Man-machine conversation analysis method, device and equipment, medium and program product
By decomposing and matching user request tasks, combined with the generalization processing of large language models and the execution of interface tools, the problem of insufficient intelligence in human-computer dialogue analysis is solved, and more accurate analysis results are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 国投融合科技股份有限公司
- Filing Date
- 2025-12-31
- Publication Date
- 2026-04-14
AI Technical Summary
Existing human-computer dialogue systems lack sufficient intelligence and adaptability, leading to inaccurate analysis results.
By breaking down the user-input request task, generating a sequence of request subtasks, and matching them with a preset set of business seed questions, the system uses a large language model for generalization and API tools for execution to generate accurate analysis results.
It improves the intelligence and adaptability of human-computer dialogue, obtains more accurate analysis results, and meets user needs.
Smart Images
Figure CN121860044A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of human-computer interaction technology, specifically to a human-computer dialogue analysis method, apparatus, equipment, medium, and program product. Background Technology
[0002] In multi-source data analysis and intelligent decision-making scenarios, enterprises typically rely on databases, file systems, and business documents to analyze data from human-computer dialogues. However, existing human-computer dialogue systems have significant shortcomings in terms of intelligence and adaptability, leading to inaccurate analysis results. Summary of the Invention
[0003] At least one embodiment of the present invention provides a human-computer dialogue analysis method, apparatus, device, medium, and program product to solve the problem that the existing human-computer dialogue has obvious deficiencies in intelligence and adaptability, resulting in inaccurate analysis results.
[0004] To solve the above-mentioned technical problems, the present invention is implemented as follows:
[0005] In a first aspect, embodiments of the present invention provide a human-computer dialogue analysis method, comprising:
[0006] The user-input request task is broken down into a sequence of request subtasks;
[0007] The request subtasks in the request subtask sequence are matched with the business seed questions in the preset business seed question set to obtain multiple target business seed questions corresponding to the request subtask sequence;
[0008] Based on the interface tool corresponding to the target business seed problem, execute the request subtask sequence to obtain the analysis results corresponding to the request task.
[0009] Optionally, the human-computer dialogue analysis method, wherein the user-input request task is broken down to generate a sequence of request sub-tasks, includes:
[0010] The user's intent is obtained by parsing the user's input request task based on the large language model.
[0011] Based on the user intent and a pre-built knowledge vector library, the request task is broken down to generate a sequence of request subtasks.
[0012] Optionally, the human-computer dialogue analysis method further includes:
[0013] Obtain business knowledge documents, which include at least one of the following: business table structure, business documents, and the set of business seed questions;
[0014] Based on the heading levels and / or paragraph boundaries in the business knowledge document, the document is sliced to obtain multiple knowledge text slices.
[0015] The knowledge text slices are encoded to obtain knowledge vector data;
[0016] The knowledge vector data is mapped and stored with the corresponding tags of the knowledge text slices to construct the knowledge vector library.
[0017] Optionally, the human-computer dialogue analysis method, wherein matching the request subtasks in the request subtask sequence with business seed questions in a preset set of business seed questions to obtain multiple target business seed questions corresponding to the request subtask sequence includes:
[0018] The request subtasks in the request subtask sequence are generalized according to the large language model to obtain the generalized request subtasks; wherein, the generalization process includes at least one of synonym replacement, sentence rewriting, scenario expansion and logical equivalence transformation;
[0019] The generalized request subtasks are matched with business seed questions in a preset set of business seed questions to obtain multiple target business seed questions corresponding to the request subtask sequence.
[0020] Optionally, the human-computer dialogue analysis method, wherein matching the request subtasks in the request subtask sequence with business seed questions in a preset set of business seed questions to obtain multiple target business seed questions corresponding to the request subtask sequence includes:
[0021] If the request subtasks in the request subtask sequence are matched with the business seed questions in the preset business seed question set, and no target business seed question is obtained corresponding to the first request subtask in the request subtask sequence, then the first request subtask after generalization is filled in with fields and configured logically to obtain the target business seed question corresponding to the first request subtask.
[0022] Optionally, the human-computer dialogue analysis method further includes:
[0023] Based on the large language model, generate the interface tool corresponding to the target business seed problem corresponding to the first request subtask;
[0024] According to the interface tool library, obtain the interface tool corresponding to the target business seed problem corresponding to the second request subtask in the request subtask sequence; wherein, the target business seed problem corresponding to the second request subtask is obtained by matching the request subtasks in the request subtask sequence with the business seed problems in the preset business seed problem set; wherein, the interface tool library is constructed based on the business knowledge document and the large language model.
[0025] Optionally, in the aforementioned human-computer dialogue analysis method, when the number of multiple interface tools generated based on the large language model exceeds a certain threshold, the method further includes:
[0026] The edit distance of the abstract syntax tree between any two of the multiple interface tools is taken as the logical similarity between any two of the interface tools.
