Complex finance and taxation problem retrieval method and device based on large model agent

Abstract

The invention discloses a finance and taxation complex problem retrieval method and device based on a large model agent. The method comprises the following steps of: performing session context supplementation on a user question retrieved by a user according to a questioning habit of the retrieved user and a preset question mode, obtaining a supplementary question description, disassembling the supplementary question description to obtain a plurality of sub-question descriptions of the supplementary question description, dynamically allocating weights for the plurality of sub-question descriptions, and determining the priority of each sub-question description; performing deep retrieval on the sub-problem descriptions according to the priorities, obtaining knowledge fragments of the sub-problem descriptions and the correlation of the corresponding sub-problem descriptions, obtaining a correlation result, regenerating accurate retrieval words of the sub-problem descriptions according to the correlation result, performing retrieval again, and obtaining final knowledge fragments of retrieval; and according to the question type, the knowledge background and the question style of the user question, carrying out personalized fusion on the final knowledge fragment to obtain a question answer of the user question and returning the question answer to the user front end.

Description

Technical Field

[0001] The present invention relates to the technical field of financial and taxation problem retrieval, and more specifically, to a method and device for retrieval of complex financial and taxation problems based on a large model intelligent agent. Background Art

[0002] In today's business environment, financial and tax issues are becoming increasingly complex and diverse. With the development of the economy and the continuous adjustment and updating of tax policies, enterprises and individuals are facing huge challenges in handling financial and tax affairs.

[0003] Traditional tax processing methods often rely on manual experience and limited knowledge base, which is inefficient and prone to errors. It is difficult for humans to fully and accurately understand and analyze complex tax issues, resulting in wrong decisions or non-compliance with regulatory requirements, which may cause economic losses and legal risks to enterprises and individuals.

[0004] In terms of knowledge acquisition and application, existing technologies are insufficient in understanding the context of user questions, deeply breaking down complex problems, and accurately judging the relevance of knowledge fragments. This makes the acquired knowledge inaccurate and incomplete, affecting the final processing effect.

[0005] At the same time, in terms of learning and applying existing experience, there is a lack of effective methods to use the answers to similar problems to improve the ability to handle new problems. It is impossible to quickly extract valuable information from a large amount of historical data to guide reasoning and answers to new questions.

[0006] In summary, existing technologies have many limitations when dealing with complex issues in the financial and tax fields, and there is an urgent need for an innovative and comprehensive solution to improve the efficiency, accuracy and adaptability of financial and tax processing. Summary of the invention

[0007] In view of the deficiencies in the prior art, the present invention provides a method and device for retrieving complex financial and taxation issues based on a large model intelligent agent.

[0008] According to one aspect of the present invention, a method for retrieving complex financial and taxation issues based on a large model agent is provided, comprising:

[0009] Supplement the conversation context of the user's question according to the questioning habits of the searching user and the preset question pattern, and obtain the supplementary question description;

[0010] The supplementary problem description is decomposed to obtain multiple sub-problem descriptions of the supplementary problem description, and weights are dynamically assigned to the multiple sub-problem descriptions according to the real-time importance of each sub-problem description and the user's specific demand scenario, so as to determine the priority of each sub-problem description;

[0011] Perform a deep search on each sub-problem description according to the priority, obtain the knowledge fragments of each sub-problem description, and judge the relevance between the knowledge fragments and the corresponding sub-problem descriptions based on the semantic understanding technology of BGE-Reranker to obtain the relevance results. Regenerate the precise search terms of each sub-problem description according to the relevance results and re-search to obtain the final knowledge fragments retrieved;

[0012] The final knowledge fragments are personalized and integrated according to the user's question type, knowledge background and questioning style, and the answer to the user's question is obtained and returned to the user front end.

