Complex decision-oriented intent-driven path inference atlas question-answering system and method

Abstract

The invention provides a complex decision-oriented intention-driven path inference atlas question-answering system and method. According to the system, semantic analysis is carried out on a question input by a user, key information such as a target object, a region, an event and time is extracted, the intention type of the question is recognized, and a matched reasoning strategy is dynamically selected according to the intention type. And generating a plurality of candidate reasoning paths on the basis of the knowledge graph, carrying out weighted evaluation on the reasoning paths from a plurality of dimensions such as logic consistency, information integrity, problem correlation and data support degree, screening to obtain a target reasoning path, and outputting a reasoning result. The system displays a complete reasoning process while giving a conclusion, and the interpretability and credibility of a result are improved. The method is suitable for complex decision analysis scenes with multiple entities, multiple relations and strong time sequence constraints, and the reasoning accuracy and the decision support capability can be effectively improved.

Description

Intent-Driven Path Reasoning Graph Question Answering System and Method for Complex Decision Making Technical Field

[0001] This invention relates to the field of interdisciplinary technology of artificial intelligence and knowledge graphs, and more specifically, to an intent-driven path reasoning graph question-answering system and method for complex decision-making. Background Technology

[0002] With the rapid development of information technology and artificial intelligence, the demand for complex decision analysis based on multi-source data is constantly increasing. In practical applications, the questions raised by users often have characteristics such as multiple entities, multiple relationships, and strong temporal constraints. Simply relying on keyword matching or static query methods is no longer sufficient to meet the requirements for depth of understanding and reliability of reasoning. Especially in scenarios involving target object behavior analysis, regional situation assessment, and risk evaluation, how to conduct effective correlation analysis on massive amounts of structured and unstructured data has become one of the core issues facing the field of intelligent decision support.

[0003] Most existing graph-based question answering or decision analysis systems are based on pre-built knowledge graphs, outputting query results through entity relationship matching or rule-driven methods. These technologies typically employ fixed inference strategies, lacking specific differentiation for different question types and failing to dynamically adjust the inference approach based on the user's actual intent. When faced with path-related or temporal evolution problems, existing technologies often only provide superficial association results, failing to uncover deeper information hidden in multi-hop relationships or temporal evolution, thus limiting the accuracy and practicality of the inference results.

[0004] On the other hand, some existing technologies primarily focus on static relationships between entities during the reasoning process, lacking effective modeling of time factors and dynamic change characteristics. For problems involving time conditions, existing systems often use time merely as a simple filtering condition, failing to systematically analyze the trends and stage-specific characteristics of entities or events over time, thus making it difficult to support the judgment and prediction of complex evolutionary processes. This insufficient time-series modeling capability is particularly prominent in complex decision-making scenarios.

[0005] Furthermore, existing graph reasoning systems generally suffer from insufficient interpretability in their result presentation. Most systems only output the final conclusion or answer, failing to provide a clear reasoning path, key node relationships, and scoring criteria, making it difficult for users to understand the specific logical process that led to the conclusion. When the reasoning results are used to assist decision-making, the lack of a transparent and traceable reasoning process significantly reduces the credibility and application value of the results.

[0006] Meanwhile, existing technologies typically employ single or implicit judgment criteria in path selection and evaluation, lacking a mechanism for multi-dimensional comprehensive evaluation of inference paths, making it difficult to distinguish the reliability of different inference paths. This makes it difficult for the system to consistently output the most valuable result when faced with multiple possible explanatory paths, and also increases the risk of misjudgment.

[0007] In summary, existing technologies for graph-based question answering and reasoning analysis of complex decision-making problems generally suffer from problems such as insufficient adaptability of reasoning strategies, limited ability to fuse temporal information, uninterpretable reasoning processes, and imperfect path evaluation mechanisms.

[0008] Therefore, there is an urgent need for intent-driven path reasoning graph question answering systems and methods for complex decision-making to solve these problems. Summary of the Invention

[0009] The purpose of this invention is to solve the technical problems mentioned in the background art above, and to provide an intent-driven path reasoning graph question answering system for complex decision-making, comprising: a question parsing module, used to parse user-input questions based on a natural language processing model, extract target object entities, region entities, event entities, and time entities from the questions, and construct a semantic representation of the questions; an intent recognition and strategy selection module, used to generate question features based on the semantic representation, identify the intent type corresponding to the questions, and select a reasoning strategy matching the intent type from a variety of preset reasoning strategies; a reasoning path generation module, used to generate multiple candidate reasoning paths in the knowledge graph based on entity relationships and path constraints according to the selected reasoning strategy; a path evaluation and selection module, used to evaluate the candidate reasoning paths in multiple dimensions and select the target reasoning path; and a result generation and visualization module, used to generate the final answer based on the target reasoning path and display the reasoning process in a structured manner.

