Question answering method and device for multi-level intention recognition, electronic equipment and storage medium

Abstract

The invention discloses a multilevel intention recognition question and answer method and device, electronic equipment and a storage medium, through the application, semantic analysis is performed on natural language query of a user by means of a large model, whether the query is a target service related problem is accurately judged, a traditional rule or keyword-based matching mode is replaced, and the user experience is improved. The natural language processing capability is effectively improved; performing service type classification on queries through the semantic understanding capability of the large model, and dynamically generating corresponding service processing flows, thereby breaking through the limitation of a traditional static architecture, and adapting to a compound service scene; and finally, in combination with a dynamic generation process, vector retrieval is carried out by utilizing a retrieval enhancement generation knowledge base, so that a large model generates accurate answers based on retrieved business knowledge, and answer stiffness caused by dependence on a fixed knowledge base is avoided. The technical effects of improving natural language query understanding accuracy, flexibly adapting to various business scenes, generating accurate answers fitting business knowledge and optimizing user consultation experience and business processing efficiency are achieved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of data processing, and particularly relates to a multi-level intent recognition question and answer method and device, electronic equipment and storage medium. BACKGROUND

[0002] As a core application in the field of insurance information technology, the life insurance business question and answer system widely serves key business links such as product consultation, contract processing, claim operation and claim settlement process. With the increasing complexity of insurance business and the diversification of user needs, the existing system mostly adopts a static processing architecture based on rules or keyword matching, and its technical system mainly relies on the cooperative work of a preset semantic analysis module and a fixed knowledge base. However, the traditional system has significant limitations when dealing with complex business scenarios, and its technical evolution has always failed to break through the lack of natural language processing capability and business. SUMMARY

[0003] The present disclosure provides a multi-level intent recognition question and answer method, device, electronic equipment and storage medium.

[0004] According to a first aspect of the present disclosure, a multi-level intent recognition question and answer method is provided, which comprises: receiving a natural language query input by a user, and using a large model to perform semantic analysis on the query content to determine whether it is a target business-related question; if it is determined to be the target business-related question, performing business type classification on the query content based on the semantic understanding capability of the large model; generating a corresponding business processing flow dynamically according to the business type classification result; combining the dynamically generated business processing flow, using retrieval enhancement to generate a knowledge base for vector retrieval on the query content, and generating a precise answer containing business knowledge based on the retrieval result through the large model.

[0005] Optionally, the receiving a natural language query input by a user, and using a large model to perform semantic analysis on the query content to determine whether it is a target business-related question comprises: using a pre-trained embedding model to convert the natural language query into a vector representation; calculating the similarity threshold of the query vector and a preset life insurance business feature vector, and determining that it is a target business-related question when the similarity is greater than or equal to the similarity threshold.

[0006] Optionally, the performing business type classification on the query content based on the semantic understanding capability of the large model comprises: performing preliminary matching in combination with a preset business type keyword set; The cosine similarity of the query text and each business type semantic template is calculated by a large model, and the business type corresponding to the maximum similarity is selected as the classification result.

[0007] Optionally, the dynamically generating a corresponding business processing flow according to the business type classification result comprises: For product and security issues, an interactive flow containing greetings, questions, and conditional branches is generated based on a preset domain-specific language template; For contract and claim issues, an execution flow is generated by parsing the step nodes and conditional branches in the preset domain-specific language template.

[0008] Optionally, the vector retrieval of the query content by the knowledge base generated by retrieval enhancement in combination with the dynamically generated business processing flow comprises: An approximate nearest neighbor search algorithm is used to retrieve knowledge fragments with a similarity greater than a preset threshold from the vector database; The retrieved knowledge fragments are input into the large model as context to generate precise answers containing business knowledge.

[0009] Optionally, the method further comprises: According to the business type classification result, the retrieval result is stored using a partitioning strategy of the cache database, wherein the partitioning key is generated based on the business type and the user identifier.

[0010] According to a second aspect of the present disclosure, a multi-level intent recognition question and answer device is provided, comprising: A receiving unit is configured to receive a natural language query input by a user, and perform semantic analysis on the query content using a large model to determine whether it is a target business-related question; A classification unit is configured to, if it is determined to be the target business-related question, perform business type classification on the query content based on the semantic understanding ability of the large model; A generation unit is configured to dynamically generate a corresponding business processing flow according to the business type classification result; A retrieval unit is configured to, in combination with the dynamically generated business processing flow, generate a knowledge base for vector retrieval of the query content using retrieval enhancement, and generate precise answers containing business knowledge based on the retrieval result through a large model.