[0013] Optionally, perform a deep search on each sub-problem description according to the priority to obtain knowledge fragments of each sub-problem description, including:

[0014] Determine the index library for retrieval by combining the time of the retrieval user's question, the source channel of the question, and the multimodal information of the relevance of historical questions;

[0015] Dynamically adjust the index library based on the source channels of user questions to obtain the latest index library;

[0016] According to the priority, the description of each sub-problem is deeply searched in the latest index library to obtain the knowledge fragments of each sub-problem description.

[0017] Optionally, a deep search is performed on the description of each sub-problem in the latest index library according to the priority to obtain the knowledge fragments of the description of each sub-problem, including:

[0018] Knowledge graph enhancement technology is used to perform structured representation and association analysis of knowledge in the financial and taxation fields based on grid problem descriptions, and to obtain knowledge fragments describing each sub-problem.

[0019] Optionally, the final knowledge fragments are personalized and integrated according to the question type, knowledge background, and questioning style of the user's question to obtain the answer to the user's question, including:

[0020] Using the big financial and tax model, we can infer the key points and potential needs of the user's problem based on the specific asset characteristics, disposal methods, and financial status factors in the user's problem;

[0021] Based on the key points and potential needs of users' questions, combined with real-time changes in fiscal and tax policies and market dynamics, the final knowledge fragments are personalized and integrated to generate answers to questions.

[0022] According to another aspect of the present invention, a device for retrieving complex financial and taxation issues based on a large model agent is provided, comprising:

[0023] A supplementation module is used to supplement the conversation context of the user question retrieved by the user according to the questioning habits of the retrieval user and the preset question mode, and obtain a supplementary question description;

[0024] A weight allocation module is used to decompose the supplementary problem description to obtain multiple sub-problem descriptions of the supplementary problem description, and dynamically allocate weights to the multiple sub-problem descriptions according to the real-time importance of each sub-problem description and the user's specific demand scenario, so as to determine the priority of each sub-problem description;

[0025] The generation module is used to perform in-depth search on each sub-problem description according to the priority, obtain the knowledge fragments of each sub-problem description, and judge the relevance between the knowledge fragments and the corresponding sub-problem descriptions based on the semantic understanding technology of BGE-Reranker to obtain the relevance results. According to the relevance results, the precise search terms of each sub-problem description are re-generated and searched again to obtain the final knowledge fragments retrieved;

[0026] The fusion module is used to personalize the final knowledge fragments according to the question type, knowledge background and questioning style of the user's question, obtain the answer to the user's question and return it to the user front end.

[0027] Optionally, the generation module performs a deep search on each sub-problem description according to the priority to obtain knowledge fragments of each sub-problem description, including:

[0028] A determination submodule is used to determine the index library for retrieval by combining the time of the question asked by the retrieval user, the source channel of the question, and the multimodal information of the relevance of the historical questions;

[0029] The adjustment submodule is used to dynamically adjust the index library according to the source channel of the user's questions and obtain the latest index library;

[0030] The retrieval submodule is used to perform in-depth retrieval of each sub-problem description in the latest index library according to priority, and obtain the knowledge fragments of each sub-problem description.

[0031] Optionally, the retrieval submodule includes:

[0032] The acquisition unit is used to use the knowledge graph enhancement technology to perform structured representation and association analysis on the knowledge in the financial and taxation field according to the grid problem description, and obtain the knowledge fragments described in each sub-problem.

[0033] Optionally, the fusion module includes:

[0034] The reasoning submodule is used to use the financial and taxation big model to infer the key points and potential needs of the user's problem based on the specific asset characteristics, disposal methods, and financial status factors in the user's problem;

[0035] The generation sub-module is used to personalize the final knowledge fragments and generate answers to questions based on the key points and potential needs of user questions, combined with real-time changes in fiscal and tax policies and market dynamics.

[0036] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and the computer program is used to execute the method described in any one of the above aspects of the present invention.

[0037] According to another aspect of the present invention, an electronic device is provided, comprising: a processor; a memory for storing instructions executable by the processor; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the method described in any one of the above aspects of the present invention.