[0010] As a preferred technical solution of the present invention, the problem parsing module includes a named entity recognition unit and a relation extraction unit, which are used to extract target object entities, region entities, event entities and time entities from user problems, and construct a semantic representation framework for the problem.

[0011] As a preferred technical solution of the present invention, the intent recognition and strategy selection module includes an intent classification unit, which is used to identify user questions as path-related questions or time-series evolution questions based on question characteristics, and activate the corresponding reasoning strategy accordingly.

[0012] As a preferred technical solution of the present invention, when the problem is a path association problem, the reasoning strategy is a multi-hop path reasoning strategy. The multi-hop path reasoning strategy includes seed node determination, relationship expansion and multi-path search steps, which are used to generate multiple candidate reasoning paths in the knowledge graph.

[0013] As a preferred technical solution of the present invention, when the problem is a temporal evolution problem, the reasoning strategy is a temporal reasoning strategy, which includes temporal semantic parsing, temporal data construction and alignment, and temporal relationship reasoning steps, used to generate a reasoning path that reflects the changes of entities or events over time.

[0014] As a preferred technical solution of the present invention, the path evaluation and selection module adopts a multi-dimensional weighted scoring mechanism to comprehensively evaluate the candidate reasoning paths. The multi-dimensional evaluation includes at least logical consistency, information integrity, problem relevance, and data support.

[0015] As a preferred technical solution of the present invention, in the multi-dimensional weighted scoring mechanism, the weight coefficients corresponding to each evaluation dimension are adaptively adjusted according to the intent type of the question.

[0016] As a preferred technical solution of the present invention, the result generation and visualization module is also connected to an external document library, which is used to trigger retrieval operations based on key nodes in the reasoning path, so as to supplement relevant background information and enhance the credibility and interpretability of the reasoning results.

[0017] This invention provides an intent-driven path reasoning graph question answering method for complex decision-making, comprising the following steps: Step 1: Extracting entity and semantic information from the user-input question; Step 2: Identifying the question intent based on the entity and semantic information, and selecting a reasoning strategy matching the question intent; Step 3: Generating multiple candidate reasoning paths in the knowledge graph based on the selected reasoning strategy; Step 4: Evaluating the candidate reasoning paths from multiple dimensions, and selecting the target reasoning path; Step 5: Generating an answer based on the target reasoning path and displaying the reasoning process.

[0018] As a preferred technical solution of the present invention, in step four, a comprehensive credibility score is calculated for each candidate reasoning path, and the comprehensive credibility score is calculated according to the following formula: in, Scoring the logical consistency of the path. Score the information integrity of the path. Scoring the relevance of the path to the user's problem. Score the data support for the path. , , , These are the weighting coefficients for the corresponding evaluation dimensions.

[0019] Compared with existing technologies, the intent-driven path reasoning graph question answering system and method for complex decision-making proposed in this invention takes the user's question intent as the core driver. Through the parsing of question semantics and key entities, it achieves automatic identification and classification of different types of questions, and dynamically selects an appropriate reasoning strategy based on the identification results. Compared with existing technical solutions that use fixed reasoning methods, this invention can employ different reasoning mechanisms for path-related questions and temporal evolution questions, effectively avoiding the problem of insufficient adaptability of a single strategy in complex scenarios, thereby improving the relevance of the reasoning results and the overall analysis efficiency.

[0020] This invention introduces a knowledge graph-based multi-path generation and multi-dimensional evaluation mechanism during the reasoning process. It comprehensively evaluates candidate reasoning paths from multiple dimensions, including logical consistency, information completeness, question relevance, and data support, and selects the target reasoning path through a weighted scoring method. This mechanism can effectively distinguish the reliability of different paths when multiple potential explanation paths exist, reducing reasoning bias caused by improper path selection, improving the stability and credibility of reasoning results, and enhancing the system's practical value in complex decision analysis.

[0021] Furthermore, this invention not only provides the final reasoning conclusion in the results output stage, but also simultaneously offers a complete reasoning path and key node information. It can also supplement the reasoning results with external documentation resources, ensuring good transparency and traceability of the reasoning process. By presenting the reasoning conclusion along with its supporting evidence, this invention significantly improves users' understanding and trust in the system's analysis results, enhancing the interpretability and reliability of decision support results in practical applications. Figure 1 is a system block diagram of the present invention; Figure 2 is a method flowchart of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings 1-2. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0024] The present invention will be further described below with reference to specific embodiments, but the scope of protection of the present invention is not limited to the following embodiments. Equivalent substitutions or modifications made by those skilled in the art without departing from the technical concept of the present invention should all fall within the scope of protection of the present invention.

