Big model-based file handling process knowledge base construction and multi-round memory question and answer method

Abstract

The invention discloses a large model-based document handling flow knowledge base construction and multi-round memory question and answer method, which comprises the following steps of: respectively acquiring region information associated with document handling items and flow key information corresponding to various document handling items to form a target document; performing knowledge segmentation on data in the target document, performing vectorization processing on segmented contents by adopting an open source semantic vector and a reordering model, storing a result into a local vector database to construct a file handling flow knowledge base, and configuring a target function; an original retrieval question input by a user is processed in combination with the constructed multi-round memory question and answer intelligent model and the document handling flow knowledge base, reply content is output in a point division mode, and the reply content is automatically associated with a direct link corresponding to a document handling item and a document handling flow chart; and updating the file handling process knowledge base based on question feedback and executing a new round of memory questions and answers.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a method for constructing a case handling process knowledge base based on a large model and multi-round memory question and answer. Background Art

[0002] With the deepening of the "Internet + Government Affairs Service", government affairs services are accelerating the transformation from offline to online, and the way for the public to handle affairs is gradually changing from "running on-site" to "handling online". However, the popularization of online services has not completely replaced the demand for offline handling, and the online and offline collaborative service ability has become a key link in improving the efficiency of government affairs services.

[0003] In the scenario of online retrieval of case handling processes, the public's consulting needs for links such as process guidelines, material preparation, and location guidelines show characteristics such as high frequency, multi-round, and context dependence. The current main consulting channels include four types: local government affairs service websites, local treasure (through) WeChat official accounts, artificial customer service, and government affairs Q&A assistants. The first two use traditional government affairs retrieval mechanisms and lack the ability of dialogue interaction, often making the public face a relatively high cognitive threshold during the search process and making it difficult to quickly locate the required matters and relevant guidelines. Traditional artificial customer service is limited by working hours and concurrent processing capabilities, and is prone to queuing congestion and response delays during peak consulting periods. Existing government affairs Q&A assistants generally lack effective memory and semantic association with regional attribution, matter categories, and dialogue history, resulting in users having to repeat inputting background information in multi-round questions, affecting the interaction efficiency and overall service experience. Summary of the Invention

[0004] In view of at least one defect or improvement requirement of the prior art, the present invention provides a method for constructing a case handling process knowledge base based on a large model and multi-round memory question and answer. By constructing a standardized case handling process knowledge base, the problem of inaccurate association and extraction of the administrative region and business type to which a case handling matter belongs in the retrieval problem is solved, and the intelligent retrieval function of case handling knowledge with multi-round memory ability is realized.

[0005] To achieve the above objectives, according to the first aspect of the present invention, a method for constructing a knowledge base for handling procedures based on a large model and for multi-round memory-based question answering is provided. This method includes: acquiring regional information associated with handling items and key process information corresponding to various handling items, forming a target document; segmenting the data in the target document into knowledge segments, using an open-source semantic vector and reordering model to vectorize the segmented content, storing the results in a local vector database to construct a knowledge base for handling procedures and configuring target functions; processing the user's original search question using the constructed multi-round memory-based question answering intelligent model and the handling procedure knowledge base, and outputting response content in a point-by-point manner, wherein the response content automatically associates the direct links and handling procedure flowcharts corresponding to the handling items; updating the handling procedure knowledge base based on question feedback and executing a new round of memory-based question answering.

[0006] In an exemplary embodiment, the step of performing knowledge segmentation on the data in the target document includes: using the first row of the header row in the data of the target document as the key and each row of data as the corresponding value; and converting the data in the target document into a dictionary list in the form of key-value pairs.

[0007] In an exemplary embodiment, the process of processing the original retrieval question input by the user by combining the constructed multi-turn memory question-answering intelligent model and the handling process knowledge base includes: using the multi-turn memory question-answering intelligent model in combination with custom model instructions to analyze the original retrieval question, extracting regional information and matter information from the original retrieval question to obtain extraction results; and generating target output based on the extraction results.