[0038] Beneficial effects of the present invention:

[0039] Improve answer accuracy: By supplementing conversation context, breaking down questions and optimizing retrieval, judging the relevance of fragments, and other measures, we ensure that the big model can rely on accurate knowledge fragments when answering complex financial and taxation questions, thereby reducing the occurrence of "hallucination" phenomena.

[0040] Enhanced reasoning ability: The RAG-based Few Shot learning method enables large models to better master reasoning skills and improve their ability to answer specific types of questions by learning the solution process of similar questions.

[0041] Improve user experience: The implementation of the present invention enables complex problems in the financial and tax fields to be answered quickly and accurately, thereby improving user work efficiency and satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings:

[0043] Figure 1 It is a flowchart of a method for retrieving complex financial and taxation issues based on a large model agent provided by an exemplary embodiment of the present invention;

[0044] Figure 2 It is another flow chart of a method for retrieving complex financial and taxation issues based on a large model agent provided by an exemplary embodiment of the present invention;

[0045] Figure 3 It is a schematic diagram of the structure of a complex taxation problem retrieval device based on a large model agent provided by an exemplary embodiment of the present invention;

[0046] Figure 4 This is a structure of an electronic device provided by an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0047] Below, the exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention, and it should be understood that the present invention is not limited to the exemplary embodiments described here.

[0048] It should be noted that the relative arrangement of components and steps, the numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless specifically stated otherwise.

[0049] Those skilled in the art can understand that the terms "first" and "second" in the embodiments of the present invention are only used to distinguish different steps, devices or modules, etc., and neither represent any specific technical meaning nor indicate the necessary logical order between them.

[0050] It should also be understood that, in the embodiments of the present invention, “plurality” may refer to two or more than two, and “at least one” may refer to one, two or more than two.

[0051] It should also be understood that any component, data or structure mentioned in the embodiments of the present invention can generally be understood as one or more, unless explicitly limited or otherwise indicated in the context.

[0052] In addition, the term "and / or" in the present invention is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present invention generally indicates that the associated objects before and after are in an "or" relationship.

[0053] It should also be understood that the description of the various embodiments of the present invention focuses on the differences between the various embodiments, and the same or similar aspects thereof can be referenced to each other, and for the sake of brevity, they will not be described one by one.

[0054] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0055] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0056] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0057] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0058] Embodiments of the present invention can be applied to electronic devices such as terminal devices, computer systems, servers, etc., which can operate with many other general or special computing system environments or configurations. Examples of well-known terminal devices, computing systems, environments and / or configurations suitable for use with electronic devices such as terminal devices, computer systems, servers, etc. include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network personal computers, small computer systems, large computer systems, and distributed cloud computing technology environments including any of the above systems, etc.

[0059] Electronic devices such as terminal devices, computer systems, servers, etc. can be described in the general context of computer system executable instructions (such as program modules) executed by computer systems. Generally, program modules can include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers can be implemented in a distributed cloud computing environment, where tasks are performed by remote processing devices linked through a communication network. In a distributed cloud computing environment, program modules can be located on local or remote computing system storage media including storage devices.

[0060] Exemplary Methods

[0061] Figure 1 FIG. 1 is a flow chart of a method for retrieving complex taxation issues based on a large model agent provided by an exemplary embodiment of the present invention. This embodiment can be applied to electronic devices, such as Figure 1 As shown, the financial and taxation complex problem retrieval method 100 based on the large model agent includes the following steps:

[0062] Step 101, supplementing the user question retrieved by the user with a conversation context according to the questioning habits of the searching user and a preset question pattern, and obtaining a supplementary question description;

[0063] Step 102, decomposing the supplementary problem description to obtain multiple sub-problem descriptions of the supplementary problem description, and dynamically assigning weights to the multiple sub-problem descriptions according to the real-time importance of each sub-problem description and the user's specific demand scenario, and determining the priority of each sub-problem description;

[0064] Step 103, perform a deep search on each sub-problem description according to the priority, obtain the knowledge fragment of each sub-problem description, and judge the correlation between the knowledge fragment and the corresponding sub-problem description based on the semantic understanding technology of BGE-Reranker to obtain the correlation result, regenerate the precise search terms of each sub-problem description according to the correlation result, and re-search to obtain the final knowledge fragment retrieved;

[0065] Step 104 , the final knowledge fragments are personalized and integrated according to the question type, knowledge background and questioning style of the user's question, and the answer to the user's question is obtained and returned to the user front end.