[0011] Optionally, the receiving unit is further configured to: A pre-trained embedding model is used to convert the natural language query into a vector representation; The similarity threshold of the query vector and the preset life insurance business feature vector is calculated, and when the similarity is greater than or equal to the similarity threshold, it is determined to be a target business-related question.

[0012] Optionally, the classification unit is further configured to: Initial matching is performed based on a pre-defined set of keywords related to business types; The cosine similarity between the query text and the semantic templates of each business type is calculated using a large model, and the business type with the highest similarity is selected as the classification result.

[0013] Optionally, the generation unit is further configured to: For product-related and security-related issues, an interactive flow including greetings, questions, and conditional branches is generated based on preset domain-specific language templates; For contract-related and claims-related issues, an execution flow is generated by parsing the step nodes and conditional branches in a preset domain-specific language template.

[0014] Optionally, the retrieval unit is further configured to: An approximate nearest neighbor search algorithm is used to retrieve knowledge fragments from a vector database whose similarity to the query vectors is greater than a preset threshold. The retrieved knowledge fragments are used as context input to a large model to generate accurate answers that include business knowledge.

[0015] Optionally, the device further includes: The retrieval unit is also used to classify results according to business type and store retrieval results using a partitioning strategy of the cache database, where the partition key is generated based on business type and user identifier.

[0016] According to a third aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect above.

[0017] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described in the first aspect above.

[0018] According to a fifth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method described in the first aspect above.

[0019] The multi-level intent recognition question-answering method, apparatus, electronic device, and storage medium disclosed herein, through the use of a large model to perform semantic analysis on user natural language queries, accurately determines whether the question is related to the target business, replacing the traditional rule-based or keyword-based matching method, effectively improving natural language processing capabilities; furthermore, the semantic understanding capability of the large model is used to classify queries into business types and dynamically generate corresponding business processing flows, breaking the limitations of traditional static architecture and adapting to complex business scenarios; finally, combined with the dynamically generated flow, a knowledge base is generated using retrieval enhancement for vector retrieval, allowing the large model to generate accurate answers based on the retrieved business knowledge, avoiding the rigidity of answers caused by relying on a fixed knowledge base. Therefore, it can solve the technical problems of existing life insurance business question-answering systems that adopt a static processing architecture, have insufficient natural language processing capabilities, and are difficult to adapt to complex business scenarios, achieving the technical effects of improving the accuracy of natural language query understanding, flexibly adapting to various business scenarios, generating accurate and business-related answers, and optimizing user consultation experience and business processing efficiency.

[0020] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0021] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 A flowchart illustrating a multi-level intent recognition question-answering method provided in an embodiment of this disclosure; Figure 2 A schematic diagram of the structure of a question-answering device for multi-level intent recognition provided in an embodiment of this disclosure; Figure 3 A schematic block diagram of an example electronic device provided for embodiments of this disclosure. Detailed Implementation

[0022] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0023] The following description, with reference to the accompanying drawings, outlines a question-answering method, apparatus, electronic device, and storage medium for multi-level intent recognition, according to embodiments of the present disclosure.

[0024] Figure 1This is a flowchart illustrating a multi-level intent recognition question-answering method provided in an embodiment of the present disclosure.

[0025] like Figure 1 As shown, the method includes the following steps: Step 101: Receive the natural language query input by the user, and use the large model to perform semantic analysis on the query content to determine whether it is a question related to the target business. The system first receives queries input by users in natural language. Users can complete the input through pre-defined user interfaces such as web pages, mobile applications, or application programming interfaces (APIs). This input design is simple and intuitive, allowing users to easily express their business inquiries. The system then invokes a large-scale model to perform semantic analysis on the received queries. This large-scale model technology refers to a large neural network model with powerful natural language processing and understanding capabilities, capable of processing and analyzing massive amounts of text data and extracting the core semantic information behind the text. During semantic analysis, the large-scale model performs deep semantic understanding of the user's natural language query, identifying business-related elements and the user's core needs within the query. Based on a pre-defined target business scope, it determines whether the query falls within the scope of the target business, providing a preliminary basis for subsequent business processing.

