Problem processing method and device based on artificial intelligence, computer equipment and medium

Abstract

The invention belongs to the technical field of artificial intelligence, and relates to an artificial intelligence-based question processing method and device, computer equipment and a storage medium, and the method comprises the steps: receiving a question text input by a user; performing task analysis on the problem text to obtain a task analysis result; retrieving a target strategy unit related to the task analysis result from an experience library; fusing the target strategy unit and the question text to obtain a target prompt sequence; based on the graph neural network, performing strategy path selection and combination processing on the target strategy unit to obtain a target strategy path; based on the target prompt sequence, using a large language model to execute processing on the target strategy path, and generating corresponding answer data; and if the answer data passes the knowledge verification, performing output processing on the answer data. In addition, answer data may be stored in a blockchain. The method can be applied to question and answer processing scenes in the field of financial science and technology, the processing efficiency of question processing is effectively improved, and the accuracy of answer data is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology and can be applied to the financial technology field, particularly to artificial intelligence-based problem-solving methods, devices, computer equipment, and storage media. Background Technology

[0002] In the insurance and finance sector, with the continuous development of artificial intelligence technology, large language models (LLMs) have been gradually applied to various problem-solving scenarios in recent years, covering multiple aspects such as intelligent customer service, interpretation of insurance terms, claims assistance, and compliance auditing, providing strong technical support for the development of insurance business.

[0003] However, current products in the industry that use large language models to handle insurance business issues have significant shortcomings. Existing products generally rely on large language models and manually constructed prompts for problem-solving. Specifically, for different insurance business scenarios, such as health insurance Q&A, auto insurance claims review, and financial compliance checks, experts need to design corresponding experience-based prompts. This process requires experts to invest a significant amount of time and effort, resulting in extremely low efficiency in problem-solving. Furthermore, due to the subjectivity and limitations of manual design, it is impossible to ensure the high accuracy of the generated answers, making it difficult to meet the stringent quality and efficiency requirements of the insurance and financial industry.

[0004] Therefore, there is an urgent need to develop a new technology that can improve the efficiency of problem-solving in insurance business and ensure the accuracy of answers. Summary of the Invention

[0005] The purpose of this application is to propose a problem-solving method, apparatus, computer device, and storage medium based on artificial intelligence, so as to solve the technical problems of low processing efficiency and inability to guarantee the accuracy of answers in existing problem-solving methods.

[0006] Firstly, an artificial intelligence-based problem-solving method is provided, including: Receive user-input question text; The problem text is parsed based on a preset parsing strategy to obtain the corresponding parsing results; Retrieve target strategy units related to the task parsing results from a pre-set experience base; The target strategy unit and the question text are fused together to obtain the corresponding target prompt sequence; Based on a preset graph neural network, the target policy unit is processed to select and combine policy paths to obtain the corresponding target policy path; Based on the target prompt sequence, the target strategy path is processed using a preset large language model to generate corresponding answer data; Perform knowledge verification on the answer data; If the answer data passes the knowledge verification, then the answer data will be output.

[0007] Secondly, an artificial intelligence-based problem-solving device is provided, comprising: The receiving module is used to receive the question text input by the user; The parsing module is used to perform task parsing on the question text based on a preset parsing strategy to obtain the corresponding task parsing results; The retrieval module is used to retrieve target strategy units related to the task parsing results from a preset experience base; The fusion module is used to fuse the target strategy unit and the question text to obtain a corresponding target prompt sequence; The processing module is used to select and combine policy paths for the target policy unit based on a preset graph neural network to obtain the corresponding target policy path. The generation module is used to process the target strategy path based on the target prompt sequence using a preset large language model to generate corresponding answer data; The verification module is used to perform knowledge verification on the answer data; The output module is used to output the answer data if the answer data passes the knowledge verification.

[0008] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described artificial intelligence-based problem-solving method.

[0009] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned problem-solving method based on artificial intelligence.

[0010] In the aforementioned solution implemented by the AI-based problem-solving method, apparatus, computer equipment, and storage medium, the user-inputted problem text is first received; then, the problem text is parsed based on a preset parsing strategy to obtain the corresponding task parsing result; next, target strategy units related to the task parsing result are retrieved from a preset experience base; and the target strategy units and problem text are fused to obtain the corresponding target prompt sequence; subsequently, the target strategy units are processed by selecting and combining strategy paths based on a preset graph neural network to obtain the corresponding target strategy path; further, based on the target prompt sequence, a preset large language model is used to execute the target strategy path to generate the corresponding answer data; finally, the answer data undergoes knowledge verification; if the answer data passes the knowledge verification, the answer data is output. Based on the above automated processing flow, this application can automatically generate target prompt sequences and target strategy paths that match the input question text through strategy unit abstraction and cross-task experience transfer processing. This allows the system to avoid manually designing prompts for each new task, significantly reducing labor costs. Furthermore, it utilizes a large language model to process the target prompt sequences and target strategy paths to generate answer data and outputs answer data that has passed knowledge verification. This effectively improves the processing efficiency of question handling and enhances the accuracy and reliability of the generated answer data. Attached Figure Description

[0011] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is an exemplary system architecture diagram to which this application can be applied; Figure 2 This is a flowchart of an embodiment of the problem-solving method based on artificial intelligence according to this application; Figure 3 This is a schematic diagram of a structure of an embodiment of the artificial intelligence-based problem processing device according to this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0014] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0016] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0017] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0018] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0019] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0020] It should be noted that the problem-solving method based on artificial intelligence provided in this application is generally executed by a server / terminal device, and correspondingly, the problem-solving device based on artificial intelligence is generally set in the server / terminal device.