[0027] Based on the logical similarity between any two of the multiple interface tools, the multiple interface tools are divided into at least one category;
[0028] One of the interface tools belonging to the same class shall be retained.
[0029] Optionally, the human-computer dialogue analysis method, wherein executing the request subtask sequence according to the interface tool corresponding to the target business seed question to obtain the analysis result corresponding to the request task includes:
[0030] Based on the dependencies between the request subtasks in the request subtask sequence, the execution order corresponding to the request subtask sequence is obtained;
[0031] Based on the execution order and the interface tool corresponding to the target business seed problem, the sequence of request subtasks is executed to obtain the analysis results corresponding to the request tasks.
[0032] Secondly, embodiments of the present invention also provide a human-computer dialogue analysis device, comprising:
[0033] The generation module is used to break down the user-input request task and generate a sequence of request subtasks.
[0034] The matching module is used to match the request subtasks in the request subtask sequence with the business seed questions in the preset business seed question set to obtain multiple target business seed questions corresponding to the request subtask sequence.
[0035] The execution module is used to execute the request subtask sequence according to the interface tool corresponding to the target business seed problem, and obtain the analysis results corresponding to the request task.
[0036] Thirdly, embodiments of the present invention also provide a human-computer dialogue analysis device, comprising: a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the processor executes the program or instructions to implement the human-computer dialogue analysis method as described in the first aspect.
[0037] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the human-computer dialogue analysis method as described in the first aspect.
[0038] Fifthly, embodiments of the present invention also provide a computer program product, including computer instructions, which, when executed by a processor, implement the human-computer dialogue analysis method as described in the first aspect.
[0039] Compared with existing technologies, embodiments of the present invention provide a human-computer dialogue analysis method, apparatus, device, medium, and program product. The method decomposes a user-input request task into a sequence of request sub-tasks; matches the request sub-tasks in the request sub-task sequence with business seed questions in a preset set of business seed questions to obtain multiple target business seed questions corresponding to the request sub-task sequence; and executes the request sub-task sequence according to the interface tool corresponding to the target business seed questions to obtain the analysis results corresponding to the request task. This improves the level of intelligence and adaptability, obtains accurate human-computer dialogue analysis results, and meets user needs. Attached Figure Description
[0040] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:
[0041] Figure 1 This is a flowchart illustrating the human-computer dialogue analysis method described in an embodiment of the present invention;
[0042] Figure 2 This is a flowchart illustrating one embodiment of the human-computer dialogue analysis method described in this invention.
[0043] Figure 3 This is a flowchart illustrating another embodiment of the human-computer dialogue analysis method described in this invention.
[0044] Figure 4 This is a schematic diagram of the architecture of an application system for the human-computer dialogue analysis method described in an embodiment of the present invention;
[0045] Figure 5 This is a schematic diagram of the modules of the human-computer dialogue analysis device described in an embodiment of the present invention;
[0046] Figure 6 This is a hardware block diagram of the human-computer dialogue analysis device described in an embodiment of the present invention. Detailed Implementation
[0047] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.
[0048] In various embodiments of the present invention, it should be understood that the sequence number of each process described below 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.
[0049] In addition, the terms "system" and "network" are often used interchangeably in this article.
[0050] The terms "first," "second," etc., used in the specification and claims of this invention are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, not limited in number; for example, the first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.
[0051] Reference Figure 1 This invention provides a human-computer dialogue analysis method, comprising:
[0052] Step 101: Decompose the user-input request task to generate a sequence of request subtasks;
[0053] In this embodiment of the invention, a planning agent can be used to decompose the user-input request task and generate a sequence of request subtasks. Specifically, the planning agent receives the user-input request task and can decompose it according to a Large Language Model (LLM) to generate a sequence of request subtasks. The sequence of request subtasks includes multiple request subtasks, such as data query tasks, interface call tasks, and result statistics tasks.
[0054] Step 102: Match the request subtasks in the request subtask sequence with the business seed questions in the preset business seed question set to obtain multiple target business seed questions corresponding to the request subtask sequence;
[0055] The preset business seed problem can be a typical problem sample in a specific business domain. Furthermore, the preset business seed problem can be obtained by generalizing the user-input business seed problem. This generalization process includes, but is not limited to: synonym replacement; sentence rewriting; scenario-based expansion; and logical equivalence transformation.
[0056] In this embodiment of the invention, a generalized intelligent agent can be used to generalize the business seed problem input by the user to obtain the preset business seed problem.
[0057] Step 103: Execute the request subtask sequence according to the interface tool corresponding to the target business seed problem, and obtain the analysis results corresponding to the request task.