[0008] In an exemplary embodiment, generating the target output based on the extraction result includes: if the region information and the event information are successfully identified in the extraction result, outputting them according to a preset format; if the region information or the event information is not successfully identified in the extraction result, assigning the corresponding field in the target output to an empty value.

[0009] In an exemplary embodiment, the multi-turn memory question-answering intelligent model and the handling process knowledge base are combined to process the original retrieval question input by the user, and the point-by-point output response content includes: creating and maintaining the current matter information session variable and the current region information session variable; when non-empty region information and matter information are successfully extracted from the original retrieval question, the current matter information session variable and the current region information session variable are updated, otherwise they remain unchanged.

[0010] In one exemplary embodiment, after updating the current item information session variable and the current region information session variable, the method further includes: using the updated current item information session variable and the current region information session variable, as well as the original search question, to reconstruct the original search question according to a preset template.

[0011] In an exemplary embodiment, after reconstructing the original retrieval question according to a preset template, the method further includes: performing a document processing knowledge base retrieval based on the reconstructed original retrieval question; calculating semantic similarity using a semantic vector model to obtain preliminary retrieval results; optimizing and sorting similar segments in the preliminary retrieval results using a re-ranking model to obtain retrieval results; and, based on the retrieval results and a prompt word template, calling a large-scale reasoning language model to perform reasoning analysis and structured generation, and outputting response content in a point-by-point manner.

[0012] According to a second aspect of the present invention, a system for constructing a knowledge base for handling procedures based on a large model and for multi-round memory question answering is also provided, comprising: an acquisition unit, used to acquire regional information associated with handling matters and key process information corresponding to various handling matters, forming a target document; a segmentation unit, used to segment the data in the target document into knowledge segments, use an open-source semantic vector and reordering model to vectorize the segmented content, and store the results in a local vector database to construct a knowledge base for handling procedures and configure target functions; an output unit, used to process the original retrieval question input by the user in combination with the constructed multi-round memory question answering intelligent model and the knowledge base for handling procedures, and output the response content in a point-by-point manner, wherein the response content automatically associates the direct links and handling process diagrams corresponding to the handling matters; and an update unit, used to update the knowledge base for handling procedures based on the question feedback and execute a new round of memory question answering.

[0013] According to a third aspect of the invention, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer program, which is configured to execute the above-described method for constructing a knowledge base for handling procedures based on a large model and for multi-round memory question answering at runtime.

[0014] According to a fourth aspect of the present invention, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the above-described method for constructing a knowledge base for a large-scale application process and for multi-round memory-based question answering via the computer program.

[0015] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: (1) This invention provides a method for constructing a knowledge base for handling procedures based on a large model and a multi-round memory question-and-answer method. By organizing regional relationship information and handling procedure information, a knowledge base for handling procedures is constructed that associates handling procedures with detailed regional descriptions. The handling procedures are associated with their respective regions, and official links and handling flowcharts are embedded in the consultation responses to achieve authoritative jumps in service guidance and ensure the accuracy and traceability of consultation services.

[0016] (2) By utilizing a non-inference model combined with parameter extraction prompting technology, accurate extraction of regional and item information from multiple user queries is achieved. Based on this, a session variable mechanism is used to dynamically update and iterate this key information, thereby supporting automatic completion of regional and item-related content in multi-turn dialogues. In multi-turn dialogues, a contextual association mechanism between administrative regions and government service items, as well as an automatic completion mechanism for key information, are implemented, effectively solving the problem of users having to repeatedly input regions and service items, and improving interaction efficiency and service experience. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating an optional method for constructing a knowledge base for a case handling process based on a large model and for multi-round memory question answering, provided for an embodiment of this application; Figure 2 A schematic diagram illustrating an optional process for constructing a knowledge base for government affairs processing procedures, provided as an embodiment of this application; Figure 3 A schematic diagram of an optional government affairs processing knowledge base multi-round memory question-and-answer process provided for an embodiment of this application; Figure 4 This is a schematic diagram of an optional electronic device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0020] The terms "first," "second," "third," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0021] According to one aspect of the embodiments of this application, a method for constructing a knowledge base for handling procedures based on a large model and a multi-round memory question-and-answer method is provided. The following is in conjunction with... Figure 1 This application describes the method for constructing a knowledge base for handling procedures based on a large model and for multi-round memory-based question answering, as provided in the embodiments of this application.