[0021] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0022] Continue to refer to Figure 2 This document illustrates a flowchart of an embodiment of the AI-based problem-solving method according to this application. The order of steps in the flowchart can be changed, and some steps can be omitted, depending on different needs. The AI-based problem-solving method provided in this application can be applied to any scenario requiring problem-solving, and therefore can be applied to products in these scenarios, such as problem-solving products in the financial insurance field. The AI-based problem-solving method includes the following steps: Step S201: Receive the question text input by the user.

[0023] In this embodiment, the problem-solving method based on artificial intelligence runs on an electronic device (e.g., Figure 1The server / terminal device shown can acquire user-inputted question text via wired or wireless connection. It should be noted that the aforementioned wireless connection methods include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future-developed wireless connection methods. The implementing entity of this application is specifically a question processing system, which can be simply referred to as the system. This application can be applied to question processing business scenarios in the financial and insurance field, such as intelligent customer service, insurance clause interpretation, claims assistance, and compliance review. The system can receive users' natural language questions, i.e., user-inputted question text, through various channels, including but not limited to the online customer service window of the insurance institution's official website, the consultation entry in the mobile application, and voice-to-text input via customer service hotline transfers. For example, when a user clicks the "Online Consultation" button on an insurance institution's APP, an input box pops up, where the user can enter their question.

[0024] Step S202: Perform task parsing on the problem text based on a preset parsing strategy to obtain the corresponding task parsing result.

[0025] In this embodiment, the specific implementation process of performing task parsing on the problem text based on the preset parsing strategy to obtain the corresponding task parsing result will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0026] Step S203: Retrieve target strategy units related to the task parsing results from a preset experience base.

[0027] In this embodiment, the specific implementation process of retrieving the target strategy unit related to the task parsing result from the preset experience base will be described in more detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0028] Step S204: The target strategy unit and the question text are fused to obtain the corresponding target prompt sequence.

[0029] In this embodiment, the specific implementation process of fusing the target strategy unit and the question text to obtain the corresponding target prompt sequence will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0030] Step S205: Based on a preset graph neural network, the target strategy unit is processed by selecting and combining strategy paths to obtain the corresponding target strategy path.

[0031] In this embodiment, the specific implementation process of selecting and combining the target strategy unit based on the preset graph neural network to obtain the corresponding target strategy path will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0032] Step S206: Based on the target prompt sequence, the target strategy path is processed using a preset large language model to generate corresponding answer data.

[0033] In this embodiment, the specific implementation process of performing the target strategy path processing based on the target prompt sequence using a preset large language model to generate corresponding answer data will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0034] Step S207: Perform knowledge verification on the answer data.

[0035] In this embodiment, the specific implementation process of the knowledge verification of the answer data described above will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0036] Step S208: If the answer data passes the knowledge verification, then the answer data is output.

[0037] In this embodiment, once the answer data is detected to have passed the knowledge verification, the answer data is sent to the user to achieve the output of the answer data and ensure the accuracy and reliability of the answer data.

[0038] This application first receives the question text input by the user; then, it performs task parsing on the question text based on a preset parsing strategy to obtain the corresponding task parsing result; next, it retrieves the target strategy unit related to the task parsing result from a preset experience base; and then fuses the target strategy unit with the question text to obtain the corresponding target prompt sequence; subsequently, it selects and combines the strategy path for the target strategy unit based on a preset graph neural network to obtain the corresponding target strategy path; further, based on the target prompt sequence, it uses a preset large language model to execute the target strategy path to generate the corresponding answer data; finally, it performs knowledge verification on the answer data; if the answer data passes the knowledge verification, it outputs the answer data. Based on the above automated processing flow, this application, through strategy unit abstraction and cross-task experience transfer processing, can automatically generate a target prompt sequence and target strategy path that match the input question text. This allows the system to avoid manually designing prompts for each new task, significantly reducing labor costs. Furthermore, by using a large language model to process the target prompt sequence and target strategy path to generate answer data and outputting answer data that has passed knowledge verification, it can effectively improve the processing efficiency of question handling and improve the accuracy and reliability of the generated answer data.

[0039] In some alternative implementations, step S202 includes the following steps: The problem text is preprocessed to obtain the corresponding target text.

[0040] In this embodiment, preliminary preprocessing of the user-input question text is performed to improve the accuracy and efficiency of subsequent processing. Specifically, preprocessing includes removing irrelevant characters from the input text, such as redundant spaces and special symbols; performing spell checking and correction, automatically correcting some common spelling errors; and standardizing the capitalization of the text to avoid recognition problems caused by inconsistent capitalization. For example, if a user inputs "Does this critical illness insurance cover diabetic complications?", the system will correct "complications" to "complications" and remove the space before the question mark.

[0041] The target text is processed using natural language understanding to obtain the corresponding understanding results.

[0042] In this embodiment, lexical analysis, syntactic analysis, and semantic analysis, all techniques in natural language processing, can be used to deeply understand the user-inputted question text. Lexical analysis breaks down the text into individual words and labels their parts of speech. For example, "Does this critical illness insurance cover diabetic complications?" is broken down into "Does this / critical / ill / insurance / cover / diabetes / complications / ?" and labeled with their parts of speech. Syntactic analysis determines the grammatical relationships between words and constructs the grammatical structure of the sentence. Semantic analysis understands the overall meaning of the sentence, identifies the topic and key information involved in the question, and thus obtains the corresponding understanding results.

[0043] Based on the understanding results, a matching process is performed with a preset task type template to generate a task type corresponding to the target text.

