishment module 101 establishes the corresponding pending customer service task based on the intent recognition result.

[0050] In practical applications, the intent recognition model can be established based on at least one of a machine learning algorithm, such as Naive Bayes, Support Vector Machine (SVM), a deep learning model such as a Recurrent Neural Network (RNN), and a Long Short-Term Memory Network (LSTM). The trained intent recognition model can predict the user input text and identify the user's intent category.

[0051] In addition, in another optional embodiment, the task establishment module 101 can also use a task guidance system to establish a pending customer service task that matches the government question request. Specifically, a task guidance system can be pre-established to push corresponding questions based on the content of the user's possible questions to guide the user to further refine his or her needs. The user answers the questions raised by the task establishment module 101, and then the task establishment module 101 establishes a pending customer service task based on the user's answer.

[0052] For example, in the government affairs customer service system, the workflow of implementing the task establishment module 101 to establish the pending customer service task matching the government affairs question request by means of the task guidance system is as follows:

[0053] Assuming that a user initiates a government affairs question request regarding the residence certificate application process, the task establishment module 101 needs to establish a corresponding pending customer service task based on the question raised by the user.

[0054] Then, the user inputs "I want to know the application process of residence certificate", and the task establishment module 101 receives the user's question text. Then, the task guidance system pushes relevant questions to the user based on the preset question template and the government affairs application process knowledge base, such as "Are you applying for a new residence certificate or renewing it?", "Do you need to apply for a personal or unit residence certificate?", etc. The user answers the questions raised by the task establishment module 101, such as "I am a new application for a personal residence certificate." Therefore, the task establishment module 101 automatically establishes pending customer service tasks based on the user's answers and the preset logic of the task guidance system, including sorting out user needs, classifying information, assigning processing personnel, etc.

[0055] The task guidance system can provide a response to the user after completing the pending customer service task, such as providing information such as the specific process and required materials for obtaining a residence certificate, in order to resolve the user's questions or problems.

[0056] Through the guidance of the task guidance system and the process of user answers, the task establishment module 101 can establish corresponding pending customer service tasks according to the user's gradually refined needs, thereby providing more personalized and professional services. This method can not only help users express their needs more clearly, but also improve customer service work efficiency and service quality.

[0057] In the demand identification module 102, first, the intelligent government customer service model is called in the pending customer service task, and the intelligent government customer service model is used to perform an interactive dialogue with the user to obtain user demand data corresponding to the pending customer service task.

[0058] For example, suppose a user initiates a government affairs question request. The system creates a pending customer service task through the task establishment module 101. The intelligent government affairs customer service model in the demand identification module 102 is required to identify the user's demand and provide corresponding services. Exemplarily, the system calls the intelligent government affairs customer service model, greets the user and asks: "Hello, do you have any questions that need help?" The user replies: "I want to know how to apply for a residence certificate." Then, the intelligent government affairs customer service model analyzes the user's answer, extracts the user's specific demand, namely, the residence certificate application, and replies to the user: "You can go to the Citizen Service Center to apply for a residence certificate. You need to prepare identity documents and residence address proof documents." User feedback: The user may further ask: "How long does it take to apply?" The intelligent government affairs customer service model can further analyze and extract this demand.

[0059] As an optional embodiment, the intelligent government customer service model includes at least: a multimodal input layer, a multimodal feature fusion layer, a context-aware continuous dialogue layer, and a transfer learning layer.

[0060] In this optional embodiment, the intelligent government customer service model includes a multimodal input layer, a multimodal feature fusion layer, a context-aware continuous dialogue layer and a transfer learning layer.

[0061] The multimodal input layer is used to receive information from multiple input sources, such as text, voice, and images, to fully obtain the user's input information. For example, users can interact in multiple ways, such as inputting questions in text, describing requirements in voice, and uploading pictures or files. The multimodal input layer can process these different types of input information.

[0062] The multimodal feature fusion layer is used to fuse features from different data sources in the multimodal input layer to improve the richness and accuracy of feature representation. By fusing features from multiple data sources such as text, speech, and images, the model can understand user needs more comprehensively and improve its ability to recognize and understand user intent.

[0063] The context-aware continuous conversation layer is responsible for handling the continuous conversation between the user and the system, taking into account the context information of the conversation in order to better understand the user's intention and provide a coherent conversation experience. By maintaining the conversation history and context information, the model can achieve more intelligent conversation management, considering not only a single conversation but the overall conversation process, thereby providing more coherent and personalized services.

[0064] The transfer learning layer shares knowledge and model parameters between different tasks to achieve knowledge transfer on new tasks, accelerate model training and improve performance. In the government sector, the transfer learning layer can transfer knowledge and experience from existing tasks to new tasks, improve the model's ability to understand and solve specific problems in the government sector, and shorten the training time and cost of the model in the government sector.

[0065] Combining the functions of the above layers, the intelligent government customer service model can receive and understand user input information more comprehensively, provide personalized, coherent and efficient government customer service services, and improve user experience and service quality. At the same time, the introduction of the transfer learning layer can accelerate the application and optimization process of the intelligent government customer service model in the government field.

[0066] Based on the above structure, the demand identification module 102, when using the intelligent government customer service model to perform interactive dialogue with the user to obtain user demand data corresponding to the customer service task to be processed, see Figure 2 As shown, this module can implement the following steps:

[0067] 201. Generate first interaction information with a user through a multimodal input layer using an interactive dialogue generation model matched with a pending customer service task; obtain multimodal interaction data from second interaction information fed back by the user; the multimodal interaction data at least includes dialogue text, dialogue image, and dialogue audio;

[0068] 202. Extracting initial demand language features corresponding to different modal interaction data and associated interaction features between different modal interaction data from the multimodal interaction data through a multimodal feature fusion layer;

[0069] 203. Converting the initial demand language features and the associated interaction features extracted based on the second interaction information into a first interaction dialogue tracking feature sequence through the context-aware continuous dialogue layer according to the context dialogue semantic relationship and the pre-input historical interaction dialogue feature sequence;

[0070] 204. Performing semantic understanding processing on the first interactive dialogue tracking feature sequence through the context-aware continuous dialogue layer, and generating third interactive information with the user based on the semantic understanding result;

[0071] 205. Obtain multimodal interaction data from fourth interaction information fed back by the user based on the third interaction information through the multimodal input layer;

[0072] 206. Converting the first interactive dialogue tracking feature sequence, the initial demand language features and the associated interactive features extracted based on the fourth interactive information into a second interactive dialogue tracking feature sequence through the context-aware continuous dialogue layer according to the context dialogue semantic relationship and the pre-input historical interactive dialogue feature sequence;

[0073] 207. Through the transfer learning layer, the government dialogue demand recognition model is used to perform semantic understanding processing on the first interactive dialogue tracking feature sequence and the second interactive dialogue tracking feature sequence to obtain user demand data.

[0074] For example, in this demand identification module 102, the identification and interaction process of user needs in government dialogue is realized through the multimodal input layer, the multimodal feature fusion layer, the context-aware continuous dialogue layer and the transfer learning layer. The following will introduce the interaction process realized by the demand identification module 102 with examples of specific dialogues:

[0075] First, the interaction process corresponding to the generation of the first interaction information is as follows:

[0076] User: How do I apply for medical insurance?

[0077] System: Through the interactive dialogue generation model matched with the pending customer service tasks, the system generates the first interactive information: Hello, you need to apply for medical insurance. I can help you understand the relevant procedures and requirements.

[0078] Secondly, the interaction process corresponding to the second interaction information extraction is as follows:

[0079] User feedback: Uploaded relevant medical insurance information, including medical records, ID card and insurance policy.

[0080] System: Acquire multimodal interaction data, such as conversation text (text information provided by users), conversation images (uploaded medical records, ID card images), and conversation audio (user voice notes).

[0081] Again, the first interactive dialogue tracking feature generation process is as follows:

[0082] Through the context-aware continuous dialogue layer, the system extracts the initial demand language features (medical insurance processing requirements) and related interaction features (relevant information uploaded by the user), and converts them into the first interactive dialogue tracking feature sequence for further semantic understanding and processing.