[0022] Figure 1 This is a flowchart illustrating an optional method for constructing a knowledge base for a case handling process based on a large model and for multi-round memory-based question answering, as provided in an embodiment of this application. Figure 1 As shown, the process of this method may include the following steps: S102, respectively obtain the regional information associated with the handling items and the key process information corresponding to each type of handling item, and form the target document; S104, the data in the target document is segmented into knowledge segments, and the segmented content is vectorized using an open-source semantic vector and reordering model. The results are stored in a local vector database to build a document processing knowledge base and the target function is configured. S106, The original search question input by the user is processed by combining the constructed multi-round memory question-and-answer intelligent model and the handling process knowledge base, and the reply content is output in a point-by-point manner. The reply content is automatically associated with the direct link and handling process diagram corresponding to the handling item. S108, update the knowledge base of the handling process based on the feedback and execute a new round of memory questions and answers.

[0023] The embodiments of this application provide a method for constructing a knowledge base for handling procedures based on a large model and for multi-round memory question answering. This method can be applied to scenarios of 24 / 7 uninterrupted, efficient, and accurate government affairs handling procedures consultation services. For example, in the scenario of online retrieval of handling procedures, the public's consultation needs for procedures guidance, material preparation, location guidance, and other aspects are characterized by high frequency, multiple rounds, and context dependence.

[0024] This application accurately understands the procedural questions raised by the public, such as social security payment procedures and housing provident fund withdrawal procedures. It rigorously relies on existing procedural knowledge for information retrieval, resulting in a structured, point-by-point response. The response will include the corresponding official links and procedural flowcharts to ensure that inquiries are based on evidence, achieving efficient and accurate responses to government service inquiries.

[0025] Optionally, combined Figure 2 and Figure 3 As shown, the process of this method may include the following steps: Step 1: Acquire knowledge of the application process.

[0026] Two main categories of data were obtained through local government service websites, the Local Treasure (Tong) WeChat official account, and official data interfaces. The first category is regional information associated with the service items, specifically the regional relationships at the city, district / county, sub-district, and residents' committee levels, compiled into a target document (CSV file). The first line of the document is a fixed header row, and each subsequent line records a specific regional relationship. The second category is key process information corresponding to various service items, compiled into a CSV file with columns including "Item Name - Administrative Division of Item - Official Link to Item - Accepting Agency - Type of Service Item - Completion Time - Service Hall - Processing Time - Processing Location - Service Window - Office Telephone - Transportation Guide - Application Conditions - Application Materials - Permit Fees - Processing Flowchart." Each line in the document corresponds to a specific government service item, and the "Administrative Division of Item" column must include the corresponding superior region information to avoid ambiguity in region names as much as possible.

[0027] Step 2: Knowledge segmentation, vectorization, and storage.

[0028] The data in the CSV document is converted into a dictionary list in key-value pair format. Specifically, the first row (header row) serves as the key (field name), and subsequent rows serve as the corresponding values ​​(field content). Each row of data (each government service item) is combined with the first row header to form an independent dictionary, thus constructing a list covering all service items. In the subsequent knowledge segmentation process, each dictionary item in the list is divided into a separate text block, considered as a segmented knowledge piece.

[0029] Then, knowledge vectorization and storage are carried out. The open-source semantic vector and reordering model is used to vectorize the segmented content and store the results in a local vector database to build a case handling process knowledge base. The target functions of the case handling process knowledge base are configured, namely, add, delete, modify and query functions.

[0030] Step 3: Building Multi-Round Consulting Capabilities A multi-turn memory question-and-answer intelligent assistant is constructed. The assistant's multi-turn memory capability consists of five key components: voice or text input of the original question, extraction of key information of the original question, construction and updating of conversation variables, question rewriting, and retrieval of knowledge base information.

[0031] The input received by the intelligent assistant is the original search question (question_origin) posed by the user, which can be in the form of voice or text. Voice input will execute the corresponding speech-to-text service.