[0066] Specifically, the present invention has significant innovation and practicality in the field of finance and taxation, and fully utilizes memory ability, planning ability, thinking ability and retrieval generation ability. It is mainly composed of two key modules: knowledge fragment acquisition optimization and RAG-based Few Shot learning.

[0067] 1. Innovations and detailed description of knowledge fragment acquisition optimization

[0068] (1) Conversation context supplement

[0069] In the process of searching for financial and tax issues, the innovation of the present invention is that it can not only supplement the context information of this session, but also further predict and supplement the possibly relevant information by intelligently analyzing the user's questioning habits and question patterns. For example, when a user asks "Is this year's corporate income tax preferential policy applicable to small technology companies like us?", in addition to considering the company's specific financial status, industry classification and other key information mentioned by the user before, the system will also predict that the user may be interested in specific corporate income tax preferential terms based on the user's previous focus on similar issues, such as the inquiry history of the policy of additional deduction of R&D expenses. In this way, when supplementing the context, the user's potential needs will be more comprehensively considered to form a more targeted and accurate problem description, such as "Our company is a small technology company with an annual turnover of about 5 million yuan. We have paid attention to the policy of additional deduction of R&D expenses before. Is this year's corporate income tax preferential policy applicable to our situation?" Through this innovative context supplement method, the accuracy and effectiveness of the search can be greatly improved, providing users with more accurate answers.

[0070] (2) Problem analysis and retrieval

[0071] The innovation of the present invention in problem decomposition and retrieval lies in the introduction of a dynamic weight allocation mechanism. Taking "how to formulate a reasonable annual tax plan for an enterprise, taking into account the newly introduced tax incentives and the company's business expansion plan" as an example, after the large model decomposes it into multiple small questions such as "what are the newly introduced tax incentives", "what are the factors affecting the taxation of the company's business expansion plan", and "what are the basic steps and key points for formulating the annual tax plan of the enterprise", it does not simply perform parallel or sequential retrieval, but dynamically allocates weights based on the real-time importance of the problem and the user's specific demand scenarios. For example, if a series of major tax incentive policies have been newly introduced in the current user's area, then the weight of the question "what are the newly introduced tax incentives" will be increased accordingly, and in-depth retrieval will be prioritized to ensure that the most critical information is provided to the user at the first time. At the same time, in the process of parallel retrieval and sequential retrieval, adjustments are constantly made based on the feedback of the retrieval results to achieve the most optimized retrieval strategy and obtain relevant knowledge fragments comprehensively and efficiently.

[0072] (3) Fragment relevance judgment and re-examination

[0073] In the segment relevance judgment and re-examination link, the present invention innovatively introduces the semantic understanding technology based on BGE-Reranker. For the retrieved knowledge segments, the big model can not only judge the relevance from the surface keywords, but also calculate the semantic relevance between the question and the segment through BGE-Reranker. For example, when the retrieved segment is about the provisions of personal income tax, and the user's question is mainly about corporate income tax, the system will not only simply calculate the relevance, but also further analyze the potential connection between the personal income tax provisions and corporate income tax issues. If certain policy principles or calculation methods are found to have similarities, the system will intelligently extract these common information, and conduct a secondary search in combination with the characteristics of corporate income tax, and automatically generate more accurate search terms, such as "corporate income tax planning related knowledge segments similar to personal income tax principles." In this innovative way, the retrieval process is repeated until several knowledge segments that are highly relevant to the user's question and have deep value are obtained, providing a solid foundation for subsequent accurate answers.