[0044] In this embodiment, the task type of the user's question is identified by matching the understanding results based on natural language understanding with predefined task type templates. Specifically, the similarity between the text keywords and features extracted from the understanding results and the keywords and features in the predefined task type templates can be calculated. Various methods can be used for similarity calculation, such as cosine similarity and Jaccard similarity. For example, for the user's input question "What are the provisions of this critical illness insurance policy?", the keywords "provisions" and "provisions" are extracted and their similarity is calculated with the keywords in the "insurance terms interpretation" template. If the similarity exceeds a certain threshold, the question is considered to match the template. Then, based on the similarity calculation results, the most matching task type template is determined. If multiple templates have high similarity to the text, further filtering can be performed based on business rules and contextual information to select the most suitable task type. For example, if both "insurance terms interpretation" and "insurance product introduction" templates are matched, but based on the semantic information of the question, a more favorable question about terms interpretation is preferred, then "insurance terms interpretation" is selected as the final task type.

[0045] The task type is compared and mapped with tasks in a preset historical task database to obtain the corresponding processing result.

[0046] In this embodiment, the construction process of the aforementioned historical task database includes: collecting various task data previously processed by the insurance institution, including user-inputted questions, task types, processing results, and solutions. This data can come from multiple channels such as customer service records, online consultation records, and complaint handling records. Then, the collected task data is cleaned and organized to remove duplicate, erroneous, or incomplete data. Simultaneously, the data is classified and labeled according to task type, business area, etc., and corresponding tags and attributes are added to each task to facilitate subsequent querying and comparison. Subsequently, the cleaned and organized task data is stored in a database, which can be a relational database or a non-relational database, selected based on the characteristics of the data and query requirements. Database indexes are established to improve data query efficiency. At the same time, the database is regularly maintained and updated to ensure the accuracy and completeness of the data.

[0047] Furthermore, features are extracted from both the newly identified tasks and tasks in the historical task database. These features can include textual features (such as keywords and word vectors), semantic features (such as semantic roles and semantic relationships), and task type features. For example, for the new task "Does this critical illness insurance cover diabetic complications?", the keywords "critical illness insurance," "coverage," and "diabetic complications" are extracted, along with semantic features (such as inquiring about coverage). Similarly, corresponding features are extracted from tasks in the historical task database. Then, an appropriate similarity calculation method is selected based on the extracted feature type. For textual features, methods such as cosine similarity and BM25 can be used; for semantic features, semantic similarity calculation methods, such as word vector-based similarity calculation, can be used; and for task type features, direct matching is possible. Subsequently, task cluster mapping is performed, that is, based on the similarity calculation results, the new task is compared and mapped to tasks in the historical task database. Tasks with high similarity are grouped together to form task clusters. For example, the current insurance clause interpretation question is mapped to the historical "clause interpretation" task cluster, which contains experience and data from previously processed insurance clause interpretation questions and serves as the corresponding processing result.

[0048] The processing result is used as the task parsing result corresponding to the question text.

[0049] In this embodiment, the mapping results can also be evaluated and optimized. If the mapping results for some tasks are found to be inaccurate, the similarity calculation method and threshold can be further adjusted. Simultaneously, based on processing experience and solutions in the historical task database, references and suggestions are provided for new tasks, improving the efficiency and quality of task processing.

[0050] Based on the above processing flow, this application preprocesses the problem text to obtain the corresponding target text; then, it performs natural language understanding processing on the target text to obtain the corresponding understanding result; subsequently, it matches the understanding result with a preset task type template to generate a task type corresponding to the target text; next, it compares and maps the task type with tasks in a preset historical task database to obtain the corresponding processing result; finally, it uses the processing result as the task parsing result corresponding to the problem text. In this way, this application clarifies the nature and category of the problem through task parsing, providing a precise direction for subsequent experience retrieval. Only by accurately identifying the task type can relevant experience be located from the experience base, avoiding processing errors caused by retrieval of irrelevant experience. At the same time, alignment with historical tasks allows for the reference of experience and methods used in handling similar tasks previously, improving processing efficiency and accuracy, and serving as an important bridge connecting task input and experience retrieval.

[0051] In some optional implementations of this embodiment, step S203 includes the following steps: Call the preset experience library.

[0052] In this embodiment, the construction process of the aforementioned experience base consists of historical task review and analysis, and experience abstraction and strategy unit summarization. Specifically, historical task review and analysis includes: Task data collection: Establishing a dedicated data collection system to collect data on resolved tasks from various channels, including original user input, intermediate results during system processing, final solutions, and user feedback. These channels may include online customer service dialogue records, email inquiry records, complaint handling records, etc. Data processing and cleaning: Processing and cleaning the collected data to remove duplicate, erroneous, or incomplete data. For example, for customer service dialogue records, removing irrelevant chatter and retaining only information directly related to task processing. Simultaneously, standardizing the data format ensures the accuracy of subsequent analysis. Link tracing recording: Embedding a link tracing mechanism in the task processing system to record the complete reasoning path of the model when processing each historical task. This reasoning path includes the input data received by the model, the thought process at each step (such as which algorithms were used and what judgments were made), the data sources used (such as specific clause numbers in the clause library), and the rules followed. For example, when handling an insurance clause interpretation task, the recording model first identifies that the user is asking about the coverage of a specific disease, then extracts relevant clauses from the clause library for comparative analysis, and finally draws a conclusion.