[0058] In this embodiment of the invention, an execution agent (Multi Agent) can be employed to execute the request sub-task sequence according to the interface tool corresponding to the target business seed problem, and obtain the analysis results corresponding to the request task. Specifically, this execution agent, as the core agent of the perception-invocation process, schedules the interface tool corresponding to the target business seed problem through the function invocation capability of LLM, and executes it according to the request sub-task sequence output by the planning agent.
[0059] Moreover, the executing agent has autonomous perception capabilities. When the lack of an interface causes an anomaly in the interface tool scheduling, it can trigger the interface tool generation process or terminate the process.
[0060] For example, the LLM used in this embodiment of the invention is DeepSeek-r1-671b or Qwen3-32b, which serves as the core inference engine, undertaking natural language understanding tasks, generation tasks, logical reasoning tasks, etc., and driving intelligent interaction and process execution. The natural language understanding task includes recognizing user intent; the generation task includes, but is not limited to: generating analysis results; generating interface tools; the logical reasoning task includes, but is not limited to: decomposing the request task.
[0061] In one implementation, optionally, the user-input request task is broken down to generate a sequence of request subtasks, including:
[0062] Based on the LLM, the user's input request task is parsed to obtain the user's intent;
[0063] Based on the user intent and a pre-built knowledge vector library, the request task is broken down to generate a sequence of request subtasks.
[0064] In this embodiment of the invention, after receiving the user's input request task, intent parsing is performed based on the LLM's natural language understanding capability to obtain the user's intent. Based on the user's intent and the knowledge vector library of the pre-built knowledge parsing layer, the request task decomposition process is triggered to generate a request subtask sequence.
[0065] Optionally, when the decomposition process of the request task is triggered, the complex request task can be generated into a sequence of request subtasks using the OpenManus intelligent agent framework.
[0066] In one embodiment, optionally, the method further includes:
[0067] Obtain business knowledge documents, which include at least one of the following: business table structure, business documents, and the set of business seed questions;
[0068] Based on the heading levels and / or paragraph boundaries in the business knowledge document, the document is sliced to obtain multiple knowledge text slices.
[0069] The knowledge text slices are encoded to obtain knowledge vector data;
[0070] The knowledge vector data is mapped and stored with the corresponding tags of the knowledge text slices to construct the knowledge vector library.
[0071] In this embodiment of the invention, the business knowledge document may be uploaded by the user. The business table structure includes business system data to support structured data interaction; the business document includes industry processes and / or domain static knowledge, such as unstructured knowledge carriers like business rules, and a set of domain-specific terminology to ensure accurate understanding of the semantics of vertical scenarios; the business seed question set may be typical question samples from a specific business domain, such as historical question samples or user-added question samples, used for question generalization and interface tool construction.
[0072] Before step 101 above, firstly, a business knowledge document is obtained; based on the heading levels and / or paragraph boundaries in the business knowledge document, the document is sliced to obtain multiple knowledge text slices; optionally, the heading levels and / or paragraph boundaries in the business knowledge document are obtained by parsing the document's format, for example, by parsing the paragraph separators of a TXT file; the heading levels include, but are not limited to: first-level headings; second-level headings; the paragraph boundaries include, but are not limited to: line breaks; tabs; blank lines. Here, by recognizing the heading levels and / or paragraph boundaries in the business knowledge document, content awareness is achieved, and the document is sliced to obtain multiple knowledge text slices. Compared to the fixed-length slicing method used in the prior art, this method can preserve the complete semantics to a greater extent.
[0073] Then, multiple knowledge text slices can be saved to an SQLite database, and each knowledge text slice can be assigned a tag (ID) to establish a mapping relationship between the knowledge text slice and the corresponding knowledge vector data during retrieval. Here, the SQLite database is a lightweight relational database used to store structured knowledge, such as the tags of the business knowledge document, the path of the business knowledge document, and multiple knowledge text slices, to meet the needs of accurate data query and correlation analysis, thereby providing structured data management capabilities.
[0074] Furthermore, the knowledge text slices are encoded to obtain knowledge vector data. Optionally, a vector model is used to encode the knowledge text slices, transforming them into high-dimensional knowledge vector data. Here, the vector model can be the BGE-M3 model, responsible for text vectorization conversion, providing basic vector representation capabilities for similarity retrieval and retrieval enhancement, and supporting the knowledge retrieval and semantic matching process.
[0075] The knowledge vector data is mapped and stored with the corresponding tags of the knowledge text slices to construct the knowledge vector library. Optionally, the tags of the knowledge text slices and the corresponding knowledge vector data are saved to the Faiss database to establish a fast mapping between the knowledge vector data and the knowledge text slices, supporting subsequent Top-K retrieval based on semantic similarity and combined queries with knowledge text slice filtering. Here, the Faiss database provides vector similarity-based retrieval technology, storing text vectorized data to support efficient similar document and question retrieval, achieving efficient and accurate vector retrieval, and is a core dependency of the subsequent retrieval enhancement process.