[0025] This embodiment provides an intent-driven path reasoning graph-based question answering system and method for complex decision-making. This solution is primarily used to analyze and reason about complex decision problems involving multiple entities, multiple relationships, and temporal constraints, providing users with interpretable decision support results. The system as a whole uses knowledge graphs as its data foundation and intent recognition as its core driver. Through path reasoning and multi-dimensional evaluation mechanisms, it achieves efficient solutions to complex problems.

[0026] In its implementation, the system first receives a natural language question input by the user. This question typically contains various semantic elements such as the target object, region information, key events, and time conditions. The system processes the user's question through a question parsing module, which is built on natural language processing technology and includes a named entity recognition unit and a relation extraction unit. The named entity recognition unit identifies and extracts target object entities, region entities, event entities, and time entities from the user's question; the relation extraction unit identifies the semantic relationships between these entities, thus forming a structured semantic representation of the question. In this way, the original natural language question is converted into a feature representation that the system can process, providing a unified data entry point for subsequent reasoning processes.

[0027] After completing the semantic parsing of the question, the system enters the intent recognition and strategy selection phase. Based on the constructed question feature representation, the intent recognition and strategy selection module classifies and identifies the question to determine whether it is a path-related question or a temporal evolution question. Path-related questions typically focus on the relationships between multiple entities and their potential impact paths, such as the association between a target object and a specific region or event; temporal evolution questions focus more on the changing trends and evolutionary patterns of entities or events over time. The system completes the above identification through the intent classification unit and automatically activates the matching reasoning strategy based on the identification results, thereby avoiding the use of a single fixed strategy to handle all questions and improving reasoning efficiency and accuracy.

[0028] When the system identifies a user's question as a path-related question, the reasoning path generation module employs a multi-hop path reasoning strategy. This strategy uses the core entities identified during the question analysis phase as seed nodes, and expands relationships and searches paths within the knowledge graph. Specifically, the system first locates the first-degree relationship nodes directly associated with the seed nodes, and then expands these relationships at multiple levels to form multiple candidate reasoning paths. Each reasoning path consists of multiple entity nodes and their relationships, reflecting the potential links between elements such as target objects, regions, and events, thus providing multiple possible explanation paths for the question.

[0029] When the system identifies a user's question as a temporal evolution problem, the reasoning path generation module employs a temporal reasoning strategy. This strategy first parses and standardizes the time representation in the question, converting the natural language time information into a unified time range or point-in-time representation. Subsequently, the system constructs dynamic, time-related relational data within a knowledge graph, aligning and integrating entity states and event records at different time points. Based on this, it generates reasoning paths reflecting the changes in entities or events over time through temporal relationship reasoning, thereby supporting the analysis of information such as trends and stage-specific characteristics.

[0030] After generating multiple candidate reasoning paths, the system enters the path evaluation and selection phase. The path evaluation and selection module employs a multi-dimensional weighted scoring mechanism to comprehensively evaluate each candidate reasoning path. Evaluation dimensions include at least logical consistency, information completeness, question relevance, and data support. Logical consistency is used to determine whether there are semantic conflicts between the nodes and their relationships within the path; information completeness measures whether the path covers the key entities and information elements required to answer the question; question relevance measures the degree of matching between the reasoning path and the user's question; and data support assesses the reliability level of the data corresponding to each relationship within the path.

[0031] During the implementation of the method, the overall credibility score of the candidate reasoning path is calculated according to the following formula: in, Scoring the logical consistency of the path. Score the information integrity of the path. Scoring the relevance of the path to the user's problem. Score the data support for the path. , , , These are the weighting coefficients for the corresponding evaluation dimensions.

[0032] The weighting coefficients can be adaptively adjusted according to the type of problem intent. For example, in path-related problems, the weights of logical consistency and relevance can be increased, and in time-series evolution problems, the weights of information integrity and data support can be increased, so that the scoring results are more in line with specific decision-making needs.

[0033] The system ranks the candidate reasoning paths based on the scoring results and selects the highest-scoring path as the target reasoning path. Subsequently, the result generation and visualization module generates the final answer based on the target reasoning path and displays the key nodes, relationships, and reasoning order in a structured manner, allowing users to intuitively understand the formation process of the reasoning conclusion. To further enhance the credibility and interpretability of the reasoning results, the result generation and visualization module can also connect to an external document library. When specific key nodes are involved in the reasoning path, the system can automatically trigger a document retrieval operation to supplement relevant background information and integrate the retrieval results with the reasoning path for display.