[0083] Next, the third interaction information generates the corresponding interaction process as follows:

[0084] The system performs semantic understanding processing based on the feature sequence of the first interactive dialogue tracking and generates the third interactive information: Thank you for uploading the relevant information. We will process your medical insurance application as soon as possible.

[0085] Then, the interaction process corresponding to the fourth interaction information extraction is as follows:

[0086] User feedback: Inquire about the processing progress and the information that needs to be supplemented.

[0087] System: again obtains fourth interaction information fed back by the user based on the third interaction information through the multimodal input layer, including text query and possible attached pictures or files.

[0088] Furthermore, the context-aware continuous dialogue layer generates a second interactive dialogue tracking feature sequence for subsequent demand identification and understanding based on the historical interactive dialogue feature sequence, the first interactive dialogue tracking feature sequence, and the initial demand language features and associated interaction features extracted from the fourth interactive information.

[0089] Finally, the transfer learning layer uses the government dialogue demand recognition model to perform semantic understanding processing on the first interactive dialogue tracking feature sequence and the second interactive dialogue tracking feature sequence, and finally obtains user demand data, such as processing progress query and information supplement requirements.

[0090] Through the above steps, we can identify user needs and provide corresponding services and support based on multiple rounds of interactions between users and the system, combined with multimodal interaction data and contextual conversation semantic relationships. This approach based on multimodal input and continuous dialogue can improve user experience and help understand and meet user needs more intelligently.

[0091] Among them, the government dialogue demand recognition model is obtained through parameter preprocessing and fine-tuning based on various customer service tasks and corresponding historical government dialogue data. Further, optionally, the system needs to collect relevant data of various customer service tasks, including information such as problems, needs, solutions, etc. in the government field. At the same time, it is also necessary to obtain a large amount of historical government dialogue data, which contains dialogue content such as user questions, system replies, and user feedback. Before training the government dialogue demand recognition model, the collected data needs to be parameter preprocessed, including data cleaning, noise removal, tokenization, etc., to ensure the quality and consistency of the data. For multimodal data, specific preprocessing is also required, such as word segmentation of text data and feature extraction of image data. Then, the customer service task data and historical dialogue data after parameter preprocessing are used to fine-tune the government dialogue demand recognition model. During the fine-tuning process, the model will improve its understanding of the context of the government field and its ability to recognize cross-task needs by learning the patterns and rules in the data. By gradually adjusting the model parameters and the training process, the model gradually converges and improves performance. After fine-tuning is completed, the government dialogue demand recognition model needs to be evaluated, and the performance of the model on the government dialogue demand recognition task needs to be evaluated using the test set data. Here, the evaluation indicators can include accuracy, recall, F1 value, etc. to determine whether the model achieves the expected recognition effect. After the performance of the government dialogue demand recognition model is evaluated and confirmed, the government dialogue demand recognition model can be deployed to the actual system to provide government customer service with the ability to intelligently identify user needs. After the government dialogue demand recognition model is deployed, it can receive user input in real time, quickly and accurately identify user needs, and guide the subsequent dialogue interaction process.

[0092] Through the above functional steps, based on the parameter preprocessing and fine-tuning of various customer service tasks and historical government dialogue data, the government dialogue demand identification model can better adapt to the characteristics of the government field, improve the ability to accurately identify user needs, and thus provide more intelligent and efficient services for the government customer service system.

[0093] Optionally, the intelligent government customer service model further includes: a model dynamic enhancement layer. The model dynamic enhancement layer is a key component of the intelligent government customer service model, which is used to perform dialogue evaluation processing on the initial demand language features and related interaction features, and dynamically enhance the interactive dialogue generation model according to the dynamic dialogue evaluation results, thereby optimizing the performance and effect of the model.

[0094] As an optional embodiment, the demand identification module 102, after using the intelligent government customer service model to execute an interactive dialogue with the user to obtain user demand data corresponding to the customer service task to be processed, is also specifically configured to: perform dialogue evaluation processing on the initial demand language features and the associated interaction features through the model dynamic reinforcement layer to obtain a dynamic dialogue evaluation result; and use the dynamic dialogue evaluation result to dynamically strengthen the interactive dialogue generation model to achieve optimization processing of the interactive dialogue generation model.

[0095] Specifically, the model dynamic reinforcement layer will evaluate the interactive dialogue process between each round of users and the system, including analyzing and evaluating the associated interactive features of the user's demand language and the system's response, so as to understand the progress of the dialogue and the gradual evolution of user needs. Based on the dynamic dialogue evaluation results, the model dynamic reinforcement layer will dynamically strengthen and optimize the interactive dialogue generation model. This dynamic reinforcement can include parameter adjustment, model update, enhanced training, etc. to improve the performance and accuracy of the model in actual dialogue scenarios. Through the evaluation process, the model dynamic reinforcement layer will obtain dynamic dialogue evaluation results, which reflect the quality, effectiveness and user satisfaction of each round of dialogue, help the system identify advantages and disadvantages, and provide a basis for further optimization. Through the dynamic reinforcement model, the model can be dynamically adjusted according to the evaluation of real-time dialogue to adapt to different user needs and dialogue scenarios, improve the fluency and intelligence of dialogue interaction, and thus improve user experience and service quality.

[0096] In general, the model dynamic reinforcement layer plays a key role in the intelligent government customer service model. Through the dynamic processing of dialogue evaluation and model optimization, the interactive dialogue generation process is continuously optimized and fine-tuned, thereby improving the intelligence, adaptability and user satisfaction of the system. This dynamic adjustment and optimization mechanism helps to ensure the efficient operation and continuous improvement of the government customer service system.

[0097] Furthermore, in order to improve the robustness and generalization ability of the intelligent government customer service model, data enhancement techniques such as Generative Adversarial Networks or Variational Autoencoders can be used to expand the training data. At the same time, multiple optimized interactive dialogue generation models can be integrated to obtain more accurate and reliable answer results through model fusion or voting mechanisms.

[0098] As an optional embodiment, the intelligent government customer service model also includes: a model integration layer. Specifically, the model integration layer is also an important component of the intelligent government customer service model, which is used to perform multimodal data enhancement processing on historical government dialogue data corresponding to different types of customer service tasks, build interactive dialogue generation models suitable for different tasks, and integrate these models into target interactive dialogue generation models suitable for various types of customer service tasks through model fusion processing.

[0099] Based on this, the demand identification module 102, before using the intelligent government customer service model to perform an interactive dialogue with the user to obtain user demand data corresponding to the customer service task to be processed, is further specifically configured as follows:

[0100] Through the model integration layer, multimodal data enhancement processing is performed on the historical government dialogue data corresponding to different types of customer service tasks to obtain a multimodal extension set of government dialogues; using the multimodal extension set of government dialogues, multiple interactive dialogue generation models are constructed for different types of customer service tasks, and the multiple interactive dialogue generation models are integrated into a target interactive dialogue generation model suitable for different types of customer service tasks through model fusion processing.

[0101] It is understandable that the model integration layer uses historical government dialogue data for multimodal data enhancement processing, combines data in different forms such as text, voice, and images, and constructs a richer and more comprehensive multimodal expansion set of government dialogue to improve the training effect and generalization ability of the model. Based on the multimodal expansion set of government dialogue, the model integration layer constructs multiple interactive dialogue generation models for different types of customer service tasks, optimizes and models data and task requirements of different modalities to meet the specific needs and scenarios of different customer service tasks. For multiple interactive dialogue generation models, the model integration layer performs model fusion processing, integrates the advantages and characteristics of each model, and obtains a more comprehensive and comprehensive target interactive dialogue generation model. This model fusion can be achieved through technical means such as integrated learning, model stacking, and model fusion. The final integrated interactive dialogue generation model can be applied to different types of customer service tasks, has higher versatility and adaptability, can handle diverse customer needs and problems, and improve the overall efficiency and service quality of the government customer service system.