[0053] The summary of experience abstraction and strategy units includes: Pattern extraction algorithm application: Pattern extraction algorithms are used to analyze recorded reasoning paths and identify common, reusable operational patterns. These algorithms can be based on rule matching, statistical analysis, or machine learning. For example, statistical analysis reveals that in multiple insurance clause interpretation tasks, the model employs operations such as "compare clauses to the database clause by clause" and "prioritize outputting coverage conditions," demonstrating high reusability. Strategy unit abstraction: The extracted operational patterns are abstracted into general strategy units. Each strategy unit defines clear inputs, outputs, and execution logic. For example, the "compare clauses to the database clause by clause" strategy unit takes the clause content to be interpreted and related disease information as input, outputs the coverage provisions related to the disease in the clause, and executes by comparing the clause content with the disease information clause by clause according to the clause number order, extracting the matching clause content. Strategy unit classification and organization: Based on the function and applicable scenarios, strategy units are classified and stored in the experience base. Classification can be based on task type (e.g., insurance clause interpretation, claims process consultation, etc.) and the role of the strategy unit (e.g., data extraction, logical judgment, result output, etc.). At the same time, detailed metadata is added to each strategy unit, including creation time, applicable scope, usage frequency, etc., to facilitate subsequent retrieval and management.

[0054] The similarity between the task parsing results and each strategy unit contained in the experience base is calculated to obtain the corresponding similarity calculation results.

[0055] In this embodiment, the feature similarity between the new task and each strategy unit in the experience base can be calculated based on the task features obtained from task parsing (i.e., the task parsing result). A vector space model-based method can be used, representing tasks and strategy units as vectors, and measuring the degree of similarity by calculating the cosine similarity between vectors. For example, the feature vector of the current insurance clause interpretation problem can be compared with the vectors of each strategy unit in the experience base to identify strategy units with high similarity.

[0056] Based on the similarity calculation result and the preset similarity threshold, a first strategy unit that meets the similarity condition with the task parsing result is retrieved from the experience base.

[0057] In this embodiment, a similarity threshold can be set according to business needs and actual application. When the similarity between a new task and a strategy unit exceeds this threshold, the strategy unit is considered relevant to the new task. The threshold setting needs to comprehensively consider the diversity of tasks and the accuracy of similarity, avoiding including strategy units with low similarity in the search results, while also ensuring that a sufficient number of relevant strategy units can be found. Furthermore, based on the set similarity threshold, the first strategy unit with high similarity to the new task (i.e., the similarity calculation result exceeds the threshold) is retrieved from the aforementioned experience base. These strategy units are then sorted in descending order of similarity to generate a search result list.

[0058] The first strategy unit is filtered based on a preset strategy unit filtering strategy to obtain the corresponding second strategy unit.

[0059] In this embodiment, the strategy content of the above-mentioned strategy unit screening strategy includes: Contextual information analysis: Analyzing the contextual information of the new task, including the user's historical consultation records, the context of the current dialogue, etc. For example, if the user has mentioned some information related to the current task in previous dialogues, such as the name of a specific insurance product or a disease condition, then when screening strategy units, priority is given to those strategy units that can match this contextual information. Business rule matching: Screening the retrieved strategy units according to the rules and processes of insurance business. For example, when handling claims process consultation tasks, only strategy units that comply with the insurance company's claims regulations are selected, excluding those strategy units that conflict with business rules. Comprehensive evaluation and ranking: Comprehensively evaluating the retrieved strategy units by comprehensively considering factors such as similarity, contextual information matching degree, and business rule matching degree. Based on the evaluation results, the strategy units are re-ranked, and the strategy unit most suitable for the current task is selected as the final experience invocation result. For example, if a strategy unit performs well in all three aspects of similarity, contextual information matching degree, and business rule matching degree, it is ranked at the top of the search results list.

[0060] Furthermore, the first strategy unit can be filtered based on the strategy content of the above strategy unit filtering strategy, and the filtered second strategy unit can be used as the final target strategy unit.

[0061] The second strategy unit is used as the target strategy unit.

[0062] Based on the above processing flow, this application calls a preset experience base; then calculates the similarity between the task parsing result and each strategy unit contained in the experience base, obtaining the corresponding similarity calculation result; subsequently, based on the similarity calculation result and a preset similarity threshold, it retrieves the first strategy unit from the experience base that meets the similarity condition with the task parsing result; subsequently, it filters the first strategy unit based on a preset strategy unit filtering strategy to obtain the corresponding second strategy unit; finally, it uses the second strategy unit as the target strategy unit. Experience invocation is a crucial step in experience transfer, establishing a connection between effective experience summarized from historical tasks and the current new task. This application provides rich material for subsequent transfer adaptation by retrieving relevant experience from the experience base. Without accurately invoking appropriate experience, subsequent prompt sequence generation and problem handling would lack a basis; therefore, experience invocation is an important link connecting historical experience and current task handling, which helps improve the accuracy of the generated answer data.

[0063] In some alternative implementations, step S204 includes the following steps: The target strategy unit is functionally analyzed to obtain the corresponding functional analysis results.

[0064] In this embodiment, the above-mentioned functional analysis process includes: performing detailed functional analysis on the invoked target strategy units; clarifying the input requirements, processing logic, and output format of each strategy unit; for example, for the "compare clauses to the database line by line" strategy unit, its input is the insurance clause database and the clause information to be interpreted, the processing logic is to compare clauses one by one according to the clause number order, and the output is the matching result after comparison; and analyzing the relationship between the input of the strategy unit and the context of the new question; and determining how to extract the input information required by the strategy unit from the context of the new question. Taking the "prioritize output coverage conditions" strategy unit as an example, it is necessary to clarify from the new question which insurance product, which disease, or which condition the coverage conditions are about, so as to determine the specific output content.

[0065] The contextual information is obtained by extracting the context from the question text.

[0066] In this embodiment, the aforementioned context extraction process includes: using natural language processing technology to extract key information from the question text. Through methods such as word segmentation, part-of-speech tagging, and named entity recognition, entities and concepts related to the strategy unit are identified. For example, for the question "Does this critical illness insurance cover diabetic complications?", key entities such as "critical illness insurance" and "diabetic complications" are extracted. Furthermore, the contextual semantics of new questions are further understood to grasp the overall intent and background of the question. Relevant information that the user may have mentioned previously, as well as the specific context of the current question, are considered. For example, if the user mentioned a specific insurance product name in the preceding text, this needs to be taken into account when extracting information.

