CN-121981217-A - Construction method of multi-agent collaborative TRIZ contradiction matrix fused with DIKWP model

CN121981217ACN 121981217 ACN121981217 ACN 121981217ACN-121981217-A

Abstract

The invention discloses a method for constructing a multi-agent collaborative TRIZ contradiction matrix fused with DIKWP models, which relates to an artificial intelligence technology and aims at solving the problems that a closed loop is difficult to form and the like in the prior art. A processing pipeline composed of four intelligent agents with specific and cooperative functions is constructed, the original data of a technical file is automatically converted into structured contradiction matrix knowledge, the four intelligent agents are respectively an intelligent agent for identifying contradiction sentence pairs, an intelligent agent for mapping parameters, an intelligent agent for extracting technical solutions and an intelligent agent for constructing contradiction matrix, the technical file is used as the original data, the contradiction sentence pairs representing technical opposition are extracted as information after the four intelligent agents are processed, the information is mapped into standard engineering parameters, and the contradiction matrix is constructed, so that reusable structured knowledge is formed. The method has the advantages of high automation and efficiency improvement, high identification precision and interpretability and strong systematicness and logic closed loop.

Inventors

LEI FANGYUAN
LIANG MINJING
DAI QINGYUN
ZHUANG CHUXIN
HE ZHENYU

Assignees

广东技术师范大学

Dates

Publication Date: 20260505
Application Date: 20251205

Claims (10)

1. A method for constructing a multi-agent collaborative TRIZ contradictory matrix fused with DIKWP models is characterized in that a processing pipeline consisting of four agents with specific functions and collaborative work is constructed, and the original data of a technical file is automatically converted into structured contradictory matrix knowledge; the four agents are respectively an identification agent of a contradictory sentence pair, a parameter mapping agent, an agent extracted by a technical solution and an agent constructed by a contradictory matrix; The technical file is used as original data, after four agents process, contradictory sentence pairs which characterize the technical opposition are extracted as information, the information is mapped into standard engineering parameters and contradictory matrixes are constructed, and reusable structured knowledge is formed.
2. The method for constructing the multi-agent collaborative TRIZ contradiction matrix by fusing DIKWP models according to claim 1 is characterized in that the contradiction sentence pair identification agent is used for automatically identifying and extracting sentence pairs expressing technical contradictions from an input technical file, the sentence pairs describing improved sentences and deteriorated sentences are positioned and described according to a pre-designed prompt word template by utilizing the context understanding capability of a plurality of large language models, and the output of the sentence pairs is a structured contradiction sentence pair list and is used as a downstream parameter mapping agent and a technical solution to extract the input of the agent.
3. The method for constructing the multi-agent collaborative TRIZ contradictory matrix by fusing DIKWP models according to claim 2, wherein the technical document is a published patent document.
4. The method for constructing the multi-agent collaborative TRIZ contradiction matrix by fusing DIKWP models according to claim 2, wherein the contradiction sentence pair identification agent comprises a conflict detection and resolution module for removing redundant contradiction pairs, and an intelligent evaluation and verification module is provided for calling an evaluation type large language model to verify the logical consistency and TRIZ consistency of sentence pairs and ensure the output quality.
5. The method for constructing the multi-agent collaborative TRIZ contradiction matrix by fusing DIKWP models according to claim 2 is characterized in that the parameter mapping agent is used for mapping natural language contradiction description output by the contradiction sentence pair identification agent to standard engineering parameters in TRIZ theory, distinguishing technical contradiction and physical contradiction through a contradiction classifier, mapping an improvement sentence and a deterioration sentence into an improvement parameter and a deterioration parameter respectively by combining a large language model with a TRIZ parameter knowledge base for the technical contradiction, and transmitting the output structured parameter pair to a contradiction matrix to construct the intelligent body for integration.
6. The method for constructing the multi-agent collaborative TRIZ contradiction matrix by fusing DIKWP models according to claim 5, wherein the parameter mapping agent comprises a parameter conflict detection and resolution module for filtering irrelevant or redundant parameter mappings, and a multi-round verification mechanism is provided for ensuring the accuracy of the parameter mappings.
7. The method for constructing the multi-agent collaborative TRIZ contradiction matrix by fusing DIKWP models according to claim 5, wherein the technical solution extraction agent is used for extracting a technical solution corresponding to the identified contradiction sentence pair from a technical document, firstly receiving the contradiction sentence pair from the identified agent, locating and extracting a specific technical solution description for solving the contradiction in the technical document based on the contradiction context, directly referencing the original text of the outputted technical solution and attaching a position label, and transmitting the technical solution as key evidence and content to the contradiction matrix construction agent for principle induction.
8. The method for constructing the multi-agent collaborative TRIZ contradiction matrix by fusing DIKWP models according to claim 7, wherein the technical solution extraction agent comprises a conflict detection and resolution module of the solution for deduplication and filtering, and the accuracy and the correlation of the extraction solution are verified through multiple rounds of evaluation.
9. The method for constructing the multi-agent collaborative TRIZ contradiction matrix by fusing DIKWP models according to claim 7, wherein the contradiction matrix construction agent is used for integrating contradiction sentence pairs, identifying agent by parameter mapping and extracting agent by technical solution to construct a final structured TRIZ contradiction matrix, firstly receiving improved parameter-deteriorated parameter pairs by parameter mapping agent output and technical solution by technical solution to extract agent output, calling a large language model to query TRIZ contradiction matrix, identifying the invention principle applicable to the current parameter pairs and technical solution for technical contradiction, identifying the separation principle applicable to the current parameter pairs and technical solution for physical contradiction, and finally organizing the improved parameter-deteriorated parameter-invention principle into a triplet form to construct a visual and traceable contradiction matrix.
10. The method for constructing the multi-agent collaborative TRIZ contradiction matrix by fusing DIKWP models according to claim 9, wherein the contradiction matrix constructing agent comprises a technical principle verification mechanism, so that the recommended invention principle is guaranteed to be circulated in the technical solution, and rationality and consistency of principle recommendation are guaranteed through multiple rounds of iteration.