[0102] Through the model integration layer, the government customer service system can better utilize multimodal data for modeling and training, build a powerful interactive dialogue generation model suitable for various customer service tasks, and improve the system's intelligence level and effectiveness in handling government dialogues and serving users. This model integration approach can effectively integrate various resources and information, and improve the overall performance of the system and user experience.

[0103] For example, the application of the model integration layer in the intelligent government customer service model can be explained through an example. Assume that there are two different types of government customer service tasks: applying for government benefits and querying service procedures. In order to better meet the requirements of these two tasks, the model integration layer will be used to optimize the model and service performance.

[0104] For the task of applying for government benefits, historical government dialogue data can be collected and sorted, and through multimodal data enhancement processing, data in different forms such as text, voice and image can be combined to construct a multimodal expansion set of government dialogue for the task of applying for government benefits. Similarly, for the task of querying the service process, similar data enhancement processing is also performed to obtain a multimodal expansion set of government dialogue for this task. Based on the multimodal expansion set of government dialogue for the task of applying for government benefits, a special interactive dialogue generation model is constructed for this task, and modeling and training are performed based on historical data and demand characteristics. Similarly, for the task of querying the service process, an independent interactive dialogue generation model is also established. Next, through the model integration layer, multiple interactive dialogue generation models constructed for different tasks are fused, and these interactive dialogue generation models are integrated into a target interactive dialogue generation model suitable for different types of customer service tasks. In the model fusion processing, ensemble learning methods such as voting method, weighted voting method or model fusion algorithm can be used to integrate the outputs of multiple models to achieve better prediction performance.

[0105] Through the above process, the model integration layer can effectively combine the historical conversation data of different types of customer service tasks, build a multimodal data enhancement set and create an interactive dialogue generation model for specific tasks, and finally integrate these models into a target model suitable for different types of customer service tasks through model fusion processing. This approach can improve the intelligent performance and adaptability of the government customer service system under different types of tasks, and provide users with a more personalized and efficient service experience.

[0106] As an optional embodiment, the demand identification module 102, through the model integration layer, performs multimodal data enhancement processing on the historical government dialogue data corresponding to different types of customer service tasks to obtain a multimodal expansion set of government dialogues, and is specifically configured as follows:

[0107] For text data in historical government dialogue data, the keywords in the text data are replaced with synonyms or near-synonyms; and / or the text data is converted into sentence structures; and / or the dialogue sentences between different roles in the text data are expanded through dialogue bionics; and / or the text data is noise-filled to obtain expanded government dialogue text data.

[0108] For the specific configuration of multimodal data enhancement processing of historical government dialogue data corresponding to different types of customer service tasks through the model integration layer in the demand identification module 102, we can perform different types of text data processing, such as synonym replacement, sentence structure conversion, dialogue bionic expansion and noise filling. The following are examples of these processing methods for text data in historical government dialogue data:

[0109] For example, the synonym or near-synonym replacement process can be, assuming that the original conversation sentence is: "User: I want to know how to handle medical insurance procedures." Based on this, the sentence after synonym replacement is: "User: I hope to know how to apply for medical insurance procedures." In this example, by replacing the keywords "want to know" with "hope to know" and "how to handle" with "how to apply", the conversation sentences can be made richer and more diverse, increasing the diversity of historical conversation data.

[0110] For example, the sentence structure conversion process can be, assuming that the original conversation sentence is: "Government agency: Please provide your identity certificate and related materials." Based on this, the sentence after sentence structure conversion is: "Please submit your identity certificate and related materials to handle the formalities." By performing sentence structure conversion on the original conversation sentence, the expression method and sentence structure can be changed, making the data more diverse, which helps the model understand the user's intention more comprehensively.

[0111] For example, the conversation bionic expansion process can be, assuming that the original conversation sentence is: "Government agency: Please fill out the application form and submit it to the designated office location". Based on this, the conversation after conversation bionic expansion is: "User: Can I fill out the form online and upload it directly? Government agency: Yes, we support filling out and uploading materials online". By simulating the conversation interaction between different roles, new conversation expansion data can be generated, the number of conversation sentences and expression forms in the text data can be increased, the coverage of the historical conversation data set can be expanded, and the generalization ability of the model can be improved.

[0112] For example, the noise filling process can be: assuming that the original conversation sentence is: "Government agencies: Please bring your original identity certificate to the service hall for processing". Based on this, the conversation after noise filling is: "Government agencies: Please bring your identity certificate printout to the service hall for processing, and prepare the required materials". By adding noise to text data, such as inserting redundant information or grammatical errors, the diversity and complexity of the conversation data can be increased, the robustness and generalization ability of the model can be improved, and the anti-interference ability of the model can be improved.

[0113] By performing multimodal data enhancement processing on historical government dialogue data in these ways, the diversity and complexity of the data can be effectively enriched, providing a richer and more comprehensive data foundation for model training, thereby improving the performance and effect of the intelligent government customer service model in demand identification and dialogue generation. This processing method helps the model better understand user needs and provide more accurate and personalized services.

[0114] Specifically, conversational bionic augmentation is a method to augment a dataset by simulating the interactions between different characters in real conversations. This processing method can help enrich historical conversation datasets and improve the generalization ability of the model and its adaptability to different situations. The following is a specific example to explain the method of conversational bionic augmentation. Assume that the original conversation is as follows:

[0115] User: I would like to know how to apply for a disability certificate.

[0116] Government agencies: You can bring your identity card and relevant supporting documents to the community affairs reception center for processing.

[0117] Based on the above dialogue, we can get the following dialogue bionic expansion:

[0118] User: Can I apply for a disability certificate online?

[0119] Government agencies: We are sorry, our system does not support online applications at the moment. You can choose to go to the community affairs reception center to handle it.

[0120] In this example, the original conversation was biomimetically expanded to generate an interactive content that simulates a real conversation. The user asked questions about online applications, and the government agency responded accordingly. Such biomimetic expansion of the conversation can help the model better understand the different questions and needs that users may raise, and improve the generalization and response capabilities of the dialogue system.

[0121] Another example of conversation bionic expansion, assuming the original conversation is as follows:

[0122] User: What materials are needed to apply for medical insurance?

[0123] Government agencies: You will need to provide proof of identity, social security card and relevant medical certificates.

[0124] Based on the above dialogue, we can get the following dialogue bionic expansion:

[0125] User: What procedures are required for medical insurance?

[0126] Government agencies: When applying for medical insurance, you need to prepare your identity card, social security card and relevant medical certification documents.

[0127] Through bionic augmentation, we simulated variations of users’ questions and responses from government agencies. This approach can increase the diversity of conversation datasets, make the model better adapt to different questioning methods and expressions, and improve the performance of intelligent government customer service systems in demand identification and conversation generation.

[0128] Optionally, the dialogue bionic expansion is realized through rule engine and reinforcement learning. That is, first, a series of rules are pre-set to simulate the dialogue situation between government agencies and users, and different user questions and government agency responses are defined in the rules. Based on professional knowledge and experience, common problems, common responses and dialogue interaction logic are determined to establish a rule base. The conditions and actions of the rules are defined to simulate the real dialogue situation, and the trigger conditions and execution logic of the rules are formulated. Then, the pre-set rules are converted into computer executable code to build a rule engine system. The rule engine system can match the corresponding rules and generate the government agency's response according to the user's questions and the current dialogue situation. The rule engine realizes the automatic generation of dialogue, which can effectively simulate different dialogue situations and expand dialogue data. Then, the reinforcement learning framework is introduced to establish an interactive model between the intelligent agent and the environment. The intelligent agent is responsible for simulating the user's questioning behavior, and the environment is responsible for providing the corresponding government agency response. The reinforcement learning algorithm can adjust the behavior of the intelligent agent according to the reward signal to generate dialogue content that conforms to the dialogue logic and the real situation. Through reinforcement learning, the intelligent agent can continuously learn and optimize the ability to generate dialogues, and improve the quality and fidelity of the simulated dialogue. Finally, the dialogue data generated by the rule engine is used as training data, and reinforcement learning is introduced to further optimize the dialogue generation process. Under the reinforcement learning framework, the agent can learn and explore based on the dialogue data generated by the rule engine to improve the accuracy and naturalness of dialogue generation. Combining the rule engine and reinforcement learning methods can achieve the goal of dialogue bionic expansion, generate more diverse and realistic dialogue data, and improve the system's performance in dialogue generation tasks.