[0026] Step 102: If the problem is determined to be related to the target business, the query content is classified by business type based on the semantic understanding capability of the large model. The system fully leverages the semantic understanding capabilities of large-scale neural network models for classification. These models, possessing powerful natural language processing and understanding capabilities, not only identify explicit keywords in queries but also delve deeper to uncover hidden user business needs and potential connections. During the classification process, the large-scale model combines pre-defined definitions of various business types covered by the target business to perform multi-dimensional semantic analysis and feature matching on user queries. For example, when a user queries about insurance products, the large-scale model can semantically identify that the inquiry pertains to a product-related business; when a user queries about claims processing, it can accurately determine that it belongs to the claims category. Through this semantic analysis and matching process, the system can accurately categorize user queries into the corresponding business types, providing clear guidance for subsequent execution of specific processing procedures for different business types.

[0027] Step 103: Dynamically generate the corresponding business processing flow based on the business type classification results; A business process flow refers to a series of ordered operations or interactive logics defined to achieve specific business objectives. Its generation process is closely integrated with the characteristics of the specific business type identified through classification. The system first invokes a pre-defined business process framework, which contains the general logic and structure of various business processes. Then, relying on the semantic understanding and logic construction capabilities of the large model, the framework is adaptively adjusted and improved according to the core needs of the current specific business type. For example, if the classification result points to a business type that requires fixed steps, the generated flow will clearly define the sequence of each operation step and the core task of each step; if the classification result corresponds to a business type requiring multiple rounds of interactive guidance, the generated flow will focus on designing dialogue logic and information collection nodes that conform to user communication habits. The entire dynamic generation process does not require manual configuration; instead, the system automatically matches the corresponding generation rules based on the business type, quickly outputting a processing flow adapted to the current business scenario, providing clear and demand-aligned process support for subsequent user interactions or the orderly advancement of business.

[0028] Step 104: Combining the dynamically generated business processing flow, the retrieval enhancement knowledge base is used to perform vector retrieval on the query content, and based on the retrieval results, a precise answer containing business knowledge is generated through a large model.

[0029] The retrieval-enhanced generative knowledge base is a knowledge storage and retrieval module that integrates retrieval and generation technologies. Internally, it stores various business knowledge related to the target business, including business terms, processes, guidelines, and operational specifications. This knowledge is stored in vector form in a vector database for efficient retrieval. The system first clarifies the business scenario and knowledge requirements corresponding to the current query based on dynamically generated business processes. Then, it converts the user's natural language query into vector form. This conversion process relies on a pre-trained embedding model to ensure that the query vector accurately reflects the semantic core of the user's query. Next, in the vector database of the retrieval-enhanced generative knowledge base, the system uses techniques such as approximate nearest neighbor search to match knowledge vectors with high semantic similarity to the query vector, selecting knowledge content highly consistent with the requirements of the current business process as the retrieval results. Finally, the system inputs the retrieved business knowledge and the logical requirements of the dynamically generated business process into a large model. Leveraging its powerful natural language generation capabilities, the large model deeply integrates the retrieved knowledge with the business process requirements, generating accurate answers that contain precise business knowledge and conform to the current business scenario. This ensures both the professionalism and accuracy of the answers and their synergy with the business process, providing users with information support tailored to their actual needs.

[0030] In some embodiments, receiving a natural language query input by the user and performing semantic analysis on the query content using a large model to determine whether it is a question related to the target business includes: A pre-trained embedding model is used to convert natural language queries into vector representations; By calculating the similarity threshold between the query vector and the preset life insurance business feature vector, a problem is determined to be related to the target business when the similarity is greater than or equal to the similarity threshold.

[0031] The pre-trained embedding model is a model trained on massive amounts of text data. It has the ability to convert unstructured natural language text into structured vector representations that retain semantic information. This model can accurately convert the business needs and core semantics contained in user queries into vector form, providing a data foundation for subsequent judgments. The system also pre-stores preset life insurance business feature vectors. These feature vectors are generated based on a large amount of life insurance business-related text data, such as life insurance product introductions, claims process instructions, and policy maintenance operation guides. They can accurately reflect the semantic features of the life insurance business domain and serve as a benchmark for determining the business category to which the query belongs.