[0076] It should be noted that by combining vector retrieval and re-ranking models, such as BGE-rerank-v2-m3, and using vector distance calculations, such as cosine similarity, to recall relevant segments, and then using LLM to generate analysis results, the model's inherent knowledge is supplemented, and the response accuracy and domain adaptability are improved.
[0077] In one implementation, optionally, the request subtasks in the request subtask sequence are matched with business seed questions in a preset set of business seed questions to obtain multiple target business seed questions corresponding to the request subtask sequence, including:
[0078] The request subtasks in the request subtask sequence are generalized according to LLM to obtain the generalized request subtasks; wherein, the generalization process includes at least one of synonym replacement, sentence rewriting, scenario-based expansion and logical equivalence transformation;
[0079] The generalized request subtasks are matched with business seed questions in a preset set of business seed questions to obtain multiple target business seed questions corresponding to the request subtask sequence.
[0080] In this embodiment of the invention, a generalized intelligent agent can be employed. Based on a preset set of business seed questions and combined with the natural language understanding capabilities of LLM, multiple target business seed questions corresponding to the request subtask sequence can be obtained through at least one of the following: synonym replacement, sentence rewriting, scenario-based expansion, and logical equivalence transformation. The sentence rewriting includes, but is not limited to: active conversion; passive conversion; interrogative sentence switching; and declarative sentence switching. The scenario-based expansion includes, but is not limited to: supplementing the question descriptions under different business scenarios. The logical equivalence transformation is, for example, the difference between "how to query" and "what is the query method".
[0081] Since the generalized request subtasks are multi-intent and multi-representation request subtasks, the multiple target business seed problems corresponding to the generalized request subtask sequence are also multi-intent and multi-representation target business seed problems.
[0082] It should be noted that while generalizing the request subtasks in the request subtask sequence according to LLM, a domain technology dictionary can be used to ensure the professional accuracy of the generalized request subtasks, maximize the coverage of potential user interaction scenarios, and improve the compatibility of the interface tool with diverse user intents.
[0083] In one implementation, optionally, the request subtasks in the request subtask sequence are matched with business seed questions in a preset set of business seed questions to obtain multiple target business seed questions corresponding to the request subtask sequence, including:
[0084] If the request subtasks in the request subtask sequence are matched with the business seed questions in the preset business seed question set, and no target business seed question is obtained corresponding to the first request subtask in the request subtask sequence, then the first request subtask after generalization is filled in with fields and configured logically to obtain the target business seed question corresponding to the first request subtask.
[0085] In this embodiment of the invention, if the target business seed problem corresponding to the first request subtask in the request subtask sequence is not obtained in step 102 above, then the first request subtask after the above generalization process needs to be filled in with fields and configured logically according to user requirements to obtain the target business seed problem corresponding to the first request subtask.
[0086] In one embodiment, optionally, the method further includes:
[0087] Based on the LLM, generate the interface tool corresponding to the target business seed problem corresponding to the first request subtask;
[0088] According to the interface tool library, obtain the interface tool corresponding to the target business seed problem corresponding to the second request subtask in the request subtask sequence; wherein, the target business seed problem corresponding to the second request subtask is obtained by matching the request subtasks in the request subtask sequence with the business seed problems in the preset business seed problem set; wherein, the interface tool library is built based on the business knowledge document and the LLM.
[0089] In this embodiment of the invention, before step 103, the method further includes generating an interface tool corresponding to the target business seed problem. Here, generating the interface tool corresponding to the target business seed problem includes the following two methods:
[0090] Method 1: In step 102 above, the request subtasks in the request subtask sequence are matched with the business seed questions in the preset business seed question set. For the first request subtask that does not obtain a target business seed question, an interface tool corresponding to the target business seed question of the first request subtask is generated according to the LLM. Specifically, firstly, the fields of the first request subtask after the above generalization process need to be completed and the logic configured according to the user's data interaction requirements to obtain the target business seed question corresponding to the first request subtask. Then, according to the LLM, the business knowledge document, and the target business seed question corresponding to the first request subtask, an interface tool corresponding to the target business seed question of the first request subtask is generated.
[0091] Here, LLM automatically parses data interaction requirements, field mapping relationships, and business logic to generate standardized interface tools. These interface tools support multiple types of interfaces, such as RESTful and SQL queries, and can be implemented using various programming languages, including Java and Python. The entire process requires no code writing from the user, significantly reducing the technical threshold and enabling the expansion of the interface tools.
[0092] Method 2: In step 102 above, the request subtasks in the request subtask sequence are matched with the business seed questions in the preset business seed question set to obtain the second request subtask of the target business seed question. Based on the interface tool library, the interface tool corresponding to the target business seed question corresponding to the second request subtask in the request subtask sequence is obtained. Specifically, the pre-built interface tool library is invoked to directly obtain the interface tool corresponding to the target business seed question of the second request subtask in the request subtask sequence. Here, the interface tool library can be built periodically.