[0067] Based on the functional analysis results and the contextual information, corresponding instruction information is generated.

[0068] In this embodiment, specific instructions are generated based on the obtained functional analysis results and contextual information. These instructions must clearly and explicitly guide the model on how to use the strategy unit to handle the problem. For example, for the "check clauses against the insurance policy library" strategy unit, the instruction "Find the relevant clauses regarding diabetic complications for this critical illness insurance from the insurance policy library" is generated.

[0069] The instruction information is organized and processed based on a preset organization strategy to generate a corresponding target prompt sequence.

[0070] In this embodiment, the organizational strategy includes organizing the generated instructions according to the execution order of the strategy units to form a complete prompt sequence. During the organization process, the logical coherence and operability of the prompt information are ensured. For example, combining the "compare clauses to the insurance clause library" and "prioritize outputting coverage conditions" strategy units, the generated prompt sequence is: "First, search the insurance clause library for relevant clauses regarding diabetic complications of the critical illness insurance; then, compare the clause content clause by clause to determine whether diabetic complications are explicitly covered; finally, prioritize outputting the conclusion regarding the coverage conditions." Furthermore, the instruction information can be organized and processed based on the organizational strategy to generate the corresponding target prompt sequence.

[0071] Based on the above processing flow, this application performs functional analysis on the target strategy unit to obtain the corresponding functional analysis results; then it extracts the context of the problem text to obtain the corresponding context information; then it generates corresponding instruction information based on the functional analysis results and context information; subsequently, it organizes and processes the instruction information based on a preset organization strategy, thereby automatically and accurately generating a matching target prompt sequence, effectively improving the accuracy and intelligence of the generated target prompt sequence.

[0072] In some alternative implementations, step S205 includes the following steps: Construct a corresponding strategy unit relationship diagram based on the target strategy unit.

[0073] In this embodiment, the construction process of the above-mentioned strategy unit relationship graph includes: Node definition: Each experience strategy unit is defined as a node in a graph neural network (GNN). Detailed attribute information is added to each node, including the functional description, input / output format, and applicable scenarios of the strategy unit. For example, for the "Data Integrity Check" strategy unit, its node attribute is defined as "Function: Check whether the claim data submitted by the user is complete; Input: Claim data submitted by the user; Output: Judgment result of data integrity". Edge definition: Based on the execution order and dependencies between strategy units, edges are defined between nodes. The direction of the edge indicates the execution order of the strategy units, and the weight of the edge indicates the strength of the dependency. For example, in the insurance claim review process, "Data Integrity Check" must be executed before "Term Matching", so a directed edge is defined from the "Data Integrity Check" node to the "Term Matching" node, and the weight of the edge can be set according to business rules and experience.

[0074] The problem text is subjected to feature extraction processing to obtain the corresponding feature vector.

[0075] In this embodiment, the feature extraction process includes text feature extraction and context feature extraction. Specifically, the text feature extraction includes: extracting features from the question text, including keyword extraction and topic recognition. Natural language processing technology is used to transform the new question into a quantifiable feature vector. For example, for the question "How do I apply for compensation for this accident insurance?", keywords such as "accident insurance" and "compensation application" are extracted as text features. The context feature extraction includes: considering the contextual information of the question text, such as the user's historical behavior and the source of the question. This contextual information is transformed into a feature vector and fused with the text feature vector to form a comprehensive feature vector for the new task. Furthermore, the obtained text features and context features are integrated to obtain the corresponding feature vector.

[0076] The feature vector is input into the policy unit relationship graph, and the policy unit relationship graph is inferred based on a preset graph neural network to obtain the score data of each policy unit node.

[0077] In this embodiment, the feature vector of the generated new task is input into the constructed policy unit relationship graph, and inference is performed using a graph neural network. Specifically, the graph neural network learns the dependencies between policy units and the influence of contextual features by aggregating information from nodes and edges, and calculates a score for each policy unit node to obtain the corresponding score data. The higher the score, the more important the policy unit is in the current new task.

[0078] Based on the score data, the corresponding optimal strategy path is selected and combined from the strategy unit relationship graph.

[0079] In this embodiment, the optimal strategy path is selected and combined from the strategy unit relationship graph based on the scores of the strategy unit nodes. Specifically, greedy algorithms, dynamic programming, or other methods can be used to determine the optimal path. For example, strategy unit nodes with higher scores are selected and connected sequentially according to the direction of the edges to form a complete strategy path.

[0080] The optimal strategy path is taken as the target strategy path.

[0081] In this embodiment, the system can also optimize the effect of experience combinations through reinforcement learning, enabling the model to gradually learn the optimal transfer method. Specifically, the effect of experience combinations is optimized through reinforcement learning algorithms. The process of the model processing a problem is viewed as an interaction between an agent and the environment in reinforcement learning. After each selection of a policy path and generation of a processing result, corresponding rewards or penalties are given based on the accuracy and effectiveness of the result. Through continuous trial and learning, the agent gradually adjusts the selection of policy paths, enabling the model to learn the optimal transfer method. For example, if the processing result generated by the model based on the current policy path receives correct feedback, a positive reward is given, prompting the model to be more inclined to choose this policy path in subsequent similar problems.