[0032] The original retrieval question, `question_origin`, is analyzed. Using an open-source general-purpose non-inference large language model combined with custom model instructions, key information is extracted from `question_origin`, including the user's region (district) and the event information (event). Based on the extraction results, the following output is generated: If the region and event information are successfully identified, the large model will output district='the region' and event='the event to be handled' in a formatted manner.

[0033] If a certain type of information cannot be identified, the corresponding field will be left empty, for example, district='', event='processing item'. Specifically, if the first query in the current session lacks regional information, district='all regions' will be set by default. In this case, the intelligent assistant will perform a process knowledge search covering all regions based on the item information and reply in the form of "region + official link to item" sorted by similarity, prompting the user in the first line of the reply: "Please select the following relevant regions, or specify a specific region and search again."

[0034] Create and maintain two session variables: current event information (event_now) and current district information (district_now). Update these two variables when a non-empty district and event are successfully extracted from the original question; otherwise, leave them unchanged.

[0035] Using the updated event_now and district_now, along with the original search question_origin, reconstruct the original search request according to the template "Event location: {district_now}, Event business name: {event_now}, Original search question: {question_origin}".

[0036] The rewritten question is searched using a knowledge base of the handling process. A semantic vector model is used to calculate semantic similarity, and a re-ranking model is used to optimize the ranking of similar segments in the initial search results. The two most relevant segments are returned with a similarity threshold of no less than 0.7. Based on the search results, and using prompt word templates, an open-source reasoning language model is invoked for reasoning analysis and structured generation. The response content is output in a point-by-point format, and the official links and handling process flowcharts corresponding to the matter are automatically linked.

[0037] Step 4: Problem Feedback and Knowledge Updates For questions and answers raised by users, as well as knowledge information that has changed due to timeliness, business personnel must review and confirm them before using the add, delete, modify, and query functions in the knowledge base configuration interface to quickly search for and update relevant knowledge, ensuring the accuracy and timeliness of the knowledge content.

[0038] Through the above steps, target documents are formed by acquiring the regional information associated with each service item and the key process information corresponding to each type of service item. The data in the target documents is then segmented into knowledge segments, and the segmented content is vectorized using an open-source semantic vector and reordering model. The results are stored in a local vector database to construct a service item process knowledge base and configure the target function. The original search question input by the user is processed using the constructed multi-turn memory question-and-answer intelligent model and the service item process knowledge base, and the response content is output in a point-by-point manner. The response content automatically associates the direct links and service item process diagrams corresponding to the service item. Based on the question feedback, the service item process knowledge base is updated, and a new round of memory question-and-answer is executed. This standardized service item process knowledge base solves the problem of inaccurately associating and extracting the administrative region and business type of the service item in the search question, realizing an intelligent retrieval function for service item knowledge with multi-turn memory capabilities.

[0039] In an exemplary embodiment, the step of performing knowledge segmentation on the data in the target document includes: S11, the first row of the header in the target document is used as the key, and each row of data is used as the corresponding value; S12 converts the data in the target document into a dictionary list in key-value pair form.

[0040] Optionally, two types of data can be obtained through the Hubei Government Service Network: regional information related to the application items and key process information corresponding to various application items. The regional information data is in CSV format, with the first line of the document as the title row, titled "Regional Association Information". Each subsequent line records a description of the regional association relationship at each level, from district / county to sub-district office to residents' committee. Example data is shown in Table 1.

[0041] Table 1. Schematic diagram of regional information associated with each service item.

[0042] The key information of the application process is in CSV format. The first line of the document is the title line, and each of the remaining lines corresponds to a specific government application item. Example data is shown in Table 2.

[0043] Table 2. Schematic diagram of key process information for various types of application items.