[0034] At the methodological level, the question-answering process executed by the aforementioned system specifically includes: extracting entity and semantic information from user questions; identifying question intent and selecting a reasoning strategy; generating multiple reasoning paths based on the selected strategy; evaluating the reasoning paths from multiple dimensions and selecting a target reasoning path; and generating an answer based on the target reasoning path and displaying the reasoning process. Through this process, the system can provide reliable analytical support for complex decision-making problems while ensuring transparency in the reasoning process and interpretability of the results.

[0035] Through the technical solution described in this embodiment, the system realizes a path reasoning mechanism driven by the user's question intent, effectively integrating knowledge graph structure information, time dimension information, and multi-dimensional evaluation mechanism. It avoids the problems of poor adaptability and untraceable results of traditional fixed strategy reasoning methods, and has good versatility and application value.

[0036] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, but the present invention is not limited to these embodiments. Equivalent modifications made by those skilled in the art without departing from the principles of the present invention should fall within the protection scope of the present invention.

[0037] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intent-driven path reasoning graph question-answering system for complex decision-making, characterized in that, include: The question parsing module is used to parse the user-input question based on the natural language processing model, extract the target object entity, region entity, event entity and time entity from the question, and construct the semantic representation of the question; the intent recognition and strategy selection module is used to generate question features based on the semantic representation, identify the intent type corresponding to the question, and select the inference strategy that matches the intent type from a variety of preset inference strategies. The reasoning path generation module is used to generate multiple candidate reasoning paths in the knowledge graph based on entity relationships and path constraints according to the selected reasoning strategy; the path evaluation and selection module is used to evaluate the candidate reasoning paths in multiple dimensions and select the target reasoning path; the result generation and visualization module is used to generate the final answer based on the target reasoning path and display the reasoning process in a structured manner.

2. The intent-driven path reasoning graph question-answering system for complex decision-making according to claim 1, characterized in that, The problem parsing module includes a named entity recognition unit and a relation extraction unit, which are used to extract target object entities, region entities, event entities and time entities from user problems, and construct a semantic representation framework for the problem.

3. The intent-driven path reasoning graph question-answering system for complex decision-making according to claim 1, characterized in that, The intent recognition and strategy selection module includes an intent classification unit, which is used to identify user questions as path-related questions or time-series evolution questions based on question characteristics, and activate the corresponding reasoning strategy accordingly.

4. The intent-driven path reasoning graph question-answering system for complex decision-making according to claim 3, characterized in that, When the problem is a path association problem, the reasoning strategy is a multi-hop path reasoning strategy, which includes seed node determination, relationship expansion and multi-path search steps, used to generate multiple candidate reasoning paths in the knowledge graph.

5. The intent-driven path reasoning graph question-answering system for complex decision-making according to claim 3, characterized in that, When the problem is a temporal evolution problem, the reasoning strategy is a temporal reasoning strategy, which includes temporal semantic parsing, temporal data construction and alignment, and temporal relationship reasoning steps, used to generate reasoning paths that reflect the changes of entities or events over time.

6. The intent-driven path reasoning graph question-answering system for complex decision-making according to claim 1, characterized in that, The path evaluation and selection module uses a multi-dimensional weighted scoring mechanism to comprehensively evaluate candidate reasoning paths. The multi-dimensional evaluation includes at least logical consistency, information completeness, question relevance, and data support.

7. The intent-driven path reasoning graph question-answering system for complex decision-making according to claim 6, characterized in that, In the multi-dimensional weighted scoring mechanism, the weight coefficients corresponding to each evaluation dimension are adaptively adjusted according to the intent type of the question.

8. The intent-driven path reasoning graph question-answering system for complex decision-making according to claim 1, characterized in that, The result generation and visualization module is also connected to an external document library, which is used to trigger retrieval operations based on key nodes in the reasoning path to supplement relevant background information and enhance the credibility and interpretability of the reasoning results.

9. A question-answering method for intent-driven path reasoning graphs for complex decision-making, characterized in that, Includes the following steps: Step 1: Extract entity and semantic information from the user's input question; Step 2: Identify the question intent based on the entity and semantic information, and select a reasoning strategy that matches the question intent; Step 3: Based on the selected reasoning strategy, generate multiple candidate reasoning paths in the knowledge graph; Step 4: Evaluate the candidate reasoning paths from multiple dimensions and select the target reasoning path; Step 5: Generate the answer based on the target reasoning path and display the reasoning process.

10. The method according to claim 9, characterized in that, In step four, a comprehensive credibility score is calculated for each candidate reasoning path, and the comprehensive credibility score is calculated according to the following formula: ;in, Scoring the logical consistency of the path. Score the information integrity of the path. Scoring the relevance of the path to the user's problem. Score the data support for the path. 、 、 、 These are the weighting coefficients for the corresponding evaluation dimensions.

Images