[0129] Through the combined application of rule engines and reinforcement learning, it is possible to achieve automatic generation, continuous learning and optimization of dialogues, thereby generating simulated dialogue data that conforms to real dialogue scenarios, providing more diverse and intelligent data support for model training and the development of intelligent systems.

[0130] Another optional embodiment is to perform at least one of random cropping, flipping, image rotation, scaling, pixel expansion, and color conversion processing on the image data in the historical government dialogue data to obtain government dialogue image expansion data.

[0131] For example, the image data in the historical government dialogue data can be augmented by using random cropping in the image data augmentation technology. Random cropping is a commonly used data augmentation method. By randomly selecting and cropping areas on the image, diverse training data is generated to improve the generalization ability of the model. The following is an example of how to use random cropping to augment the image data in the historical government dialogue data. Assume that the original image data is, and the image data in the historical government dialogue data contains a picture showing the government affairs process. Based on this, first select a random cropping area. Randomly select an area in the original image as the cropping area. Secondly, perform random cropping. Crop the selected area to generate a new image. Finally, save the data. Save the cropped image as augmented data and record the cropping location information.

[0132] In another example of augmented data, the original image showing the government affairs process was randomly cropped to generate multiple images with different cropped areas. Each image shows a different part or scale of the government affairs process, thus forming diverse augmented data. Augmented data can be used for model training to increase the model's adaptability to different scenarios and perspectives.

[0133] Through random cropping, we can generate diversified government dialogue image augmentation data to improve data diversity and training effects. In addition to random cropping, we can also combine other image enhancement techniques such as flipping, image rotation, scaling, etc. to further enrich the augmented data and improve the performance and generalization ability of the model.

[0134] For example, pixel augmentation is a common image data enhancement technique. By transforming and operating at the pixel level of the image, new augmented data is generated, thereby increasing the diversity and richness of the data. The following is an example to illustrate pixel augmentation. Assume that the original image data is an image data in the historical government dialogue data, which contains a photo of a government staff office scene. Based on this, by performing pixel interpolation operations on the original image, the resolution of the image can be increased or decreased to generate images with different pixel densities. By performing scale operations on the image, the size of the image can be enlarged or reduced to generate images of different sizes. Filling around the image, expanding the boundary pixels of the image, and generating a new augmented image. Adding a certain degree of pixel noise to the image changes the pixel value of the image, so that the image presents a certain degree of change. Transforming the pixel value of the image, such as color adjustment, contrast increase, etc., generates images with different visual effects. Through the pixel augmentation method, multiple variants of photos of government staff office scenes can be generated, including images with different resolutions, scales, padding, noise levels, and transformation effects. These augmented data can be used to increase the model's ability to recognize different pixel features and image changes, and improve the model's robustness and generalization ability.

[0135] Through pixel augmentation technology, image data can be transformed and processed at the pixel level to generate diverse augmented data, providing richer data resources for model training, which helps improve the performance of the model in government dialogue image recognition and analysis tasks.

[0136] In another optional embodiment, for the audio data in the historical government dialogue data, at least one of the following processes is performed on the audio data: audio speed change, reverberation, noise reduction, enhancement, copying, transformation, cutting, and combination, so as to obtain the government dialogue audio expansion data.

[0137] For example, for the audio data in the historical government dialogue data, the audio data can be expanded by changing the speed of sound. Sound speed change is an audio processing technology that changes the pitch and speed of the sound by adjusting the playback speed of the audio signal, thereby generating diversified audio data. The following is an example to illustrate the processing process of sound speed change, assuming that the original audio data is: The audio data in the historical government dialogue data contains a speech recording in a government propaganda film. Based on this, the sound speed change processing is as follows: First, select the speed change ratio. Select a speed change ratio according to the needs, which can be to speed up or slow down. Second, sound speed change processing. The original audio data is speeded up or slowed down, and the audio playback speed is accelerated or slowed down, and the tone of the audio is changed accordingly. Third, save the processing results. Save the audio data after speed change processing as the expansion data, and record the speed change ratio information. Fourth, the expansion data example. Through the sound speed change processing, multiple versions of government propaganda film speech recordings that are accelerated or slowed down can be generated, and each version of the audio data has a different tone and speech speed. The audio data after speed-changing processing can be used to train speech recognition models, sentiment analysis models, etc., to increase the model's ability to recognize different speaking speeds and tones, and improve the model's robustness and generalization ability.

[0138] Through sound speed-changing processing, the playback speed and pitch of audio data can be changed to generate diverse audio expansion data, providing richer data resources for model training, which helps improve the performance of the model in government dialogue audio recognition and analysis tasks. In addition to sound speed-changing, other audio processing technologies such as reverberation, noise reduction, and enhancement can also be combined to further enrich audio expansion data and improve model performance.

[0139] Further optionally, the demand identification module 102 integrates the multiple interactive dialogue generation models into a target interactive dialogue generation model suitable for different types of customer service tasks through model fusion processing, and is specifically configured as follows:

[0140] The dialogue generation results of each of the multiple interactive dialogue generation models are obtained; based on the dialogue generation results and the pre-added noise results, the multiple interactive dialogue generation models are voted to obtain the interactive dialogue generation model to be selected; and a pre-set meta-model is used to perform a fusion analysis on the dialogue generation results of each of the interactive dialogue generation models to be selected to obtain the target interactive dialogue generation model.

[0141] For example, for the model fusion processing in the demand identification module 102, a voting method and a meta model are used to integrate multiple interactive dialogue generation models, so as to generate a target interactive dialogue generation model suitable for different types of customer service tasks. Exemplarily, it is assumed that there are three different interactive dialogue generation models: interactive dialogue generation model A, interactive dialogue generation model B, and interactive dialogue generation model C. Step 1: Obtain the dialogue generation results of each interactive dialogue generation model.

[0142] Right now:

[0143] 1. Dialogue results generated by interactive dialogue generation model A: User question -> Answers from interactive dialogue generation model A

[0144] 2. Dialogue results generated by interactive dialogue generation model B: User question -> Answers from interactive dialogue generation model B

[0145] 3. Dialogue results generated by interactive dialogue generation model C: User question -> Answers from interactive dialogue generation model C

[0146] Step 2: Select the candidate interactive dialogue generation model based on voting. For each user question, the three models generate answers respectively, and the answer with the most votes is selected by voting and identified as the candidate interactive dialogue generation model. Step 3: Use meta-model for fusion analysis. For example, the meta-model can be a weighted combination model or other neural network model. Perform fusion analysis on the dialogue generation results of each candidate interactive dialogue generation model, and generate the final result of the target interactive dialogue generation model by combining the pre-set weights or strategies.

[0147] For example, a user asks a question about the government affairs handling process, and model A answers "Please log in to the official website for inquiry", model B answers "Please call the government affairs consultation hotline for help", and model C answers "Please go to the local government service hall for consultation". Through the voting method, the answer of model C is selected as the candidate interactive dialogue generation model. The meta-model is used for fusion analysis, combining the answer results of each model with the preset analysis strategy to generate the answer of the final target interactive dialogue generation model, such as "Please go to the local government service hall for consultation, or log in to the official website for inquiry".

[0148] Through the above steps, we can use the model fusion processing method to combine the advantages of multiple interactive dialogue generation models to generate a more comprehensive target interactive dialogue generation model suitable for different types of customer service tasks, thereby improving the performance and effectiveness of the system.