[0044] In other words, the header row of the CSV document containing regional information and key process information is treated as the key, and each remaining row is treated as the value, forming a dictionary in the form of key-value pairs {"key":"value"}. During knowledge segmentation, each dictionary entry is read in one pass, and each dictionary entry, i.e., each processing step, is treated as a text block. Taking the second row of data in the key process information document as an example, the resulting dictionary format, i.e., a text block, is shown below: { Item Name: Change of Disability Category / Level "Administrative Division Where Matter Permits": "Change of Disability Category / Level", Official link for this matter: http: / / zwfw.hubei.gov.cn / ... "Accepting Agency": "Wuhan East Lake High-tech Development Zone Social Affairs Bureau" "Type of application": "...", "Completion Time": "...", Service Hall: "...", Processing time: "...", Processing Location: "...", "Service Window": "...", Office phone number: "...", "Traffic Guide": "...", Application requirements: ... Application materials: "...", "License Fee": "...", "Processing Flowchart": "http: / / zwfw.hubei.gov.cn / ..." } Furthermore, through the knowledge vectorization and storage module, the bge-m3 open-source semantic vector model is used to vectorize each text block after the above processing. The vectorized content is stored in a local PostgreSQL vector database. In addition, the knowledge base is configured with fuzzy search and text block addition functions, and each text block after segmentation is configured with deletion and modification functions.

[0045] In an exemplary embodiment, the process of processing the original retrieval question input by the user by combining the constructed multi-turn memory question-answering intelligent model and the handling process knowledge base includes: S21, using a multi-round memory question-answering intelligent model combined with custom model instructions to analyze the original retrieval question, extracting regional information and item information from the original retrieval question to obtain the extraction result; S22, Generate the target output based on the extraction results.

[0046] In this embodiment, a session is created in the intelligent assistant, and the following questions are asked sequentially in this session: Question 1: Where can I apply for a change of disability type in the Guandong Subdistrict Office of the East Lake High-tech Development Zone? Question 2: What materials are required? Question 3: What about the East Lake High-tech Development Zone? Extract key information from the original question, specifically for query question 1: The query question 1 is evaluated, and the qwen2.5-7B model is used in conjunction with the following extraction instructions for extraction. Specifically: Determine whether query question 1 is the first query question; if the result is yes, the qwen2.5-7B model should be used to extract the district as Guandong Street, East Lake High-tech Development Zone, and the event as a change in disability type. The example parameter extraction instructions are as follows: As a parameter intelligent extraction assistant, your responsibility is to accurately extract the administrative region where the user is located and the government affairs matter inquired about from the user's original question {{#question_origin#}}, and assign this information to the district and event parameters respectively.

[0047] The specific rules are as follows: When a user's question clearly includes a specific administrative region and a specific government matter, accurately identify and extract the relevant information, and assign it to district and event respectively. If the user's question contains general query terms such as "how to do it", "where to process it", "when to process it", "what materials are needed", "application conditions", "processing procedure", etc., but does not specifically mention a specific administrative region or government matter, then both district and event should be set to null values ​​to indicate that no specific region or matter information could be obtained from the question.

[0048] In one exemplary embodiment, generating the target output based on the extraction result includes: S31, if the region information and event information are successfully identified in the extraction results, output them according to the preset format; S32, if the region information or event information is not successfully identified in the extraction results, the corresponding field in the target output is assigned an empty value.

[0049] In this embodiment, regarding the above query question 2: Query question 2 is judged, and the qwen2.5-7B model is used in conjunction with the above extraction instructions for extraction. Specifically: it is determined whether query question 2 is the first query question; if the judgment result is no, the qwen2.5-7B model is used to extract and obtain district and event values ​​that are both null.

[0050] Regarding query question 3: Query question 3 is evaluated, and the qwen2.5-7B model is used in conjunction with the extraction instructions to extract relevant information. Specifically: it is determined whether query question 3 is the first query question; if the result is no, the qwen2.5-7B model is used to extract the district as East Lake High-tech Development Zone and the event as a null value.

[0051] In an exemplary embodiment, the multi-turn memory-based question-answering intelligent model and the processing workflow knowledge base are combined to process the original retrieval question input by the user, and the point-by-point output of the response content includes: S41, Create and maintain session variables for current event information and current region information; S42, when non-empty region information and item information are successfully extracted from the original retrieval question, update the current item information session variable and the current region information session variable; otherwise, leave them unchanged.