[0082] Based on the above processing flow, this application constructs a corresponding strategy unit relationship graph based on the target strategy unit; then, it performs feature extraction processing on the problem text to obtain the corresponding feature vector; subsequently, it inputs the feature vector into the strategy unit relationship graph and performs inference processing on the strategy unit relationship graph based on a preset graph neural network to obtain the score data of each strategy unit node; subsequently, based on the score data, it selects and combines the corresponding optimal strategy path from the strategy unit relationship graph; finally, it uses the optimal strategy path as the target strategy path. Thus, the transfer adaptation processing adopted in this application transforms abstract experience into actionable hints and strategy paths for the current specific problem, which is an important transformation process from abstract to concrete application of experience. It enables the model to flexibly apply experience to handle new tasks according to different problem contexts and characteristics, effectively improving the model's adaptability and processing capability. Without effective transfer adaptation, experience cannot be accurately applied to practical problems, resulting in poor model processing performance. Therefore, transfer adaptation is a key step to ensure that experience can effectively serve the current task processing, thereby improving the accuracy and processing effect of task processing.

[0083] In some optional implementations of this embodiment, step S206 includes the following steps: The target prompt sequence and the target policy path are input into the large language model.

[0084] In this embodiment, the selection of Large Language Models (LLMs) is not specifically limited and can be determined based on actual business needs, such as GPT or DeepSeek. The generated target prompt sequence is input into the LLM. For example, for an insurance claim review problem, the target prompt sequence "First, check if the claim documents are complete; if key documents are missing, prompt the user to supplement them; if the documents are complete, proceed to the clause matching stage, compare with the insurance clauses to determine if the claim complies with regulations; finally, calculate the compensation amount and generate the final answer" would serve as the initial instruction for the LLM to process the problem.

[0085] Based on the large language model, the target policy path is processed step by step according to the target prompt sequence to obtain multiple corresponding policy processing results.

[0086] In this embodiment, the aforementioned large language model executes step-by-step according to the target prompt sequence and the target strategy path. Taking "data integrity check → clause matching → compensation calculation" as an example, firstly, in the "data integrity check" step, the model analyzes the input claim data to determine if there is any missing key information, such as medical invoices or diagnostic certificates. If missing information is found, the model generates a response prompting the user to supplement the data according to the prompt sequence. When the data is complete, the model enters the "clause matching" stage. It meticulously compares the claim information with the insurance clauses, using its language understanding and logical reasoning abilities to determine whether the claim complies with regulations. Next, the "compensation amount calculation" is performed, where the model calculates the specific compensation amount based on the matching results and the compensation rules in the clauses. Thus, multiple corresponding strategy processing results are generated after each strategy step is executed.

[0087] Based on preset language generation rules, the processing results of all the strategies are integrated to obtain the corresponding integrated data.

[0088] In this embodiment, the large language model integrates the results of each step according to the rules of natural language generation (language generation rules), and uses the integrated data as the final answer data. For example, after completing the clause matching to determine that the claim complies with regulations, and calculating the specific amount of compensation, the large language model will integrate this information into a fluent and accurate sentence according to the rules of natural language generation, such as "According to the insurance terms, this claim complies with regulations, and the compensation amount is X yuan," as the final answer output.

[0089] The integrated data is used as the answer data.

[0090] Based on the above processing flow, this application inputs the target prompt sequence and target policy path into a large language model; then, based on the large language model, it executes the target policy path step by step according to the target prompt sequence, obtaining multiple corresponding policy processing results; subsequently, it integrates all policy processing results based on preset language generation rules to obtain corresponding integrated data; finally, it uses the integrated data as answer data. Thus, the entire transfer adaptation process, through steps such as prompt sequence generation, policy path selection and combination, and reinforcement learning optimization, transforms abstract experience into actionable steps for the specific problem, ultimately generating accurate and effective answer data.

[0091] In some optional implementations of this embodiment, step S207 includes the following steps: Call the preset clause library and regulation library.

[0092] In this embodiment, the aforementioned terms and conditions database can specifically be a general insurance terms and conditions database. The aforementioned regulatory database can be a general regulatory knowledge base.

[0093] The answer data is semantically validated based on the aforementioned terms library.

[0094] In this embodiment, semantic matching can be performed between the migration-generated answer data and the terms and conditions library. Specifically, semantic similarity calculation methods from natural language processing can be used to compare the answer data with the relevant terms and conditions text in the library and calculate their semantic similarity. If the similarity is higher than a set threshold, the answer data is deemed to have passed semantic verification; otherwise, it is deemed to have failed semantic verification, and consequently, knowledge verification. Furthermore, if the similarity is lower than the set threshold, it will be further marked as high-risk. For example, when semantically matching the answer generated by the large language model regarding whether critical illness insurance covers diabetic complications with the corresponding terms and conditions in the library is performed, if key information in the answer is found to be inconsistent with the terms and conditions (e.g., the answer states coverage while the terms and conditions explicitly state no coverage), the answer is marked as high-risk.

[0095] If the answer data passes semantic validation, then compliance validation is performed on the answer data based on the legal database.

[0096] In this embodiment, a regulatory knowledge base is introduced for secondary verification when compliance tasks are involved. The answer data is compared with relevant regulatory provisions in the knowledge base to check for compliance. For example, when handling compliance audits of insurance products, it is necessary to check not only whether the answer complies with the insurance terms but also whether it complies with the regulations of financial regulatory agencies regarding the sale and claims of insurance products. If the answer data is found to comply with regulations, it is determined that the answer data has passed the compliance verification; otherwise, the answer data has failed the compliance verification. Furthermore, if any discrepancies with regulations are found, timely corrections and prompts can be made.

[0097] If the answer data passes the compliance verification, it is determined that the answer data has passed the knowledge verification; otherwise, it is determined that the answer data has failed the knowledge verification.

[0098] In this embodiment, answer data is only considered to have passed knowledge verification if it simultaneously passes both semantic and knowledge verification; otherwise, it is considered to have failed knowledge verification. Additionally, a manual review interface is provided. When the system detects a risky or uncertain answer, it prompts for manual review. Manual reviewers can comprehensively examine and judge the system-generated answers based on their professional knowledge and experience. For example, for complex questions involving significant interests, or answers marked as high-risk by the system, manual reviewers can further verify relevant information to ensure the accuracy and reliability of the final output.