[0074] (4) Segment splicing and answering

[0075] In terms of fragment splicing and answering, the innovation of the present invention lies in the personalized knowledge fusion strategy. When the most relevant knowledge fragments are spliced ​​to the front end of the user's question, it is not a simple sequential splicing, but a personalized fusion based on the user's question type, knowledge background and questioning style. Suppose the user asks "What is the process of corporate value-added tax deduction", the relevant fragments obtained include "basic conditions for value-added tax deduction", "common deductible items", "requirements for deduction certificates", etc. The system will first analyze the user's professional level. If the user is a financial and tax professional, the spliced ​​answer will be more concise and clear, highlighting the key points; if the user is a beginner, the answer will be more detailed, with specific cases and explanations. At the same time, the system will also adjust the language expression of the answer according to the user's questioning style to make it more in line with the user's reading habits, thereby ensuring that the answer is highly accurate and fully justified, and also more in line with the user's personalized needs.

[0076] 2. Innovation and detailed description of RAG-based Few Shot learning

[0077] (1) Question type identification and index library specification

[0078] The innovation of the present invention in terms of question type identification and index library designation lies in the use of multimodal information fusion technology. When a user asks a question, the system not only relies on the text content to identify the question type, but also combines the user's question time, question source channel, historical question correlation and other multimodal information for comprehensive judgment. For example, if a user asks "How to make accounting entries for depreciation of fixed assets" during a specific tax declaration period, the system will take into account the time factor, more accurately identify it as an accounting entry type question, and specify the corresponding index library. At the same time, according to the different source channels of the questions, such as online consultations from corporate financial software, professional finance and tax forums, etc., the index library is dynamically adjusted to ensure that the collected question and answer pairs are more in line with the user's actual demand scenarios. The index library collects a large number of similar accounting entry-related question and answer pairs, providing rich materials for FewShot learning.

[0079] (2) Retrieval and splicing of similar question-answer pairs

[0080] In the retrieval and splicing of similar question and answer pairs, the present invention innovatively introduces knowledge graph enhancement technology. When retrieving several question and answer pairs that are most similar to the user's question from the specified index library, it does not simply rely on the traditional keyword matching algorithm, but uses the knowledge graph to perform structured representation and association analysis on the knowledge in the financial and taxation fields. For example, for the above-mentioned problem of fixed asset depreciation, the system will analyze the complex relationship between different asset types, depreciation calculation methods, and accounting entries through the knowledge graph, and retrieve similar question and answer pairs such as "depreciation calculation methods and accounting entries for different asset types" and "accounting entry processing when the depreciation period changes". At the same time, when splicing these question and answer pairs, they will be arranged in order according to the relationship path in the knowledge graph to more clearly show the ideas and methods in the answering process, and provide intuitive learning examples for the large model.

[0081] (3) Model learning and reasoning

[0082] In terms of model learning and reasoning, the innovation of the present invention lies in the adaptive learning strategy. On the basis of learning these similar question-answer pairs, the large model does not simply imitate the logic and rules therein, but can make adaptive adjustments according to the uniqueness and complexity of the user's questions. Taking "Accounting entries for the disposal of fixed assets by enterprises" as an example, after learning relevant similar question-answer pairs, the large model will analyze the specific asset characteristics, disposal methods, financial status and other factors in the user's questions to infer the key points and potential needs of the user's questions. At the same time, combined with real-time changes in fiscal and taxation policies and market dynamics, an accurate, detailed and actual final answer is generated, such as "When an enterprise disposes of fixed assets, it should follow the following steps to make accounting entries... At the same time, considering the current tax preferential policies and market conditions, there may be the following special circumstances that need attention..." Through this adaptive learning and reasoning strategy, the quality and efficiency of the model's answers to specific types of questions are effectively improved, providing users with more professional and practical fiscal and taxation solutions.