[0052] Furthermore, the session variables are constructed and updated. For query question 1, the question has descriptions of region and event information, that is, the district and event values ​​extracted and returned by qwen2.5-7B are not empty. Therefore, the session variable district_now should be assigned the value of district and the session variable event_now should be assigned the value of event.

[0053] For subsequent query question 2, the question does not contain descriptions of region and event information, that is, the district and event values ​​extracted and returned by qwen2.5-7B are empty. Therefore, the session variables district_now and event_now are kept as the district and event values ​​obtained from query question 1.

[0054] For the subsequent query question 3, the question does not contain event information description, that is, the district returned by qwen2.5-7B is East Lake High-tech Development Zone and the event value is empty. Therefore, the session variable district_now is assigned to East Lake High-tech Development Zone, and the session variable event_now is kept to be the event value obtained from query question 1.

[0055] In one exemplary embodiment, after updating the current event information session variable and the current region information session variable, the method further includes: S51. Using the updated current item information session variables and current area information session variables, as well as the original search question, reconstruct the original search question according to the preset template.

[0056] Specifically, rewrite the original query question by referring to the template "Area where the item is located: {district_now}, Business name of the item: {event_now}, Original search question: {question_origin}".

[0057] For example, the rewritten query question 1: Location of the matter: Guandong Street, East Lake High-tech Development Zone; Name of the matter: Change of disability type; Original search question: Where can I apply for a change of disability type at the Guandong Street Office of East Lake High-tech Development Zone? Rewritten query question 2: Location of the matter: Guandong Street, East Lake High-tech Development Zone; Name of the matter: Change of disability type; Original search question: What materials are needed? Rewritten query question 3: Location of the matter: East Lake High-tech Development Zone; Name of the matter: Change of disability type; Original search question: What about East Lake High-tech Development Zone? This embodiment uses a session variable mechanism to dynamically update and iterate this key information, thereby supporting the automatic completion of area and item-related content in multi-turn dialogues.

[0058] In one exemplary embodiment, after reconstructing the original search question according to a preset template, the method further includes: S61, based on the reconstructed original retrieval question, a knowledge base retrieval of the handling process is performed, and a semantic vector model is used to calculate semantic similarity to obtain preliminary retrieval results; S62, Combine the re-ranking model to optimize and rank similar segments in the preliminary search results to obtain the search results; S63, based on the search results and the prompt word template, call the reasoning language model to perform reasoning analysis and structured generation, and output the reply content in a point-by-point manner.

[0059] For example, a rewritten query question is used for semantic retrieval of the application process knowledge base. The semantic vector model is bge-m3, the re-ranking model is bge-reranker-v2-m3, and the inference language model is deepseek-r1:32B. During the semantic retrieval process, the Top K number of returned fragments is set to 2, and the semantic vector matching threshold is set to 0.7. The inference hint words are shown below: You are an intelligent consultation assistant for government service procedures, focusing on providing users with precise guidance on handling matters within the administrative region. Please base your application on the encapsulated... <context> {{#context#}}< / context> The knowledge content within the tags, combined with user questions, should be answered in a logically clear and well-organized point-by-point manner according to the following requirements: 1. Clearly identify and confirm the administrative region where the user is located and the matter to be handled; 2. Respond strictly based on official information regarding the relevant matters in the knowledge base; 3. If there are official links and flowcharts for handling matters, they must be presented in their original, complete form; 4. Adding speculative or extended content outside the knowledge base is prohibited; 5. The language used in the response should be concise, clear, and well-structured to facilitate user understanding and operation.

[0060] For knowledge that is questionable or outdated, business personnel can receive a notification in the background after providing feedback in the current session or directly through the smart assistant. After review and confirmation, the knowledge can be quickly adjusted through the "add, delete, modify, and query" capabilities configured in the knowledge base.