[0099] Based on the above processing flow, this application invokes a pre-defined clause library and regulatory library; then, it performs semantic verification on the answer data based on the clause library; if the answer data passes semantic verification, it performs compliance verification on the answer data based on the regulatory library; if the answer data passes compliance verification, it is determined that the answer data has passed knowledge verification; otherwise, it is determined that the answer data has failed knowledge verification. Thus, knowledge verification based on the clause library and regulatory library is the final guarantee for the entire business process. It ensures that the generated answer results comply with insurance clauses and financial regulations, effectively avoiding legal risks caused by illusions or erroneous transfers.

[0100] In some alternative implementations, the user information obtained is subject to user consent and complies with relevant laws and policies.

[0101] Furthermore, any software tools or components not belonging to our company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

[0102] Furthermore, compared to existing technologies, this application offers the following advantages: First, in terms of efficiency, through strategy unit abstraction and cross-task experience transfer, the system avoids manually designing prompts for each new task, significantly reducing labor costs. Second, in terms of adaptability, utilizing a dynamic experience combination mechanism, the system can autonomously and flexibly transfer historical experience to new tasks, quickly generating effective solutions even when facing new insurance products or regulations. Third, in terms of compliance and reliability, the knowledge fusion verification mechanism ensures that the output results are always consistent with insurance terms and financial regulations, avoiding legal risks caused by illusions or erroneous transfer. In summary, this solution, through a closed-loop design of "experience abstraction → autonomous transfer → knowledge verification," enables the insurance and financial large-scale model to possess the characteristics of low cost, high generalization, and strong compliance, providing stable and scalable technical support for the intelligent transformation of insurance institutions.

[0103] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0104] It should be emphasized that, to further ensure the privacy and security of the above answer data, the above product conversion data can also be stored in a blockchain node.

[0105] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0106] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0107] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0108] Further reference Figure 3 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of an artificial intelligence-based problem-solving device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0109] like Figure 3 As shown, the AI-based problem-solving device 300 described in this embodiment includes: a receiving module 301, a parsing module 302, a retrieval module 303, a fusion module 304, a processing module 305, a generation module 306, a verification module 307, and an output module 308. Wherein: The receiving module 301 is used to receive the question text input by the user; The parsing module 302 is used to perform task parsing on the question text based on a preset parsing strategy to obtain the corresponding task parsing result; The retrieval module 303 is used to retrieve target strategy units related to the task parsing results from a preset experience base; The fusion module 304 is used to fuse the target strategy unit and the question text to obtain a corresponding target prompt sequence; The processing module 305 is used to select and combine policy paths for the target policy unit based on a preset graph neural network to obtain the corresponding target policy path. The generation module 306 is used to perform execution processing on the target strategy path based on the target prompt sequence using a preset large language model to generate corresponding answer data; Verification module 307 is used to perform knowledge verification on the answer data; The output module 308 is used to output the answer data if the answer data passes the knowledge verification.

[0110] In some optional implementations of this embodiment, the parsing module 302 includes: The preprocessing submodule is used to preprocess the question text to obtain the corresponding target text; The first submodule is used to perform natural language understanding processing on the target text to obtain the corresponding understanding result; The matching submodule is used to perform matching processing based on the understanding results and the preset task type template to generate a task type corresponding to the target text. The second processing submodule is used to compare and map the task type with the tasks in the preset historical task database to obtain the corresponding processing result. The first determining submodule is used to take the processing result as the task parsing result corresponding to the question text.

[0111] In some optional implementations of this embodiment, the retrieval module 303 includes: The first submodule is used to call the preset experience library; The calculation submodule is used to calculate the similarity between the task parsing result and each strategy unit contained in the experience base, and obtain the corresponding similarity calculation result; The retrieval submodule is used to retrieve, based on the similarity calculation result and a preset similarity threshold, a first strategy unit that meets the similarity condition between the task parsing result and the experience base. The filtering submodule is used to filter the first strategy unit based on a preset strategy unit filtering strategy to obtain the corresponding second strategy unit. The second determining submodule is used to use the second strategy unit as the target strategy unit.

[0112] In some optional implementations of this embodiment, the fusion module 304 includes: The first analysis submodule is used to perform functional analysis on the target strategy unit and obtain the corresponding functional analysis results. The second processing submodule is used to perform context extraction processing on the question text to obtain the corresponding context information; A generation submodule is used to generate corresponding instruction information based on the functional analysis results and the contextual information; The organization submodule is used to organize and process the instruction information based on a preset organization strategy to generate a corresponding target prompt sequence.

[0113] In some optional implementations of this embodiment, the processing module 305 includes: A submodule is constructed to build a corresponding strategy unit relationship graph based on the target strategy unit; The extraction submodule is used to perform feature extraction processing on the problem text to obtain the corresponding feature vector; The inference submodule is used to input the feature vector into the strategy unit relationship graph and perform inference processing on the strategy unit relationship graph based on a preset graph neural network to obtain the score data of each strategy unit node. The third processing submodule is used to select and combine the corresponding optimal strategy paths from the strategy unit relationship graph based on the score data; The third determining submodule is used to select the optimal strategy path as the target strategy path.

[0114] In some optional implementations of this embodiment, the generation module 306 includes: An input submodule is used to input the target prompt sequence and the target policy path into the large language model; The execution submodule is used to perform step-by-step execution processing on the target strategy path based on the large language model and the target prompt sequence to obtain multiple corresponding strategy processing results. The integration submodule is used to integrate the processing results of all the strategies based on preset language generation rules to obtain the corresponding integrated data; The fourth determining submodule is used to use the integrated data as the answer data.