[0083] The key points of the present invention are:

[0084] Key point 1: The knowledge fragment acquisition optimization module uses memory capabilities to supplement the conversation context and improve the accuracy of retrieval;

[0085] Key point 2: The knowledge fragment acquisition optimization module uses planning capabilities to realize the decomposition and retrieval of complex problems. It decomposes a complex problem into a series of simple tasks that are easy to retrieve, and further plans the execution process of these simple tasks;

[0086] Key point 3: The knowledge fragment acquisition optimization module uses thinking ability to realize fragment relevance judgment and re-examination, fragment splicing and answering. A series of simple search results are re-thought, judged and spliced ​​to ensure the accuracy and basis of the answers.

[0087] Key point 4: The question type identification and index library specification function in the RAG-based Few Shot learning module introduces a specified knowledge base to restrict the model to execute instructions within a reference range, thereby effectively avoiding or reducing the model hallucination phenomenon.

[0088] Key point 5: Retrieval and concatenation of similar question-answer pairs uses relevant information retrieved from a large-scale knowledge base to provide additional contextual support information for Few Shot learning, significantly improving the accuracy and relevance of answers generated by large models.

[0089] Beneficial effects of the present invention:

[0090] Improve answer accuracy: By supplementing conversation context, breaking down questions and optimizing retrieval, judging the relevance of fragments, and other measures, we ensure that the big model can rely on accurate knowledge fragments when answering complex financial and taxation questions, thereby reducing the occurrence of "hallucination" phenomena.

[0091] Enhanced reasoning ability: The RAG-based Few Shot learning method enables large models to better master reasoning skills and improve their ability to answer specific types of questions by learning the solution process of similar questions.

[0092] Improve user experience: The implementation of the present invention enables complex problems in the financial and tax fields to be answered quickly and accurately, thereby improving user work efficiency and satisfaction.

[0093] Exemplary Devices

[0094] Figure 3 FIG. 1 is a schematic diagram of a structure of a complex taxation problem retrieval device based on a large model agent provided by an exemplary embodiment of the present invention. Figure 3 As shown, the device 300 includes:

[0095] The supplementation module 310 is used to supplement the user question retrieved by the user according to the questioning habits of the searching user and the preset question mode, and obtain the supplementary question description;

[0096] The weight allocation module 320 is used to decompose the supplementary problem description to obtain multiple sub-problem descriptions of the supplementary problem description, and dynamically allocate weights to the multiple sub-problem descriptions according to the real-time importance of each sub-problem description and the user's specific demand scenario, and determine the priority of each sub-problem description;

[0097] A generation module 330 is used to perform a deep search on each sub-problem description according to the priority, obtain the knowledge fragment of each sub-problem description, and determine the correlation between the knowledge fragment and the corresponding sub-problem description based on the semantic understanding technology of BGE-Reranker to obtain the correlation result, and regenerate the precise search terms of each sub-problem description according to the correlation result to re-search and obtain the final knowledge fragment retrieved;

[0098] The fusion module 340 is used to perform personalized fusion of the final knowledge fragments according to the question type, knowledge background and questioning style of the user's question, obtain the answer to the user's question and return it to the user front end.

[0099] Optionally, the generation module 330 performs a deep search on each sub-problem description according to the priority to obtain a knowledge fragment of each sub-problem description, including:

[0100] A determination submodule is used to determine the index library for retrieval by combining the time of the question asked by the retrieval user, the source channel of the question, and the multimodal information of the relevance of the historical questions;

[0101] The adjustment submodule is used to dynamically adjust the index library according to the source channel of the user's questions and obtain the latest index library;

[0102] The retrieval submodule is used to perform in-depth retrieval of each sub-problem description in the latest index library according to priority, and obtain the knowledge fragments of each sub-problem description.

[0103] Optionally, the retrieval submodule includes:

[0104] The acquisition unit is used to use the knowledge graph enhancement technology to perform structured representation and association analysis on the knowledge in the financial and taxation field according to the grid problem description, and obtain the knowledge fragments described in each sub-problem.