[0032] The system calculates the similarity between the query vector and the preset life insurance business feature vector using common vector similarity calculation methods such as cosine similarity. Simultaneously, the system pre-sets a similarity threshold that aligns with actual business needs. This threshold, determined through verification using historical business data, is used to clearly classify whether a query belongs to the target business. When the calculated similarity value is greater than or equal to this preset similarity threshold, the system determines that the user's natural language query is related to the target business, providing a basis for subsequent operations such as business type classification.

[0033] In some embodiments, the business type classification of query content based on the semantic understanding capability of the large model includes: Initial matching is performed based on a pre-defined set of keywords related to business types; The cosine similarity between the query text and the semantic templates of each business type is calculated using a large model, and the business type with the highest similarity is selected as the classification result.

[0034] The pre-defined keyword set for business types is generated in advance based on the characteristics of various business types covered by the target business. Each business type corresponds to a set of core keywords. For example, product-related businesses correspond to a set of keywords including insurance product coverage and premiums, while claims-related businesses correspond to a set of keywords including claims process, reporting, and claims review. The system scans the user's query text to identify whether it contains keywords from the keyword sets for each business type. Through preliminary matching, it filters out several candidate business types that are keyword-related to the query text, narrowing down the scope for subsequent classification. Subsequently, relying on the semantic understanding capabilities of the large-scale model, it further refines the classification. The system pre-builds corresponding semantic templates for each business type. These semantic templates are generated based on a large number of historical query samples and business documents under that business type, and can fully reflect the core semantic features of that business type. The large-scale model converts the user's query text and the semantic templates of each candidate business type into vector form, and then calculates the cosine similarity between the query text vector and the semantic template vectors of each candidate business type. The higher the cosine similarity value, the stronger the semantic relationship between the two. Finally, the system selects the one with the highest cosine similarity value from all the results and determines the corresponding business type as the final business type classification result.

[0035] In some embodiments, dynamically generating the corresponding business processing flow based on the business type classification result includes: For product-related and security-related issues, an interactive flow including greetings, questions, and conditional branches is generated based on preset domain-specific language templates; For contract-related and claims-related issues, an execution flow is generated by parsing the step nodes and conditional branches in a preset domain-specific language template.

[0036] Domain-specific language templates are dedicated language templates designed for the characteristics of various target business types, containing the structural elements and logical rules required for various processes. When the classification result is a product-related or policy maintenance-related question, the system calls the corresponding domain-specific language template to generate an interactive flow that includes a greeting, a question, and conditional branches. The greeting is used to respond amicably to user needs and quickly establish communication; the question section guides users to supplement specific information based on common inquiries for this type of business, such as asking about the coverage or premium range for product-related questions, or asking about the specific type of policy maintenance service the user needs in policy maintenance-related questions; the conditional branches automatically adjust the subsequent interaction direction based on the user's response to the question, ensuring that the process aligns with the user's actual needs.

[0037] When the classification result is a contract-related or claims-related issue, the system parses the domain-specific language template for the corresponding category, extracts the pre-defined step nodes and conditional branches from the template, and generates an ordered execution flow. The step nodes clearly define each operational step in this type of business, from initiating a request to completing the processing, such as the reporting, investigation, loss assessment, and claims settlement steps in claims-related issues. Conditional branches set up response logic for different situations that may occur in the process; for example, in contract-related issues, subsequent steps are adjusted based on whether the user's submitted materials are complete, ensuring that the execution flow can proceed in an orderly manner and adapt to various business scenarios.

[0038] In some embodiments, the step of combining dynamically generated business processing flows with retrieval enhancement to generate a knowledge base for vector retrieval of query content includes: An approximate nearest neighbor search algorithm is used to retrieve knowledge fragments from a vector database whose similarity to the query vectors is greater than a preset threshold. The retrieved knowledge fragments are used as context input to a large model to generate accurate answers that include business knowledge.