[0093] It should be noted that after generating the interface tools corresponding to the target business seed problem using the above two methods, an automated testing system can be built. By generating simulated request data, functional verification and / or abnormal scenario testing can be performed on the generated interface tools. The functional verification includes, for example, data return accuracy or parameter compatibility; the abnormal scenario testing includes, for example, null value input or format errors. This ensures the usability and stability of the interface tools and guarantees their quality.
[0094] In one implementation, optionally, if the number of multiple interface tools generated based on the LLM exceeds a certain threshold, the method further includes:
[0095] The edit distance of the Abstract Syntax Tree (AST) between any two of the multiple interface tools is used as the logical similarity between any two interface tools.
[0096] Based on the logical similarity between any two of the multiple interface tools, the multiple interface tools are divided into at least one category;
[0097] One of the interface tools belonging to the same class shall be retained.
[0098] In this embodiment of the invention, when the number of multiple interface tools generated by LLM exceeds a threshold, interface merging is initiated to optimize traffic and avoid redundancy. Specifically, the logical structure and parameter configuration of each interface tool are parsed using AST, the edit distance of the AST between any two interface tools is calculated, and the logical similarity between the interface tools is quantified based on the edit distance of the AST. Based on the logical similarity, a clustering algorithm is used to group multiple interface tools with duplicate functions or logical redundancy into one category, and one interface tool is retained among the multiple interface tools belonging to the same category.
[0099] Here, when retaining one of the multiple interface tools belonging to the same category, the interface tool with the optimal execution path can be retained, and the interface tools that are not retained can be deleted.
[0100] In one implementation, optionally, the request subtask sequence is executed according to the interface tool corresponding to the target business seed problem to obtain the analysis results corresponding to the request task, including:
[0101] Based on the dependencies between the request subtasks in the request subtask sequence, the execution order corresponding to the request subtask sequence is obtained;
[0102] Based on the execution order and the interface tool corresponding to the target business seed problem, the sequence of request subtasks is executed to obtain the analysis results corresponding to the request tasks.
[0103] In this embodiment of the invention, the dependencies between the request subtasks in the request subtask sequence are used to indicate the order in which the request subtasks are executed. For example, the third request subtask is executed before the fourth request subtask; a data query task is executed before a result statistics task. Thus, when executing the request subtask sequence, the third request subtask must be executed first, followed by the fourth request subtask. Here, the request subtasks in the request subtask sequence can be obtained through planning agent analysis.
[0104] Based on the dependencies between the request subtasks in the request subtask sequence, the execution order corresponding to the request subtask sequence is obtained. The execution order of the request subtask sequence can be a Workflow task flow. According to the execution order and the interface tool corresponding to the target business seed problem, the request subtask sequence is executed to obtain the analysis results corresponding to the request tasks. That is, the request subtask sequence is executed according to the Workflow task flow order, realizing the structured and process-oriented decomposition of complex requirements.
[0105] The pre-built interface tool library in this embodiment of the invention can support the execution agent to call the request subtask sequence.
[0106] It should be noted that the interface tool library can add or remove interface tools according to system functions, that is, the interface tools are pluggable, and the interface tools may include at least one of the following:
[0107] AskHuman tool: When the user's input of the request task is unclear, such as when key parameters are missing, it automatically triggers human-computer interaction to request supplementary information from the user, ensuring the accuracy of the execution of the request task;
[0108] Termination Tool: When the requested task is being executed and a situation arises requiring termination, this termination tool is invoked to output the final result.
[0109] File access tool: invokes business knowledge documents to complete data reading and content retrieval, providing information support for the execution of the requested task;
[0110] Generate new interface tools: When the existing interface tool library cannot meet the requirements of the request task, an interface tool is generated based on the request task input by the user and the business knowledge document, and its usability is verified in conjunction with automated testing tools.
[0111] Optionally, the interface tools corresponding to the request subtask sequence can be sorted according to the execution order and reordering model to obtain the sorted interface tools.
[0112] Furthermore, the execution results of the requested subtask sequence are summarized, and the execution results are integrated and the logic is sorted out through LLM to obtain the analysis results corresponding to the requested task, that is, the final response, and output a complete solution that matches the user input of the requested task.
[0113] Figure 2 This is a flowchart illustrating one embodiment of the human-computer dialogue analysis method described in this invention. Figure 2 As shown, the method includes a pre-construction process, as detailed below:
[0114] The user inputs a business seed question, and the LLM is used to generalize the business seed question to obtain a business seed question with multiple intents and expressions.
[0115] Based on the generalized business seed problem and LLM, generate interface tools;
[0116] Test the interface tools;
[0117] The business documents and business table structures are parsed and their content is sliced to obtain multiple knowledge text slices. These multiple knowledge text slices and the tested interface tools are stored in an SQLite database. The knowledge vector data corresponding to the multiple knowledge text slices and the tested interface tools are vectorized and stored in a Faiss database. In this way, an interface tool library, a business knowledge base, and a business table structure can be built based on the SQLite database.