Description

Construction method of multi-agent collaborative TRIZ contradiction matrix fused with DIKWP model Technical Field The invention relates to an artificial intelligence technology, in particular to a method for constructing a multi-agent collaborative TRIZ contradiction matrix fused with DIKWP models. Background The traditional TRIZ contradiction matrix construction method is highly dependent on manual analysis and induction of domain experts. Along with the rapid increase of patent data volume and the trend of complex data sources, the method has the inherent defects of firstly, efficiency bottleneck, time and labor consumption for manually processing massive patent texts, difficulty in meeting the requirement of modern research and development on efficiency, secondly, consistency, and poor reproducibility, which are caused by the fact that the built matrix standard is different in subjective judgment of experts. In order to improve the automation level, the prior art introduces natural language processing and a single agent method, for example, a single large language model is adopted to extract and map the end-to-end contradiction. However, such methods have structural limitations in that complex contradictory recognition and matrix construction tasks are compressed into a single reasoning process, and a task decomposition and phase coordination mechanism is lacking. This results in insufficient recognition accuracy of the system in the face of implicit semantics and complex logic, often fragmented output results and poor stability, and difficulty in forming a contradictory matrix with logic closed loop and strong interpretability. Disclosure of Invention The invention aims to provide a method for constructing a multi-agent collaborative TRIZ contradiction matrix fused with DIKWP models, so as to solve the problems in the prior art. According to the method for constructing the multi-agent collaborative TRIZ contradiction matrix by fusing DIKWP models, a processing pipeline composed of four agents with specific and collaborative functions is constructed, and the original data of a technical file is automatically converted into structured contradiction matrix knowledge; the four agents are respectively an identification agent of a contradictory sentence pair, a parameter mapping agent, an agent extracted by a technical solution and an agent constructed by a contradictory matrix; The technical file is used as original data, after four agents process, contradictory sentence pairs which characterize the technical opposition are extracted as information, the information is mapped into standard engineering parameters and contradictory matrixes are constructed, and reusable structured knowledge is formed. The method for constructing the multi-agent collaborative TRIZ contradiction matrix fused with DIKWP model has the advantages that: (1) The method has the advantages of high automation and efficiency improvement, realizes full-flow automation from contradiction identification to matrix construction based on patent original data through multi-agent division cooperation, thoroughly gets rid of heavy dependence on manual expert, and can cope with large-scale patent data analysis. (2) The recognition accuracy and the interpretability are high, a DIKWP model is fused, a hierarchical semantic understanding framework is provided for contradiction recognition, and the preliminary result is subjected to feedback correction through the cooperation among multiple intelligent agents and a multi-round verification mechanism, so that the accuracy of complex technical contradiction recognition and the traceability of the result are remarkably improved. (3) The system and logic closed loop are strong, the complex matrix construction task is decomposed into four intelligent agent stages with complementary functions, and the four intelligent agent stages are connected in series through information flow, so that a logic tight processing closed loop is formed, the fragmentation problem caused by one-time reasoning of a single intelligent agent is avoided, and the output contradictory matrix has high structuring degree. Drawings FIG. 1 is a schematic flow chart of a method for constructing a multi-agent collaboration TRIZ contradiction matrix according to the present invention. Detailed Description As shown in FIG. 1, the method for constructing the multi-Agent collaborative TRIZ contradiction matrix fused with DIKWP model automatically and systematically converts the original Data of the technical file into the structured contradiction matrix Knowledge knowledges by constructing a processing pipeline consisting of four Agent TRIZ-agents which are specially and cooperatively operated. In one embodiment, the technical document is published patent text. Firstly, a DIKWP model is innovatively used as a top-level guiding framework to be used as a complete path of knowledge evolution all the time, a technical file is used as original data,