[0105] Optionally, the fusion module 340 includes:

[0106] The reasoning submodule is used to use the financial and taxation big model to infer the key points and potential needs of the user's problem based on the specific asset characteristics, disposal methods, and financial status factors in the user's problem;

[0107] The generation sub-module is used to personalize the final knowledge fragments and generate answers to questions based on the key points and potential needs of user questions, combined with real-time changes in fiscal and tax policies and market dynamics.

[0108] Exemplary Electronic Devices

[0109] Figure 4 This is a structure of an electronic device provided by an exemplary embodiment of the present invention. Figure 4 As shown, the electronic device 40 includes one or more processors 41 and a memory 42 .

[0110] The processor 41 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0111] The memory 42 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 41 may run the program instructions to implement the methods of the software programs of the various embodiments of the present invention described above and / or other desired functions. In one example, the electronic device may also include: an input device 43 and an output device 44, which are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0112] In addition, the input device 43 may also include, for example, a keyboard, a mouse, etc.

[0113] The output device 44 can output various information to the outside, and can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto.

[0114] Of course, to simplify, Figure 4 Only some of the components related to the present invention in the electronic device are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, the electronic device may further include any other appropriate components according to specific application conditions.

[0115] Exemplary computer program products and computer-readable storage media

[0116] In addition to the above-mentioned methods and devices, an embodiment of the present invention may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the method according to various embodiments of the present invention described in the above-mentioned "Exemplary Method" section of this specification.

[0117] The computer program product may be written in any combination of one or more programming languages ​​to write program code for performing the operations of the embodiments of the present invention, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0118] In addition, an embodiment of the present invention may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, enable the processor to execute the steps of the method according to various embodiments of the present invention described in the above “Exemplary Method” section of this specification.

[0119] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, system or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0120] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details disclosed above are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0121] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0122] The block diagrams of the devices, systems, equipment, and systems involved in the present invention are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, systems, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The word "such as" used here refers to the phrase "such as but not limited to", and can be used interchangeably with it.

[0123] The method and system of the present invention may be implemented in many ways. For example, the method and system of the present invention may be implemented by software, hardware, firmware or any combination of software, hardware, firmware. The above order of steps for the method is only for illustration, and the steps of the method of the present invention are not limited to the order specifically described above, unless otherwise specifically stated. In addition, in some embodiments, the present invention may also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present invention. Thus, the present invention also covers a recording medium storing a program for executing the method according to the present invention.

[0124] It should also be noted that in the system, device and method of the present invention, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. The above description of the disclosed aspects is provided to enable any technician in the field to make or use the present invention. Various modifications to these aspects are very obvious to those skilled in the art, and the general principles defined here can be applied to other aspects without departing from the scope of the present invention. Therefore, the present invention is not intended to be limited to the aspects shown here, but in accordance with the widest range consistent with the principles and novel features disclosed here.

[0125] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present invention to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A complex taxation problem retrieval method based on a large model agent, characterized in that: include: Supplement the conversation context of the user's question according to the questioning habits of the searching user and the preset question pattern, and obtain the supplementary question description; Decomposing the supplementary problem description to obtain multiple sub-problem descriptions of the supplementary problem description, and dynamically assigning weights to the multiple sub-problem descriptions according to the real-time importance of each sub-problem description and a user-specific demand scenario, and determining the priority of each sub-problem description; Perform a deep search on each sub-problem description according to the priority, obtain the knowledge fragment of each sub-problem description, and judge the correlation between the knowledge fragment and the corresponding sub-problem description based on the semantic understanding technology of BGE-Reranker to obtain the correlation result, regenerate the precise search terms of each sub-problem description according to the correlation result, and re-search to obtain the final knowledge fragment retrieved; The final knowledge fragments are personalized and integrated according to the question type, knowledge background and questioning style of the user's question, and the answer to the user's question is obtained and returned to the user front end.