[0149] As an optional embodiment, in the government affairs diversion model, there are an input layer, a demand feature extraction layer, a government affairs demand matching layer and a target matching layer, which are used to analyze the government affairs demand of user demand data to obtain a matching target government affairs processing flow. The functions and implementation steps of each layer are described in detail below: Figure 3 As shown, the demand analysis module 103 uses the government affairs diversion model to perform government affairs demand analysis on the user demand data, and obtains the target government affairs processing flow matching the user demand data, which is specifically implemented in the following steps:

[0150] 301. Extract corresponding government demand information from user demand data through the input layer;

[0151] 302. Converting the government demand information into corresponding multimodal government demand features through the demand feature extraction layer;

[0152] 303. Through the government affairs demand matching layer, the multimodal government affairs demand characteristics are input into the demand classification model to perform government affairs demand classification processing, so as to obtain the user's cross-modal government affairs demand characteristics;

[0153] 304. Through the target matching layer, a meta-learning training model is adopted to perform process decomposition and rapid learning processing on cross-modal government affairs demand characteristics to obtain the matching target government affairs processing process.

[0154] In the input layer, government demand information is extracted from user demand data, including government processing flow information, government item types, and related government attribute information. Government-related information, such as problem descriptions and keywords, is extracted from the demand data input by users. The extracted information is preprocessed and standardized to ensure the accuracy and completeness of the information.

[0155] In actual applications, government demand information at least includes: government processing flow information, government matter type, and related government attribute information. For example, suppose a user needs to know how to apply for a government service such as medical assistance. In this scenario, the user puts forward a specific demand, and the government demand information may include the following: government processing flow information, that is, guiding the user on how to complete the application process for medical assistance in accordance with the prescribed steps and procedures. For example, it is necessary to explain how the user should fill in the application form, submit relevant materials, and conduct an interview. Government matter type refers to the specific type of government service, which is medical assistance in this case. Government matter type can help the system accurately locate the specific service content required by the user so as to provide relevant guidance and assistance. Related government attribute information, including relevant government attribute information such as the user's region, personal identity information, and health status, is used to ensure that the matched government processing flow meets the user's actual demand background. For example, there may be subtle differences in medical assistance policies in different regions, and the system needs to consider the specific regulations in the user's region. In this example, the comprehensive use of government demand information can help the government service system accurately understand the user's needs, provide corresponding guidance and assistance, ensure that users obtain accurate and personalized government service information, and improve user satisfaction and service efficiency.

[0156] In the demand feature extraction layer, the government demand information is converted into multimodal government demand features. In this layer, text features, image features, and audio features are extracted based on the input government demand information. Each type of feature is processed and encoded and converted into a unified multimodal representation. It is ensured that different types of features can effectively represent the key information of government demand. In practical applications, multimodal government demand features include: at least one of text government demand features, image government demand features, and audio government demand features. For example, suppose a user uploads a medical report containing text and pictures through the government platform and records a voice describing the condition, hoping that the system can help him understand the relevant medical policies and service processes. In this scenario, the user provides multimodal data, and the system can extract the following multimodal government demand features to better understand the user's needs: text government demand features, that is, the system can extract government demand features from the text information provided by the user. For example, the system can analyze the text content through natural language processing technology, extract keywords, entities, topics and other information, to understand the user's description of the condition, demand, etc. Image government demand features, that is, the user uploads a picture of the medical report, and the system can extract government demand features from the image. For example, the system can use computer vision technology to analyze the content of medical reports in pictures, identify text information, medical images, symptoms of the disease, etc., to understand the user's condition and needs. Audio government demand features, that is, the user recorded a voice describing the condition, and the system can extract government demand features from the audio data. For example, the system can use speech recognition technology to convert voice content into text, and then analyze key information in the text content, such as disease description, demands, etc. In this example, by comprehensively using multimodal government demand features such as text, images, and audio, the system can understand the user's needs more comprehensively and provide users with accurate and personalized government service information and guidance, thereby improving user experience and service effects.

[0157] In the government demand matching layer, the government demand classification model is used to classify and process the multimodal government demand features to obtain the user's cross-modal government demand features, including text-image fusion government demand features, text-audio government demand features, and image-audio government demand features. First, the multimodal government demand features are input into the demand classification model for classification and identification. Secondly, different types of features are combined to generate cross-modal government demand features, such as combining text and image information or text and audio information. Finally, ensure that the classification model can correctly identify and match different types of government demand features.

[0158] In practical applications, cross-modal government needs include at least: text-image fusion government needs, text-audio government needs, and image-audio government needs. For example, suppose a user submits a government report containing text descriptions, related pictures, and voice descriptions through a government platform, and hopes that the system can provide policies and service processes related to specific matters based on this information.

[0159] For the text-image fusion government demand feature, the text description submitted by the user and the relevant pictures can be fused together to generate the text-image fusion government demand feature. For example, the system can use text information to describe the details of the disease, and at the same time combine the content of the government report in the picture, combine the text description with the image information, and form a more comprehensive and accurate government demand description feature.

[0160] For the text and audio government affairs demand features, the user provides a text description and a voice description, and the text and audio government affairs demand features can be extracted. By correlating and analyzing the text description with the voice description, the system can better understand the user's medical condition and demand demands. For example, the voice description can be converted into text through speech-to-text technology, and then integrated with the text information provided by the user for analysis.

[0161] For image and audio government demand features, the system can extract image and audio government demand features when users upload pictures and voice descriptions of government reports. The system can combine the government report content in the image data and the condition description in the voice description to generate richer and more detailed demand features. For example, the system can use computer vision and speech processing technology to jointly analyze image and voice data to more fully understand user needs.

[0162] By comprehensively using these cross-modal government service demand characteristics, the system can understand users' needs more accurately, provide users with personalized and comprehensive government service recommendations, and improve the quality and efficiency of government services.

[0163] In the target matching layer, the cross-modal government demand features are decomposed and quickly learned through the meta-learning training model to obtain the matching target government processing process, including joint representation learning and modal weight learning. The specific steps are to input the cross-modal government demand features into the meta-learning model for process decomposition and rapid learning. Through joint representation learning, different types of demand features are combined to generate a comprehensive representation. Through modal weight learning, the weights of different modal features in the matching process are determined to ensure the effectiveness and importance of the features.

[0164] In practical applications, rapid learning and processing at least include: joint representation learning and modal weight learning. For example, suppose an intelligent government affairs processing system needs to process cross-modal government affairs requirements submitted by users. The system uses rapid learning and processing including joint representation learning and modal weight learning to achieve efficient demand analysis and processing. The following is an example: A user uploaded a government affairs requirement containing text descriptions, related pictures and voice recordings through the government affairs platform, hoping that the system can automatically identify and give corresponding government affairs processing suggestions. The system needs to quickly learn and process the user's cross-modal demand characteristics to match the appropriate government affairs processing process.

[0165] In joint representation learning, the text, image and voice data submitted by users are uniformly represented through joint representation learning technology to retain the association information between different modal data and help the system better understand user needs. For example, the system maps text, image and voice data to a shared representation space through a deep learning network, so that the association between different modal data can be effectively captured.

[0166] In modal weight learning, the importance weights of different modal data in cross-modal demand matching are learned through modal weight learning technology, so as to better utilize different modal data information based on joint representation learning. For example, the system can dynamically adjust the weights according to the contribution of different modal data to government demand matching, so as to more accurately match the appropriate government processing procedures.

[0167] In this way, the combined use of joint representation learning and modal weight learning technology can achieve rapid learning and accurate matching when processing users' cross-modal government affairs needs, provide personalized government affairs processing suggestions, and provide users with effective service support and decision-making references.

[0168] Through these levels of processing, the government affairs diversion model can effectively analyze user demand data and match the target government affairs processing flow, thereby improving the efficiency and accuracy of government services.

[0169] Further optionally, the meta-learning training model is constructed based on an N-order meta-learning optimizer, where N is the order of each model parameter in the meta-learning training model. t+1 It is expressed as the following mathematical expression:

[0170]

[0171] Among them, θ t is the t-th model parameter, t is the number of iterations of the meta-learning optimizer, x is the cross-modal government demand feature, is the second-order gradient vector, is the first-order gradient, is the Hessian matrix, is the inverse matrix of the Hessian matrix, l represents the loss function, and α is the learning rate.