[0061] According to another aspect of the embodiments of this application, a joint data virtualization system for implementing the above-described large-model-based process knowledge base construction and multi-turn memory question-answering method is also provided. This system may include: The acquisition unit is used to acquire regional information associated with each type of application and key process information corresponding to each type of application, and to form a target document. The segmentation unit is used to segment the data in the target document into knowledge segments, use an open-source semantic vector and reordering model to vectorize the segmented content, and store the results in a local vector database to build a document processing knowledge base and configure the target function. The output unit is used to process the original search question input by the user by combining the constructed multi-turn memory question-answering intelligent model and the handling process knowledge base, and output the reply content in a point-by-point manner. The reply content is automatically associated with the direct link and handling process diagram corresponding to the handling item. The update unit is used to update the knowledge base of the handling process based on the feedback of the problem and to execute a new round of memory questions and answers.

[0062] Through the aforementioned modules, target documents are formed by acquiring the regional information associated with each service item and the key process information corresponding to various service items. The data in the target documents is then segmented into knowledge segments, and the segmented content is vectorized using an open-source semantic vector and reordering model. The results are stored in a local vector database to construct a service item process knowledge base and configure the target function. The original search question input by the user is processed using the constructed multi-turn memory question-and-answer intelligent model and the service item process knowledge base, and the response content is output in a point-by-point manner. The response content automatically associates the direct links and service item process diagrams corresponding to the service item. Based on the question feedback, the service item process knowledge base is updated, and a new round of memory question-and-answer is executed. This standardized service item process knowledge base solves the problem of inaccurately associating and extracting the administrative region and business type of the service item in the search question, realizing an intelligent retrieval function for service item knowledge with multi-turn memory capabilities.

[0063] According to another aspect of the embodiments of this application, a storage medium is also provided. Optionally, in this embodiment, the storage medium can be used to execute the program code of any of the above-mentioned methods for constructing a knowledge base for a case handling process based on a large model and for multi-round memory question answering in the embodiments of this application.

[0064] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: S1, obtain the regional information associated with each case and the key process information corresponding to each type of case, and form the target document; S2, the data in the target document is segmented into knowledge segments, and the segmented content is vectorized using an open-source semantic vector and reordering model. The results are stored in a local vector database to build a document processing knowledge base and the target function is configured. S3, combining the constructed multi-round memory question-and-answer intelligent model and the handling process knowledge base, processes the original search question input by the user and outputs the reply content in a point-by-point manner, wherein the reply content automatically associates the direct link and handling process diagram corresponding to the handling item; S4 updates the knowledge base of the handling process based on the feedback and executes a new round of memory-based question and answer.

[0065] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated in this embodiment.

[0066] The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0067] According to another aspect of the embodiments of this application, an electronic device is also provided for implementing the above-described method for constructing a knowledge base for handling procedures based on a large model and for multi-round memory question answering. The electronic device may be a server, a terminal, or a combination thereof.

[0068] Figure 4 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application, such as... Figure 4 As shown, it includes a processor 402, a communication interface 404, a memory 406, and a communication bus 408. The processor 402, communication interface 404, and memory 406 communicate with each other via the communication bus 408. Memory 406 is used to store computer programs; When processor 402 executes a computer program stored in memory 406, it performs the following steps: S1, obtain the regional information associated with each case and the key process information corresponding to each type of case, and form the target document; S2, the data in the target document is segmented into knowledge segments, and the segmented content is vectorized using an open-source semantic vector and reordering model. The results are stored in a local vector database to build a document processing knowledge base and the target function is configured. S3, combining the constructed multi-round memory question-and-answer intelligent model and the handling process knowledge base, processes the original search question input by the user and outputs the reply content in a point-by-point manner, wherein the reply content automatically associates the direct link and handling process diagram corresponding to the handling item; S4 updates the knowledge base of the handling process based on the feedback and executes a new round of memory-based question and answer.

[0069] Optionally, the communication bus can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4The symbol is represented by a single thick line, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned electronic device and other devices.