[0118] Figure 3 This is a flowchart illustrating another embodiment of the human-computer dialogue analysis method described in this invention. Figure 3 As shown, the method includes a real-time dialogue process, as detailed below:
[0119] The user-input request task is divided into a sequence of request subtasks based on the planning agent and LLM;
[0120] Vectorize the data in the interface tool library, business knowledge base, and business table structure, and based on retrieval enhancement and combined with the natural language understanding capabilities of LLM, obtain interface tools or generate interface tools.
[0121] The execution agent executes the interface tool corresponding to the requested subtask sequence, processes the data to obtain analysis results, and displays the interface tool.
[0122] Figure 4 This is a schematic diagram of the architecture of an application system for the human-computer dialogue analysis method described in an embodiment of the present invention. The application system for the human-computer dialogue analysis method can be simply referred to as a human-computer dialogue analysis system, such as... Figure 4 As shown, this human-computer dialogue analysis system adopts a three-layer architecture, including a knowledge parsing layer, an autonomous construction layer, and an intelligent platform layer. These three layers work together to achieve natural language-driven data understanding, automatic interface construction, and intelligent task execution. The descriptions of each layer are as follows:
[0123] Knowledge parsing layer: Responsible for parsing multi-source heterogeneous data, including databases, files (Excel, CSV), documents (PDF, Word, TXT), etc. This human-computer dialogue analysis system uses a data parsing SDK and vector model to vectorize text, and combines a re-ranking model for high-precision sorting, forming a unified knowledge index structure;
[0124] The self-construction layer is the core task orchestration engine of this human-computer dialogue analysis system. By combining LLM with Retrieval-augmented Generation (RAG), it realizes automatic generation of natural language questions into executable interfaces, task parsing, and tool scheduling.
[0125] Intelligent Platform Layer: This layer serves as the user interaction and task management center, supporting natural language dialogue, task status tracking, and visualization. The platform can provide customized knowledge responses and access controls based on different user roles (such as operators, managers, and analysts).
[0126] In addition, this human-computer dialogue analysis system is used to build a central hub and an interface tool library, as detailed below:
[0127] The workflow building interface allows users to easily upload files and manage the build process through simple operations (such as drag and drop, configuration), lowering the technical barrier to platform use and empowering business personnel to participate in platform building. Simultaneously, users can view seed issues under construction and interface tools after construction, providing tool search, version control, and permission management functions. This supports tool reuse and sharing, creating a standardized tool ecosystem and improving platform building efficiency and functional reusability.
[0128] This human-computer dialogue analysis system supports task tracing and data tracing, as detailed below:
[0129] View the entire task execution process data (such as task breakdown path, subtask execution logs, and API call details), support task backtracking, provide data basis for model optimization and process iteration, assist in troubleshooting data-related issues (such as knowledge conflicts and data errors), and ensure the platform's data quality, maintainability, and continuous evolution capabilities.
[0130] This human-computer dialogue analysis system supports visualization, as detailed below:
[0131] By presenting answers to user questions through charts, dashboards, and other formats, we help users intuitively understand business logic and data value, thus assisting in decision-making and optimization.
[0132] This human-computer dialogue analysis system supports multi-turn dialogues, as detailed below:
[0133] Based on LLM and RAG, it supports multi-turn contextual interaction, maintains dialogue state (such as historical questions, answers, user intent), realizes natural and coherent multi-turn communication, solves the needs of complex business consultation and process guidance, and improves the user interaction experience.
[0134] It should be noted that the human-computer dialogue analysis method described in this embodiment of the invention can be applied to scenarios such as intelligent data analysis, enterprise decision support, knowledge retrieval and question answering, and process automation. Typical application scenarios include, but are not limited to: natural language querying of data reports; automatic generation of business analysis interfaces; multi-dimensional data trend prediction and interpretation; cross-system data fusion and unified analysis: the system has dynamic knowledge update capabilities, and when data structures or rules change, it can automatically reconstruct the knowledge and interface library, achieving adaptive evolution of the system.
[0135] In summary, the human-computer dialogue analysis method described in the embodiments of the present invention has the following advantages:
[0136] Low barrier to entry and high efficiency: Business personnel can build the system without coding, greatly reducing development and usage costs;
[0137] Adaptive and Continuous Evolution: Possesses self-learning and dynamic update capabilities to adapt to business changes;
[0138] Cross-system collaboration capability: Breaking down data silos and enabling multi-source data linkage and business process integration;
[0139] Intelligent interactive experience: Supports multi-turn dialogue, semantic understanding and task tracing to improve user experience;
[0140] Flexible and scalable: Supports multi-agent expansion and modular deployment, suitable for multiple application scenarios.