[0039] The enhanced knowledge base stores business knowledge in vector form within a vector database. These knowledge vectors are generated from various business-related knowledge fragments through a pre-trained embedding model, accurately carrying the semantic features of the knowledge and laying the foundation for efficient retrieval. During the retrieval process, the system employs an approximate nearest neighbor search algorithm. This algorithm can quickly traverse massive amounts of knowledge vector data and calculate the similarity between each knowledge vector and the user's query vector, significantly improving retrieval efficiency compared to traditional retrieval methods and adapting to the growing scale of the knowledge base. Simultaneously, the system filters the calculation results based on a preset similarity threshold, retaining only knowledge fragments with a similarity greater than this threshold. This threshold is set according to the knowledge relevance requirements of the business scenario, effectively excluding irrelevant or low-relevance knowledge fragments and ensuring the matching degree of the retrieval results with the current query and business processing flow. Subsequently, the system uses the filtered knowledge fragments as contextual information, inputting them along with the user's query and business processing flow logic into a large model. Based on this specific business knowledge context, the large model can accurately grasp user needs and business specifications, thereby generating answers containing accurate business knowledge, ensuring the professionalism and relevance of the answers.

[0040] In some embodiments, the method further includes: Based on the business type classification results, a partitioning strategy is used in the cache database to store the retrieval results, where the partition key is generated based on the business type and user identifier.

[0041] After classifying the business types of user queries, the system uses a cache database partitioning strategy to store the search results output by the enhanced knowledge base. The cache database boasts efficient data read / write capabilities, enabling rapid response to data access requests. Partitioned storage further optimizes data management efficiency and avoids access latency issues caused by centralized data storage. The core of the partitioning strategy lies in dividing the cache database into multiple independent storage areas, each corresponding to a specific data category. This allows subsequent data queries to quickly locate the target storage area, reducing the data traversal scope.

[0042] The partition key is generated based on the business type classification result and the user identifier. The business type classification result clarifies the business category to which the search results belong, such as product category, claims category, etc., while the user identifier is used to distinguish the search data of different users. The partition key formed by the combination of the two can ensure that the search data of the same business type is stored relatively centrally, while also achieving the isolation of data from different users and avoiding data confusion. This storage method not only allows the system to quickly locate the corresponding cache partition based on the business type and user identifier when the same user initiates a query of the same business type again, without having to re-execute the knowledge base vector search operation, greatly shortening the query response time, but also effectively organizes the search data of different users and different business types, ensuring the orderliness and manageability of cached data, and improving the overall system operating efficiency.

[0043] Corresponding to the multi-level intent recognition question-answering method described above, this invention also proposes a multi-level intent recognition question-answering device. Since the device embodiments of this invention correspond to the method embodiments described above, details not disclosed in the device embodiments can be referred to in the method embodiments described above, and will not be repeated here.

[0044] Figure 2 This is a schematic diagram of the structure of a multi-level intent recognition question-answering device provided in an embodiment of this disclosure, as shown below. Figure 2 As shown, it includes: The receiving unit 21 is used to receive natural language queries input by the user and use a large model to perform semantic analysis on the query content to determine whether it is a question related to the target business. Classification unit 22 is used to classify the query content by business type based on the semantic understanding capability of the large model if it is determined to be a problem related to the target business. The generation unit 23 is used to dynamically generate the corresponding business processing flow based on the business type classification results; The retrieval unit 24 is used to combine dynamically generated business processing flows, utilize retrieval enhancement to generate a knowledge base to perform vector retrieval of query content, and generate accurate answers containing business knowledge based on the retrieval results through a large model.

[0045] Furthermore, in one possible implementation of this disclosure, the receiving unit 21 is further configured to: A pre-trained embedding model is used to convert natural language queries into vector representations; By calculating the similarity threshold between the query vector and the preset life insurance business feature vector, a problem is determined to be related to the target business when the similarity is greater than or equal to the similarity threshold.

[0046] Furthermore, in one possible implementation of this disclosure, the classification unit 22 is further configured to: Initial matching is performed based on a pre-defined set of keywords related to business types; The cosine similarity between the query text and the semantic templates of each business type is calculated using a large model, and the business type with the highest similarity is selected as the classification result.

[0047] Furthermore, in one possible implementation of this disclosure embodiment, the generation unit 23 is further configured to: For product-related and security-related issues, an interactive flow including greetings, questions, and conditional branches is generated based on preset domain-specific language templates; For contract-related and claims-related issues, an execution flow is generated by parsing the step nodes and conditional branches in a preset domain-specific language template.

[0048] Furthermore, in one possible implementation of this disclosure, the retrieval unit 24 is further configured to: An approximate nearest neighbor search algorithm is used to retrieve knowledge fragments from a vector database whose similarity to the query vectors is greater than a preset threshold. The retrieved knowledge fragments are used as context input to a large model to generate accurate answers that include business knowledge.