[0070] The memory may include RAM, or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0071] The processors mentioned above can be general-purpose processors, including but not limited to: CPU (Central Processing Unit), NP (Network Processor), etc.; they can also be DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0072] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0073] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0074] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0075] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0076] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of embodiments of this disclosure upon considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

[0077] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0078] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing a knowledge base for handling procedures based on a large model and a multi-round memory-based question-and-answer method, characterized in that, include: Obtain the regional information associated with each case and the key process information corresponding to each type of case, and form a target document; The data in the target document is segmented into knowledge segments. The segmented content is vectorized using an open-source semantic vector and reordering model. The results are stored in a local vector database to build a document processing knowledge base and configure the target function. The system combines the constructed multi-round memory question-and-answer intelligent model with the process knowledge base to process the original search questions input by the user and outputs the response content in a point-by-point manner. The response content is automatically associated with the direct links and process flowcharts corresponding to the matters to be handled. The knowledge base for handling procedures is updated based on feedback, and a new round of memory-based question and answer is executed.

2. The method for constructing a knowledge base for the handling process based on a large model and for multi-round memory-based question answering as described in claim 1, characterized in that, The step of performing knowledge segmentation on the data in the target document includes: Use the first row of the header in the target document as the key, and each row of data as the corresponding value; Convert the data in the target document into a list of key-value pairs.

3. The method for constructing a knowledge base for the handling process based on a large model and for multi-round memory-based question answering as described in claim 1, characterized in that, The process of handling the original retrieval question input by the user, which combines the constructed multi-turn memory question-answering intelligent model and the processing procedure knowledge base, includes: The original retrieval question is analyzed using a multi-turn memory question-answering intelligent model combined with custom model instructions, and the regional information and item information in the original retrieval question are extracted to obtain the extraction results; The target output is generated based on the extraction results.

4. The method for constructing a knowledge base for the handling process based on a large model and for multi-round memory-based question answering as described in claim 3, characterized in that, The step of generating the target output based on the extraction result includes: If the region information and event information are successfully identified in the extraction results, they are output according to a preset format; If no area information or event information is successfully identified in the extraction results, the corresponding field in the target output will be assigned an empty value.

5. The method for constructing a knowledge base for the handling process based on a large model and for multi-round memory-based question answering as described in claim 1, characterized in that, The multi-round memory question-answering intelligent model and the processing workflow knowledge base are combined to process the user's original search question, and the point-by-point output response content includes: Create and maintain session variables for current event information and current region information; When non-empty region information and event information are successfully extracted from the original search query, update the current event information session variable and the current region information session variable; otherwise, leave them unchanged.

6. The method for constructing a knowledge base for handling procedures based on a large model and for multi-round memory-based question answering as described in claim 5, characterized in that, After updating the current event information session variable and the current region information session variable, the method further includes: Using the updated current item information session variables and current area information session variables, along with the original search question, reconstruct the original search question according to the preset template.

7. The method for constructing a knowledge base for handling procedures based on a large model and for multi-round memory-based question answering as described in claim 6, characterized in that, After reconstructing the original search question according to the preset template, the method further includes: Based on the reconstructed original retrieval question, a knowledge base retrieval of the application process is performed, and a semantic vector model is used to calculate semantic similarity to obtain preliminary retrieval results; By combining a re-ranking model to optimize and rank similar segments in the initial search results, the final search results are obtained. Based on the search results and the prompt word template, the inference language model is invoked to perform inference analysis and structured generation, and the response content is output in a point-by-point manner.

8. A system for constructing a knowledge base for handling procedures based on a large model and a multi-turn memory-based question-and-answer system, characterized in that: include: The acquisition unit is used to acquire regional information associated with each type of application and key process information corresponding to each type of application, and to form a target document. The segmentation unit is used to segment the data in the target document into knowledge segments, use an open-source semantic vector and reordering model to vectorize the segmented content, and store the results in a local vector database to build a document processing knowledge base and configure the target function. The output unit is used to process the original search question input by the user by combining the constructed multi-turn memory question-answering intelligent model and the handling process knowledge base, and output the reply content in a point-by-point manner. The reply content is automatically associated with the direct link and handling process diagram corresponding to the handling item. The update unit is used to update the knowledge base of the handling process based on the feedback of the problem and to execute a new round of memory questions and answers.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the method of any one of claims 1 to 7.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method of any one of claims 1 to 7 through the computer program.

Images