[0141] Reference Figure 5 This invention also provides a human-computer dialogue analysis device, comprising:
[0142] The generation module 501 is used to break down the user-input request task and generate a sequence of request subtasks.
[0143] Matching module 502 is used to match the request subtasks in the request subtask sequence with the business seed questions in the preset business seed question set to obtain multiple target business seed questions corresponding to the request subtask sequence;
[0144] The execution module 503 is used to execute the request subtask sequence according to the interface tool corresponding to the target business seed problem, and obtain the analysis results corresponding to the request task.
[0145] Optionally, in the aforementioned human-computer dialogue analysis device, the generation module 501 is specifically used for:
[0146] The user's intent is obtained by parsing the user's input request task based on the large language model.
[0147] Based on the user intent and a pre-built knowledge vector library, the request task is broken down to generate a sequence of request subtasks.
[0148] Optionally, the human-computer dialogue analysis device further includes:
[0149] The acquisition module is used to acquire business knowledge documents, which include at least one of the following: business table structure, business documents, and the set of business seed questions;
[0150] The slicing module is used to slice the business knowledge document according to the heading level and / or paragraph boundaries in the business knowledge document to obtain multiple knowledge text slices;
[0151] The encoding module is used to encode the knowledge text slices to obtain knowledge vector data;
[0152] The storage module is used to map and store the knowledge vector data with the corresponding tags of the knowledge text slices, thereby constructing the knowledge vector library.
[0153] Optionally, in the aforementioned human-computer dialogue analysis device, the matching module 502 is specifically used for:
[0154] The request subtasks in the request subtask sequence are generalized according to the large language model to obtain the generalized request subtasks; wherein, the generalization process includes at least one of synonym replacement, sentence rewriting, scenario expansion and logical equivalence transformation;
[0155] The generalized request subtasks are matched with business seed questions in a preset set of business seed questions to obtain multiple target business seed questions corresponding to the request subtask sequence.
[0156] Optionally, in the aforementioned human-computer dialogue analysis device, the matching module 502 is specifically used for:
[0157] If the request subtasks in the request subtask sequence are matched with the business seed questions in the preset business seed question set, and no target business seed question is obtained corresponding to the first request subtask in the request subtask sequence, then the first request subtask after generalization is filled in with fields and configured logically to obtain the target business seed question corresponding to the first request subtask.
[0158] Optionally, the human-computer dialogue analysis device further includes:
[0159] The first interface module is used to generate the interface tool corresponding to the target business seed problem corresponding to the first request subtask based on the large language model.
[0160] The second interface module is used to obtain the interface tool corresponding to the target business seed problem corresponding to the second request subtask in the request subtask sequence according to the interface tool library; wherein, the target business seed problem corresponding to the second request subtask is obtained by matching the request subtasks in the request subtask sequence with the business seed problems in the preset business seed problem set; wherein, the interface tool library is constructed based on the business knowledge document and the large language model.
[0161] Optionally, the human-computer dialogue analysis device further includes:
[0162] The determination module is used to take the edit distance of the abstract syntax tree between any two of the multiple interface tools as the logical similarity between any two of the interface tools.
[0163] The segmentation module is used to classify the multiple interface tools into at least one class based on the logical similarity between any two of the interface tools.
[0164] A retention module is used to retain one of the interface tools among multiple interface tools belonging to the same class.
[0165] Optionally, in the aforementioned human-computer dialogue analysis device, the execution module 503 is specifically used for:
[0166] Based on the dependencies between the request subtasks in the request subtask sequence, the execution order corresponding to the request subtask sequence is obtained;
[0167] Based on the execution order and the interface tool corresponding to the target business seed problem, the sequence of request subtasks is executed to obtain the analysis results corresponding to the request tasks.
[0168] It should be noted that the human-computer dialogue analysis device provided in the embodiments of the present invention can execute the above-described human-computer dialogue analysis method. Therefore, all embodiments of the above-described human-computer dialogue analysis method are applicable to the human-computer dialogue analysis device and can achieve the same or similar technical effects.
[0169] This invention also provides a human-computer dialogue analysis device, such as... Figure 6 As shown, it includes:
[0170] The processor 601, memory 602, transceiver 603, and programs or instructions stored in the memory 602 and executable on the processor 601; when the processor 601 executes the programs or instructions, it implements the various processes of the above-described human-computer dialogue analysis method embodiments and achieves the same technical effect. To avoid repetition, these will not be described again here.
[0171] The transceiver 603 is used to receive and send data under the control of the processor 601.
[0172] Among them, Figure 6In this context, the bus architecture can include any number of interconnected buses and bridges, specifically connecting various circuits of one or more processors represented by processor 601 and memory represented by memory 602. The bus architecture can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. Transceiver 603 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. For different user equipment, the user interface 604 can also be an interface capable of connecting external or internal devices, including but not limited to keypads, displays, speakers, microphones, joysticks, etc.