[0172] In the above or following embodiments, the intelligent government customer service application system also includes a construction module, which is specifically configured to: collect government data involved in government processes to build or update a government data lake; and build an intelligent government customer service model and / or a government diversion model based on the government data lake.

[0173] The government affairs data at least includes government affairs question records, government affairs reply records, government affairs processing procedures and corresponding government affairs processing materials in multiple government affairs fields. In practical applications, the government affairs processing materials include text materials and / or image materials.

[0174] Assume that the government affairs field includes at least one of the following: public services, healthcare, and education. In the intelligent government affairs customer service application system, a construction module is included, which is responsible for collecting government affairs data involved in the government affairs process to build or update the government affairs data lake, and build an intelligent government affairs customer service model and / or a government affairs diversion model based on the government affairs data lake. Government affairs data includes government affairs question records, government affairs reply records, government affairs processing procedures, and corresponding government affairs processing materials in multiple government affairs fields, where government affairs processing materials include text materials and / or image materials, and the government affairs fields cover public services, healthcare, education, etc.

[0175] For example, suppose a government intelligent customer service platform is dedicated to providing government consulting services in the field of medical health. Specifically, crawler technology and data interfaces can be used to collect government question records, response records, processing procedures and related processing materials in the field of medical health, including medical information, disease diagnosis and treatment procedures, medical images, etc., to build a government data lake in the field of medical health. Furthermore, based on the government data lake in the field of medical health, the system uses natural language processing and machine learning technologies to build an intelligent government customer service model. The model can understand user questions, answer questions, and provide suggestions for related government processing procedures and materials, thereby realizing intelligent government consulting services. Next, the system also builds a government diversion model to automatically divert users' government needs to different processing procedures and processing materials according to the content of the user's questions and the type of needs. For example, for different types of medical problems, the system can divert user needs to corresponding medical information, disease diagnosis and treatment procedures, medical images and other resources to meet the user's personalized needs.

[0176] Through the intelligent government customer service application system constructed above, users can more conveniently obtain government information and services in the field of medical and health, and improve government processing efficiency and user experience.

[0177] As an optional embodiment, the government data lake includes at least: a government processing knowledge base and a government question and answer base. Figure 4 As shown, when the construction module collects government data involved in the government process to build or update the government data lake, it is specifically configured as follows:

[0178] 401. Obtain government text data related to government processes;

[0179] 402. Perform text processing and entity recognition from government text data to build a government knowledge graph;

[0180] 403. Build a government data lake based on the government knowledge graph;

[0181] 404. Store the pre-processed government affairs processing knowledge and question-and-answer data in the government affairs text data in corresponding storage areas in the government affairs data lake, and set access permissions for the storage areas.

[0182] Assuming that a government intelligent customer service system is aimed at the public service field, taking social security processing as an example, the government text data at least includes: policy documents, processing flow descriptions, service guides, and government Q&A texts. For example, government text data includes but is not limited to: social security policy documents, social security processing flow descriptions, social security service guides, social security related Q&A texts and other information. The government knowledge graph contains various government processing entities, entity attributes, and entity relationships. For example, the government knowledge graph contains various social security process links and corresponding authority ends, entity attributes, and entity relationships, such as insured persons, payment standards, and medical insurance benefits.

[0183] In the government affairs intelligent customer service system, each government affairs processing entity corresponds to a process link and / or a processing authority end in the government affairs processing process. At the same time, the access rights are associated with the processing authority of the government affairs processing entity corresponding to the storage area. In the government affairs intelligent customer service system, it is assumed that the government affairs intelligent customer service system is a city housing provident fund service, which involves the following government affairs processing entities: personal account inquiry, deposit details inquiry, and loan processing procedures. Personal account inquiry corresponds to the account information inquiry link in the provident fund processing process. Deposit details inquiry corresponds to the historical deposit record inquiry link in the provident fund processing process. The loan processing process corresponds to the loan application and approval link in the provident fund processing process.

[0184] It is understandable that the access rights are associated with the processing entity rights as follows: First, only provident fund contributors with personal accounts can query personal account information, so the storage area corresponding to the personal account query entity is only open to contributors. Second, the query of deposit details may require review by the provident fund center staff, so its access rights will be associated with the corresponding staff rights. Third, the loan processing process requires approval and authorization, and the access rights are associated with the rights of the approval personnel and relevant business departments. In this way, by associating the access rights with the processing rights of the government processing entity corresponding to the storage area, the government intelligent customer service system can ensure data security and reasonable access to information, while protecting privacy and providing efficient government service processes.

[0185] In the embodiment of the present application, through the synergy of the above modules, the intelligent government customer service system can automatically and intelligently understand user needs, quickly match appropriate government processing procedures, and provide users with personalized and efficient government services, assisting in improving the level of government services. In addition, the synergy of the above modules also greatly improves the matching accuracy of the target government processing procedures, ensures the accuracy of intelligent government customer service, avoids irrelevant answers or overly vague answers, and further improves the user's government customer service experience.

[0186] In another embodiment of the present application, an intelligent government customer service application method is also provided. Figure 5 As shown, the method includes:

[0187] 501. In response to a government affairs question request initiated by a user, a pending customer service task matching the government affairs question request is established; the pending customer service task is used to guide the user to further feedback his or her own government affairs needs;

[0188] 502. In the pending customer service task, call the intelligent government affairs customer service model, and use the intelligent government affairs customer service model to perform an interactive dialogue with the user to obtain user demand data corresponding to the pending customer service task;

[0189] 503. Adopting the government affairs diversion model, perform government affairs demand analysis on the user demand data to obtain a target government affairs processing flow matching the user demand data.

[0190] Further optionally, the intelligent government customer service model includes at least: a multimodal input layer, a multimodal feature fusion layer, a context-aware continuous dialogue layer, and a transfer learning layer;

[0191] The demand identification module, when using the intelligent government affairs customer service model to perform an interactive dialogue with a user to obtain user demand data corresponding to the pending customer service task, is specifically configured as follows:

[0192] The first interaction information between the user and the multimodal input layer is generated by adopting the interactive dialogue generation model matched with the pending customer service task; the multimodal interaction data is obtained from the second interaction information fed back by the user; the multimodal interaction data at least includes the dialogue text, dialogue image, and dialogue audio;

[0193] Extracting initial demand language features corresponding to different modal interaction data and associated interaction features between different modal interaction data from the multimodal interaction data through the multimodal feature fusion layer;

[0194] The context-aware continuous dialogue layer converts the initial demand language features and the associated interaction features extracted based on the second interaction information into a first interaction dialogue tracking feature sequence according to the context dialogue semantic relationship and the pre-input historical interaction dialogue feature sequence; performs semantic understanding processing on the first interaction dialogue tracking feature sequence, and generates third interaction information with the user based on the semantic understanding result;

[0195] Acquire the multimodal interaction data from fourth interaction information fed back by the user based on the third interaction information through the multimodal input layer;

[0196] The context-aware continuous dialogue layer converts the first interactive dialogue tracking feature sequence, the initial demand language features and the associated interactive features extracted based on the fourth interactive information into a second interactive dialogue tracking feature sequence according to the context dialogue semantic relationship and the previously input historical interactive dialogue feature sequence;

[0197] Through the transfer learning layer, a government dialogue demand recognition model is used to perform semantic understanding processing on the first interactive dialogue tracking feature sequence and the second interactive dialogue tracking feature sequence to obtain the user demand data;

[0198] Among them, the government affairs dialogue demand identification model is obtained through parameter preprocessing and fine-tuning based on various customer service tasks and corresponding historical government affairs dialogue data.