[0115] In some optional implementations of this embodiment, the verification module 307 includes: The second submodule is used to call the preset clause library and regulation library; The first verification submodule is used to perform semantic verification on the answer data based on the terms library; The second verification submodule is used to perform compliance verification on the answer data based on the legal database if the answer data passes semantic verification. The determination submodule is used to determine that the answer data passes the knowledge verification if the answer data passes the compliance verification, and otherwise determine that the answer data fails the knowledge verification.

[0116] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0117] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with components 41-43 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0118] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0119] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may also include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions for problem-solving methods based on artificial intelligence. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0120] In some embodiments, the processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions stored in the memory 41 or to process data, for example, to execute computer-readable instructions of the artificial intelligence-based problem-solving method.

[0121] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0122] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the artificial intelligence-based problem-solving method described above.

[0123] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0124] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A problem-solving method based on artificial intelligence, characterized in that, Includes the following steps: Receive user-input question text; The problem text is parsed based on a preset parsing strategy to obtain the corresponding parsing results; Retrieve target strategy units related to the task parsing results from a pre-set experience base; The target strategy unit and the question text are fused together to obtain the corresponding target prompt sequence; Based on a preset graph neural network, the target policy unit is processed to select and combine policy paths to obtain the corresponding target policy path; Based on the target prompt sequence, the target strategy path is processed using a preset large language model to generate corresponding answer data; Perform knowledge verification on the answer data; If the answer data passes the knowledge verification, then the answer data will be output.

2. The problem-solving method based on artificial intelligence according to claim 1, characterized in that, The step of performing task parsing on the question text based on a preset parsing strategy to obtain the corresponding task parsing result specifically includes: The question text is preprocessed to obtain the corresponding target text; The target text is processed using natural language understanding to obtain the corresponding understanding results; Based on the understanding results, a matching process is performed with a preset task type template to generate a task type corresponding to the target text; The task type is compared and mapped with tasks in a preset historical task database to obtain the corresponding processing result. The processing result is used as the task parsing result corresponding to the question text.

3. The problem-solving method based on artificial intelligence according to claim 1, characterized in that, The step of retrieving the target strategy unit related to the task parsing result from the preset experience base specifically includes: Call the preset experience library; The similarity between the task parsing result and each strategy unit contained in the experience base is calculated to obtain the corresponding similarity calculation result; Based on the similarity calculation result and the preset similarity threshold, a first strategy unit that meets the similarity condition with the task parsing result is retrieved from the experience base; The first strategy unit is filtered based on a preset strategy unit filtering strategy to obtain the corresponding second strategy unit; The second strategy unit is used as the target strategy unit.

4. The problem-solving method based on artificial intelligence according to claim 1, characterized in that, The step of fusing the target strategy unit and the question text to obtain the corresponding target prompt sequence specifically includes: Perform functional analysis on the target strategy unit to obtain the corresponding functional analysis results; The context extraction process is performed on the question text to obtain the corresponding context information; Based on the functional analysis results and the contextual information, corresponding instruction information is generated; The instruction information is organized and processed based on a preset organization strategy to generate a corresponding target prompt sequence.

5. The problem-solving method based on artificial intelligence according to claim 1, characterized in that, The step of selecting and combining policy paths for the target policy unit based on a preset graph neural network to obtain the corresponding target policy path specifically includes: Construct a corresponding strategy unit relationship diagram based on the target strategy unit; The problem text is subjected to feature extraction processing to obtain the corresponding feature vector; The feature vector is input into the strategy unit relationship graph, and the strategy unit relationship graph is inferred based on a preset graph neural network to obtain the score data of each strategy unit node. Based on the score data, select and combine the corresponding optimal strategy paths from the strategy unit relationship graph; The optimal strategy path is taken as the target strategy path.

6. The problem-solving method based on artificial intelligence according to claim 1, characterized in that, The step of performing execution processing on the target strategy path using a preset large language model based on the target prompt sequence to generate corresponding answer data specifically includes: The target prompt sequence and the target policy path are input into the large language model; Based on the large language model, the target strategy path is executed step by step according to the target prompt sequence to obtain multiple corresponding strategy processing results; Based on preset language generation rules, the processing results of all the strategies are integrated to obtain the corresponding integrated data. The integrated data is used as the answer data.

7. The problem-solving method based on artificial intelligence according to claim 1, characterized in that, The step of performing knowledge verification on the answer data specifically includes: Call the preset clause library and regulation library; The answer data is semantically validated based on the aforementioned terms and conditions database. If the answer data passes semantic validation, then the answer data undergoes compliance validation based on the legal database. If the answer data passes the compliance verification, it is determined that the answer data has passed the knowledge verification; otherwise, it is determined that the answer data has failed the knowledge verification.

8. A problem-solving device based on artificial intelligence, characterized in that, include: The receiving module is used to receive the question text input by the user; The parsing module is used to perform task parsing on the question text based on a preset parsing strategy to obtain the corresponding task parsing results; The retrieval module is used to retrieve target strategy units related to the task parsing results from a preset experience base; The fusion module is used to fuse the target strategy unit and the question text to obtain a corresponding target prompt sequence; The processing module is used to select and combine policy paths for the target policy unit based on a preset graph neural network to obtain the corresponding target policy path. The generation module is used to process the target strategy path based on the target prompt sequence using a preset large language model to generate corresponding answer data; The verification module is used to perform knowledge verification on the answer data; The output module is used to output the answer data if the answer data passes the knowledge verification.

9. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the problem-solving method based on artificial intelligence as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the problem-solving method based on artificial intelligence as described in any one of claims 1 to 7.

Images