[0049] Furthermore, in one possible implementation of this disclosure, the apparatus further includes: The retrieval unit 24 is also used to classify the results according to the business type and store the retrieval results using the partitioning strategy of the cache database, wherein the partition key is generated based on the business type and the user identifier.

[0050] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of the embodiments of this disclosure, and the principle is the same. Therefore, the embodiments of this disclosure are not limited thereto.

[0051] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0052] Figure 3A schematic block diagram of an example electronic device 400 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0053] like Figure 3 As shown, device 400 includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in ROM (Read-Only Memory) 402 or a computer program loaded from storage unit 408 into RAM (Random Access Memory) 403. RAM 403 may also store various programs and data required for the operation of device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. I / O (Input / Output) interface 405 is also connected to bus 404.

[0054] Multiple components in device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of monitors, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0055] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, CPUs (Central Processing Units), GPUs (Graphics Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processors), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as a multi-level intent recognition question-answering method. For example, in some embodiments, the multi-level intent recognition question-answering method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed on device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of the methods described above can be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to perform the aforementioned multi-level intent recognition question-and-answer method by any other suitable means (e.g., by means of firmware).

[0056] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System-on-Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0057] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0058] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0059] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0060] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.

[0061] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service system that addresses the shortcomings of traditional physical hosts and VPS (Virtual Private Server) services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0062] It's important to note that artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0063] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0064] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A multi-stage intent recognition question and answer method, characterized by, The method comprises the following steps: receiving a natural language query input by a user, and performing semantic analysis on the query content by using a large model to determine whether it is a target business-related question; if it is determined to be the target business-related question, performing business type classification on the query content based on the semantic understanding capability of the large model; generating a corresponding business processing flow dynamically according to the business type classification result; combining the dynamically generated business processing flow, performing vector retrieval on the query content by using a retrieval-enhanced knowledge base, and generating a precise answer containing business knowledge by the large model based on the retrieval result.

2. The method of claim 1, wherein, The method comprises the following steps: using a pre-trained embedding model to convert the natural language query into a vector representation; calculating the similarity threshold of the query vector and the preset life insurance business feature vector, and determining that it is a target business-related question when the similarity is greater than or equal to the similarity threshold.

3. The method of claim 1, wherein, The method comprises the following steps: combining a preset business type keyword set for preliminary matching; calculating the cosine similarity of the query text and each business type semantic template by using the large model, and selecting the business type corresponding to the maximum similarity as the classification result.

4. The method of claim 1, wherein, The method comprises the following steps: for product and insurance-related questions, generating an interactive flow containing greetings, questions, and conditional branches based on a preset domain-specific language template; for contract and claim-related questions, generating an execution flow by analyzing the step nodes and conditional branches in the preset domain-specific language template.

5. The method of claim 1, wherein, The method comprises the following steps: using an approximate nearest neighbor search algorithm to retrieve knowledge fragments with a similarity greater than a preset threshold from a vector database; inputting the retrieved knowledge fragments as context into the large model to generate a precise answer containing business knowledge.

6. The method of claim 1, wherein, The method further comprises the following steps: based on the business type classification result, using a partitioning strategy of a cache database to store the retrieval result, wherein the partitioning key is generated based on the business type and the user identifier. 7.A question and answer device of multi-level intent recognition, characterized by, The method comprises the following steps: a receiving unit configured to receive a natural language query input by a user, and perform semantic analysis on the query content by using a large model to determine whether it is a target business-related question; a classification unit configured to perform business type classification on the query content based on the semantic understanding capability of the large model if it is determined to be the target business-related question; a generation unit configured to generate a corresponding business processing flow dynamically according to the business type classification result; a retrieval unit configured to perform vector retrieval on the query content by using a retrieval-enhanced knowledge base in combination with the dynamically generated business processing flow, and generate a precise answer containing business knowledge by the large model based on the retrieval result.

8. An electronic device, comprising: The method comprises the following steps: at least one processor; and a memory connected in communication with the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.

9. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, The computer instructions are for causing the computer to perform the method of any one of claims 1-6.

10. A computer program product, characterised in that, A computer program comprising instructions which, when executed by a processor, implement the method of any one of claims 1-6.

Images