[0199] Further optionally, the intelligent government customer service model further includes: a model dynamic reinforcement layer;

[0200] The demand identification module, after using the intelligent government affairs customer service model to perform an interactive dialogue with the user to obtain the user demand data corresponding to the customer service task to be processed, is further specifically configured to:

[0201] Through the dynamic reinforcement layer of the model, the initial demand language features and the associated interaction features are processed for dialogue evaluation to obtain a dynamic dialogue evaluation result;

[0202] The interactive dialogue generation model is dynamically enhanced using the dynamic dialogue evaluation result to achieve optimization processing of the interactive dialogue generation model.

[0203] Further optionally, the intelligent government customer service model further includes: a model integration layer;

[0204] The demand identification module, before using the intelligent government customer service model to perform an interactive dialogue with the user to obtain the user demand data corresponding to the pending customer service task, is further specifically configured as follows:

[0205] Through the model integration layer, multimodal data enhancement processing is performed on historical government dialogue data corresponding to different types of customer service tasks to obtain a multimodal expansion set of government dialogues;

[0206] The multimodal expansion set of government dialogue is adopted to construct multiple interactive dialogue generation models for different types of customer service tasks respectively, and the multiple interactive dialogue generation models are integrated into a target interactive dialogue generation model suitable for different types of customer service tasks through model fusion processing.

[0207] Further optionally, the demand identification module performs multimodal data enhancement processing on the historical government dialogue data corresponding to different types of customer service tasks through the model integration layer to obtain a multimodal expansion set of government dialogues, which is specifically configured as follows:

[0208] For the text data in the historical government dialogue data, the keywords in the text data are replaced with synonyms or near-synonyms; and / or, the text data is converted into sentence structures; and / or, the dialogue sentences between different roles in the text data are expanded with dialogue bionics; and / or, the text data is filled with noise to obtain government dialogue text expansion data;

[0209] For the image data in the historical government dialogue data, performing at least one of random cropping, flipping, image rotation, scaling, pixel expansion, and color conversion processing on the image data to obtain government dialogue image expansion data;

[0210] For the audio data in the historical government dialogue data, at least one of the following processes is performed on the audio data: sound speed change, reverberation, noise reduction, enhancement, copying, transformation, cutting, and combination to obtain the government dialogue audio expansion data.

[0211] Further optionally, the demand identification module integrates multiple interactive dialogue generation models into a target interactive dialogue generation model suitable for different types of customer service tasks through model fusion processing, and is specifically configured as follows:

[0212] Obtaining respective dialogue generation results of multiple interactive dialogue generation models;

[0213] Based on the dialogue generation results and the pre-added noise results, voting is performed on multiple interactive dialogue generation models to obtain an interactive dialogue generation model to be selected;

[0214] The preset meta-model is used to perform fusion analysis on the dialogue generation results of the selected interactive dialogue generation models to obtain the target interactive dialogue generation model.

[0215] Further optionally, the government affairs diversion model at least includes: an input layer, a demand feature extraction layer, a government affairs demand matching layer, and a target matching layer;

[0216] The demand analysis module adopts the government affairs diversion model to perform government affairs demand analysis on the user demand data, and when obtaining the target government affairs processing flow matching the user demand data, is specifically configured as follows:

[0217] Extracting corresponding government affairs demand information from the user demand data through the input layer, wherein the government affairs demand information at least includes: government affairs processing flow information, government affairs item type, and related government affairs attribute information;

[0218] The government affairs demand information is converted into corresponding multimodal government affairs demand features through the demand feature extraction layer; the multimodal government affairs demand features include: at least one of text government affairs demand features, image government affairs demand features, and audio government affairs demand features;

[0219] Through the government demand matching layer, the multimodal government demand features are input into the demand classification model for government demand classification processing to obtain the user's cross-modal government demand features; the cross-modal government demand features at least include: text-image fusion government demand features, text-audio government demand features, and image-audio government demand features;

[0220] Through the target matching layer, a meta-learning training model is adopted to perform process decomposition and rapid learning processing on the cross-modal government affairs demand characteristics to obtain the matching target government affairs processing process; the rapid learning processing at least includes: joint representation learning and modal weight learning.

[0221] Further optionally, the meta-learning training model is constructed based on an N-order meta-learning optimizer, where N is the order of each model parameter in the meta-learning training model;

[0222] Among them, the t+1th model parameter θ in the meta-learning training model t+1 It is expressed as the following mathematical expression:

[0223]

[0224] Among them, θt is the t-th model parameter, t is the number of iterations of the meta-learning optimizer, x is the cross-modal government demand feature, is the second-order gradient vector, is the first-order gradient parameter, is the Hessian matrix, is the inverse matrix of the Hessian matrix, l represents the loss function, and α is the learning rate.

[0225] Further optionally, the system further includes a building module, which is specifically configured to:

[0226] Collect government data involved in government processes to build or update the government data lake;

[0227] Constructing the intelligent government affairs customer service model and / or the government affairs diversion model based on the government affairs data lake;

[0228] Among them, the government data at least includes government question records, government response records, government processing procedures and corresponding government processing materials in multiple government fields; the government processing materials include text materials and / or image materials; the government fields include at least one of the following: public services, medical health, and education.

[0229] Further optionally, the government data lake includes at least: a government processing knowledge base, a government question and answer base;

[0230] When collecting government data involved in government processes to build or update the government data lake, the construction module is specifically configured as follows:

[0231] Acquire government affairs text data related to government affairs processes, wherein the government affairs text data at least includes: policy documents, processing flow descriptions, service guidelines, and government affairs Q&A texts;

[0232] Performing text processing and entity recognition on the government text data to construct a government knowledge graph; the government knowledge graph includes various government processing entities, entity attributes, and entity relationships; each government processing entity corresponds to a process link and / or a processing authority end in the government processing process;

[0233] Constructing the government data lake based on the government knowledge graph;

[0234] The preprocessed government affairs processing knowledge and question-and-answer data in the government affairs text data are respectively stored in corresponding storage areas in the government affairs data lake, and access rights to the storage areas are set; the access rights are associated with the processing rights of the government affairs processing entity corresponding to the storage area.

[0235] In the embodiment of the present application, the above method can automatically and intelligently understand user needs, quickly match appropriate government affairs processing procedures, and provide users with personalized and efficient government affairs services, thereby assisting in improving the level of government affairs services. In addition, the above method also greatly improves the matching accuracy of the target government affairs processing procedures, ensures the accuracy of intelligent government affairs customer service, avoids irrelevant answers or overly vague answers, and further improves the user's government affairs customer service experience.

[0236] In another embodiment of the present application, an electronic device is provided, comprising: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0237] Memory, used to store computer programs;

[0238] The processor is used to implement the intelligent government affairs customer service application system described in the method embodiment when executing the program stored in the memory.

[0239] The communication bus 1140 mentioned in the above electronic device may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus 1140 may be divided into an address bus, a data bus, a control bus, etc.

[0240] For ease of representation, Figure 6 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0241] The communication interface 1120 is used for communication between the above electronic device and other devices.

[0242] The memory 1130 may include a random access memory (RAM) or a non-volatile memory (non-volatile memory), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0243] The above-mentioned processor 1110 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0244] Accordingly, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed, can implement each step that can be performed by an electronic device in the above method embodiment.