2. The method according to claim 1, characterized in that: Perform a deep search on each sub-problem description according to the priority to obtain knowledge fragments of each sub-problem description, including: Determine the index library for retrieval by combining the time of the retrieval user's question, the source channel of the question, and the multimodal information of the relevance of historical questions; Dynamically adjust the index library according to the source channel of the user question to obtain the latest index library; A deep search is performed on each sub-problem description in the latest index library according to the priority to obtain the knowledge fragment of each sub-problem description.

3. The method according to claim 2, characterized in that Performing a deep search on each sub-problem description in the latest index library according to the priority to obtain the knowledge fragment of each sub-problem description includes: The knowledge graph enhancement technology is used to perform structured representation and association analysis on the knowledge in the finance and taxation field according to the grid problem description, and the knowledge fragments described in each sub-problem are obtained.

4. The method according to claim 1, characterized in that: The final knowledge fragments are personalized and integrated according to the question type, knowledge background and questioning style of the user's question to obtain the answer to the user's question, including: Using the big financial and tax model, we can infer the key points and potential needs of the user's problem based on the specific asset characteristics, disposal methods, and financial status factors in the user's problem; According to the key points of the user's question and the potential needs, combined with real-time fiscal and tax policy changes and market dynamics, the final knowledge fragment is personalized and integrated to generate the answer to the question.

5. A complex taxation problem retrieval device based on a large model agent, characterized in that: include: A supplementation module is used to supplement the conversation context of the user question retrieved by the user according to the questioning habits of the retrieval user and the preset question mode, and obtain a supplementary question description; A weight allocation module is used to decompose the supplementary problem description to obtain multiple sub-problem descriptions of the supplementary problem description, and dynamically allocate weights to the multiple sub-problem descriptions according to the real-time importance of each sub-problem description and the user's specific demand scenario, so as to determine the priority of each sub-problem description; A generation module is used to perform a deep search on each sub-problem description according to the priority, obtain the knowledge fragment of each sub-problem description, and judge the correlation between the knowledge fragment and the corresponding sub-problem description based on the semantic understanding technology of BGE-Reranker to obtain the correlation result, and regenerate the precise search terms of each sub-problem description according to the correlation result to re-search and obtain the final knowledge fragment retrieved; The fusion module is used to perform personalized fusion of the final knowledge fragments according to the question type, knowledge background and questioning style of the user's question, obtain the answer to the user's question and return it to the user front end.

6. The device according to claim 5, characterized in that In the generation module, each sub-problem description is deeply searched according to the priority to obtain the knowledge fragments of each sub-problem description, including: A determination submodule is used to determine the index library for retrieval by combining the time of the question asked by the retrieval user, the source channel of the question, and the multimodal information of the relevance of the historical questions; An adjustment submodule, used to dynamically adjust the index library according to the source channel of the user question to obtain the latest index library; The retrieval submodule is used to perform a deep search on each sub-problem description in the latest index library according to the priority level to obtain the knowledge fragment of each sub-problem description.

7. The device according to claim 6, characterized in that Retrieval submodule, including: An acquisition unit is used to use knowledge graph enhancement technology to perform structured representation and association analysis on the knowledge in the field of finance and taxation according to the grid problem description, and to obtain the knowledge fragments described in each sub-problem.

8. The device according to claim 5, characterized in that Fusion modules, including: The reasoning submodule is used to use the financial and taxation big model to infer the key points and potential needs of the user's problem based on the specific asset characteristics, disposal methods, and financial status factors in the user's problem; The generation submodule is used to personalize and integrate the final knowledge fragments according to the key points of the user's question and the potential needs, combined with real-time fiscal and tax policy changes and market dynamics, to generate the answer to the question.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and the computer program is used to execute the method according to any one of claims 1 to 4.

10. An electronic device, characterized in that: The electronic device comprises: processor; a memory for storing instructions executable by the processor; The processor is used to read the executable instructions from the memory and execute the instructions to implement the method described in any one of claims 1 to 4.

Images