[0173] The processor 601 is responsible for managing the bus architecture and general processing, while the memory 602 can store the data used by the processor 601 when performing operations.
[0174] This invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the above-described human-computer dialogue analysis method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.
[0175] This invention also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they implement the various processes of the above-described human-computer dialogue analysis method embodiments and achieve the same technical effects. To avoid repetition, they will not be described again here.
[0176] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0177] 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 the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The 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 (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.
[0178] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.
Claims
1. A human-computer dialogue analysis method, characterized in that, include: The user-input request task is broken down into a sequence of request subtasks; The request subtasks in the request subtask sequence are matched with the business seed questions in the preset business seed question set to obtain multiple target business seed questions corresponding to the request subtask sequence; Based on the interface tool corresponding to the target business seed problem, execute the request subtask sequence to obtain the analysis results corresponding to the request task.
2. The method according to claim 1, characterized in that, The user-input request task is broken down into a sequence of request subtasks, including: The user's intent is obtained by parsing the user's input request task based on the large language model. Based on the user intent and a pre-built knowledge vector library, the request task is broken down to generate a sequence of request subtasks.
3. The method according to claim 2, characterized in that, The method further includes: Obtain business knowledge documents, which include at least one of the following: business table structure, business documents, and the set of business seed questions; Based on the heading levels and / or paragraph boundaries in the business knowledge document, the document is sliced to obtain multiple knowledge text slices. The knowledge text slices are encoded to obtain knowledge vector data; The knowledge vector data is mapped and stored with the corresponding tags of the knowledge text slices to construct the knowledge vector library.
4. The method according to claim 1, characterized in that, The request subtasks in the request subtask sequence are matched with the business seed questions in a preset set of business seed questions to obtain multiple target business seed questions corresponding to the request subtask sequence, including: The request subtasks in the request subtask sequence are generalized according to the large language model to obtain the generalized request subtasks; wherein, the generalization process includes at least one of synonym replacement, sentence rewriting, scenario expansion and logical equivalence transformation; The generalized request subtasks are matched with business seed questions in a preset set of business seed questions to obtain multiple target business seed questions corresponding to the request subtask sequence.
5. The method according to claim 4, characterized in that, The request subtasks in the request subtask sequence are matched with the business seed questions in a preset set of business seed questions to obtain multiple target business seed questions corresponding to the request subtask sequence, including: If the request subtasks in the request subtask sequence are matched with the business seed questions in the preset business seed question set, and no target business seed question is obtained corresponding to the first request subtask in the request subtask sequence, then the first request subtask after generalization is filled in with fields and configured logically to obtain the target business seed question corresponding to the first request subtask.
6. The method according to claim 5, characterized in that, The method further includes: Based on the large language model, generate the interface tool corresponding to the target business seed problem corresponding to the first request subtask; According to the interface tool library, obtain the interface tool corresponding to the target business seed problem corresponding to the second request subtask in the request subtask sequence; wherein, the target business seed problem corresponding to the second request subtask is obtained by matching the request subtasks in the request subtask sequence with the business seed problems in the preset business seed problem set; wherein, the interface tool library is constructed based on the business knowledge document and the large language model.
7. The method according to claim 6, characterized in that, When the number of multiple interface tools generated based on the large language model exceeds a certain threshold, the method further includes: The edit distance of the abstract syntax tree between any two of the multiple interface tools is taken as the logical similarity between any two of the interface tools. Based on the logical similarity between any two of the multiple interface tools, the multiple interface tools are divided into at least one category; One of the interface tools belonging to the same class shall be retained.
8. The method according to claim 1, characterized in that, Based on the interface tool corresponding to the target business seed problem, execute the request subtask sequence to obtain the analysis results corresponding to the request task, including: Based on the dependencies between the request subtasks in the request subtask sequence, the execution order corresponding to the request subtask sequence is obtained; Based on the execution order and the interface tool corresponding to the target business seed problem, the sequence of request subtasks is executed to obtain the analysis results corresponding to the request tasks.
9. A human-computer dialogue analysis device, characterized in that, include: The generation module is used to break down the user-input request task and generate a sequence of request subtasks. The matching module is used to match the request subtasks in the request subtask sequence with the business seed questions in the preset business seed question set to obtain multiple target business seed questions corresponding to the request subtask sequence. The execution module is used to execute the request subtask sequence according to the interface tool corresponding to the target business seed problem, and obtain the analysis results corresponding to the request task.
10. A human-computer dialogue analysis device, characterized in that, include: A processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the processor, when executing the program or instructions, implements the human-computer dialogue analysis method as described in any one of claims 1 to 8.
11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the human-computer dialogue analysis method as described in any one of claims 1 to 8.
12. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the human-computer dialogue analysis method as described in any one of claims 1 to 8.