Claims

1. An intelligent government customer service application system, characterized in that: The system comprises: The task establishment module is configured to respond to the government affairs question request initiated by the user and establish a pending customer service task matching the government affairs question request; the pending customer service task is used to guide the user to further feedback his own government affairs needs; The demand identification module is configured to call the intelligent government customer service model in the pending customer service task, and use the intelligent government customer service model to perform an interactive dialogue with the user to obtain user demand data corresponding to the pending customer service task; A demand analysis module is configured to adopt a government affairs diversion model to perform government affairs demand analysis on the user demand data to obtain a target government affairs processing flow matching the user demand data; The intelligent government customer service model includes at least the following four layers: a multimodal input layer, a multimodal feature fusion layer, a context-aware continuous dialogue layer, and a transfer learning layer; The demand identification module, when using the intelligent government affairs customer service model to perform an interactive dialogue with a user to obtain user demand data corresponding to the pending customer service task, is specifically configured as follows: The first interaction information between the user and the multimodal input layer is generated by adopting the interactive dialogue generation model matched with the pending customer service task; the multimodal interaction data is obtained from the second interaction information fed back by the user; the multimodal interaction data includes at least two of the following data: dialogue text, dialogue image, and dialogue audio; The multimodal feature fusion layer extracts initial demand language features corresponding to different modal interaction data and associated interaction features between different modal interaction data from the multimodal interaction data; the initial demand language features are user demand information contained in the multimodal interaction data; the associated interaction features are cross-user information associated with each other between the multimodal interaction data; The context-aware continuous dialogue layer converts the initial demand language features and the associated interaction features extracted based on the second interaction information into a first interaction dialogue tracking feature sequence according to the context dialogue semantic relationship and the pre-input historical interaction dialogue feature sequence; performs semantic understanding processing on the first interaction dialogue tracking feature sequence, and generates third interaction information with the user based on the semantic understanding result; Acquire the multimodal interaction data from fourth interaction information fed back by the user based on the third interaction information through the multimodal input layer; The context-aware continuous dialogue layer converts the first interactive dialogue tracking feature sequence, the initial demand language features and the associated interactive features extracted based on the fourth interactive information into a second interactive dialogue tracking feature sequence according to the context dialogue semantic relationship and the previously input historical interactive dialogue feature sequence; Through the transfer learning layer, a government dialogue demand recognition model is used to perform semantic understanding processing on the first interactive dialogue tracking feature sequence and the second interactive dialogue tracking feature sequence to obtain the user demand data; Among them, the government affairs dialogue demand identification model is obtained through parameter preprocessing and fine-tuning based on various customer service tasks and corresponding historical government affairs dialogue data.

2. The intelligent government affairs customer service application system according to claim 1 is characterized in that: The intelligent government affairs customer service model also includes: a model dynamic reinforcement layer; The demand identification module, after using the intelligent government affairs customer service model to perform an interactive dialogue with the user to obtain the user demand data corresponding to the customer service task to be processed, is further specifically configured to: Through the dynamic reinforcement layer of the model, the initial demand language features and the associated interaction features are processed for dialogue evaluation to obtain a dynamic dialogue evaluation result; The interactive dialogue generation model is dynamically enhanced using the dynamic dialogue evaluation result to achieve optimization processing of the interactive dialogue generation model.

3. The intelligent government affairs customer service application system according to claim 2 is characterized in that: The intelligent government customer service model also includes: a model integration layer; The demand identification module, before using the intelligent government customer service model to perform an interactive dialogue with the user to obtain the user demand data corresponding to the pending customer service task, is further specifically configured as follows: Through the model integration layer, multimodal data enhancement processing is performed on historical government dialogue data corresponding to different types of customer service tasks to obtain a multimodal expansion set of government dialogues; The multimodal expansion set of government dialogue is adopted to construct multiple interactive dialogue generation models for different types of customer service tasks respectively, and the multiple interactive dialogue generation models are integrated into a target interactive dialogue generation model suitable for different types of customer service tasks through model fusion processing.

4. The intelligent government affairs customer service application system according to claim 3 is characterized in that: The demand identification module performs multimodal data enhancement processing on the historical government dialogue data corresponding to different types of customer service tasks through the model integration layer to obtain a multimodal expansion set of government dialogues, and is specifically configured as follows: For the text data in the historical government dialogue data, the keywords in the text data are replaced with synonyms or near-synonyms; and / or the text data is converted into sentence structures; and / or the dialogue sentences between different roles in the text data are expanded by dialogue bionics; and / or, performing noise filling on the text data to obtain government dialogue text expansion data; For the image data in the historical government dialogue data, performing at least one of random cropping, flipping, image rotation, scaling, pixel expansion, and color conversion processing on the image data to obtain government dialogue image expansion data; For the audio data in the historical government dialogue data, at least one of the following processes is performed on the audio data: sound speed change, reverberation, noise reduction, enhancement, copying, transformation, cutting, and combination to obtain the government dialogue audio expansion data.

5. The intelligent government affairs customer service application system according to claim 3 is characterized in that: The demand identification module integrates multiple interactive dialogue generation models into a target interactive dialogue generation model suitable for different types of customer service tasks through model fusion processing, and is specifically configured as follows: Obtaining respective dialogue generation results of multiple interactive dialogue generation models; Based on the dialogue generation results and the pre-added noise results, voting is performed on multiple interactive dialogue generation models to obtain an interactive dialogue generation model to be selected; The preset meta-model is used to perform fusion analysis on the dialogue generation results of the selected interactive dialogue generation models to obtain the target interactive dialogue generation model.

6. The intelligent government affairs customer service application system according to claim 1 is characterized in that: The government affairs diversion model includes at least the following four layers: input layer, demand feature extraction layer, government affairs demand matching layer, and target matching layer; The demand analysis module adopts the government affairs diversion model to perform government affairs demand analysis on the user demand data, and when obtaining the target government affairs processing flow matching the user demand data, is specifically configured as follows: Through the input layer, corresponding government affairs demand information is extracted from the user demand data, and the government affairs demand information includes at least the following three information: government affairs processing flow information, government affairs item type, and related government affairs attribute information; The government affairs demand information is converted into corresponding multimodal government affairs demand features through the demand feature extraction layer; the multimodal government affairs demand features include: at least two of text government affairs demand features, image government affairs demand features, and audio government affairs demand features; Through the government demand matching layer, the multimodal government demand features are input into the demand classification model for government demand classification processing to obtain the user's cross-modal government demand features; the cross-modal government demand at least includes: at least one of text-image fusion government demand features, text-audio government demand features, and image-audio government demand features; Through the target matching layer, a meta-learning training model is adopted to perform process decomposition and rapid learning processing on the cross-modal government affairs demand characteristics to obtain the matching target government affairs processing process; the rapid learning processing at least includes: joint representation learning and modal weight learning.

7. The intelligent government affairs customer service application system according to claim 6 is characterized in that: The meta-learning training model is constructed based on an N-order meta-learning optimizer, where N is the order of each model parameter in the meta-learning training model; Among them, the t+1th model parameter in the meta-learning training model It is expressed as the following mathematical expression: ， in, is the t-th model parameter, is the number of iterations of the meta-learning optimizer, For the cross-modal government affairs demand characteristics, is the second-order gradient vector, is the first-order gradient parameter, is the Hessian matrix, is the inverse of the Hessian matrix, represents the loss function, is the learning rate.

8. The intelligent government affairs customer service application system according to claim 1 is characterized in that: The system also includes a building block, which is specifically configured as follows: Collect government data involved in government processes to build or update the government data lake; Constructing the intelligent government affairs customer service model and / or the government affairs diversion model based on the government affairs data lake; Among them, the government data at least includes government question records, government response records, government processing procedures and corresponding government processing materials in multiple government fields; the government processing materials include text materials and / or image materials; the government fields include at least one of the following: public services, medical health, and education.

9. The intelligent government affairs customer service application system according to claim 8 is characterized in that: The government data lake at least includes: a government processing knowledge base and a government question and answer base; When collecting government data involved in government processes to build or update the government data lake, the construction module is specifically configured as follows: Acquire government affairs text data related to government affairs processes, wherein the government affairs text data includes at least the following four data: policy documents, processing flow descriptions, service guidelines, and government affairs question and answer texts; Performing text processing and entity recognition on the government text data to construct a government knowledge graph; the government knowledge graph includes various government processing entities, entity attributes and entity relationships; each government processing entity corresponds to a process link and / or a processing authority end in the government processing process; Constructing the government data lake based on the government knowledge graph; The preprocessed government affairs processing knowledge and question-and-answer data in the government affairs text data are respectively stored in corresponding storage areas in the government affairs data lake, and access rights to the storage areas are set; the access rights are associated with the processing rights of the government affairs processing entity corresponding to the storage area.

Images