CN-122021834-A - Industrial chain map construction method, device, equipment and storage medium

CN122021834ACN 122021834 ACN122021834 ACN 122021834ACN-122021834-A

Abstract

The invention provides an industrial chain map construction method, a device, equipment and a storage medium, which relate to the technical field of artificial intelligence, wherein the method comprises the steps of obtaining analysis demand information of a user; the method comprises the steps of inputting analysis requirement information into an industrial analyzer agent to obtain a capability problem tree output by the industrial analyzer agent, wherein the capability problem tree is obtained by decomposing the industrial analysis requirement by the industrial analyzer agent based on a recursive decomposition strategy, comprises a plurality of capability problems, one capability problem is used for defining a knowledge boundary of an entity related to a target industrial chain or determining a relation of the entity, and inputting the capability problem tree into the knowledge engineer agent to obtain an industrial chain map of the target industrial chain output by the knowledge engineer agent. By the method, the method can flexibly adapt to the diversified analysis requirements of users and the dynamically evolving industrial chain structure, so that the generated industrial chain map can accord with the user expectation, and the construction effect of the industrial chain map is guaranteed.

Inventors

ZHANG YUNLIANG
WANG LIJUN
LI LINNA
LI ZIYOU
WANG XIAOBO
GAN PING

Assignees

中国科学技术信息研究所

Dates

Publication Date: 20260512
Application Date: 20260407

Claims (10)

1. The industrial chain map construction method is characterized by comprising the following steps of: the method comprises the steps of acquiring analysis demand information of a user, wherein the analysis demand information comprises a target industry chain and industry analysis demands of the user on the target industry chain; The capability problem tree is obtained by decomposing the industrial analysis requirements by the industrial analyst intelligent agent based on a recursive decomposition strategy, and comprises a plurality of capability problems, wherein one capability problem is used for defining a knowledge boundary of an entity related to the target industrial chain or determining a relationship of the entity; And inputting the capability problem tree into a knowledge engineer agent, and obtaining an industrial chain map of the target industrial chain output by the knowledge engineer agent.
2. The industrial chain map construction method according to claim 1, wherein the capability problem tree is generated by the industrial analyst agent based on the steps of: determining a top-level capacity problem based on the analysis demand information, and taking the top-level capacity problem as a current decomposition level problem; Performing entity identification on the current decomposition level problem to determine a domain core entity, wherein the domain core entity is the entity related to the target industrial chain in the current decomposition level problem; inputting the current decomposition level problem and the domain core entity into a pre-trained text embedding model to obtain a text embedding vector output by the text embedding model; Based on the text embedded vector, carrying out similarity retrieval in a capability problem template library, and screening out a target capability problem template, wherein the target capability problem template is the capability problem template with the highest vector similarity with the text embedded vector; Generating a composite prompt word based on the current decomposition level problem, the domain core entity, the target capability problem template and a predefined capability problem type; Inputting the compound prompt word into a large language model to obtain a sub-capability problem of the current decomposition level problem output by the large language model; And taking the sub-capability problem or the top-level capability problem as the current decomposition level problem, returning to the step of carrying out entity identification on the current decomposition level problem, and determining a domain core entity until a preset problem granularity threshold is reached, and generating the capability problem tree.
3. The method for building an industrial chain map according to claim 1, wherein after the analysis requirement information is input to an industrial analyst agent to obtain a capability problem tree output by the industrial analyst agent, further comprising: The coverage is used for measuring the coverage degree of the capability problem tree on the industrial analysis requirement, the logic is used for measuring the hierarchical rationality of the capability problem tree, the executability is used for measuring the specificity and operability of leaf node problems, and the leaf node problems are the capability problems without sub-capability problems in the capability problem tree; if it is determined that the capability problem tree fails the quality assessment based on the coverage, the logic, and the executability, generating a feedback report; And correcting the capability problem tree by the industrial analyst agent based on the feedback report.
4. The industrial chain map construction method according to claim 1, wherein the industrial chain map is generated by the knowledge engineer agent based on the steps of: Decomposing the capability problem tree to generate at least two capability problem subtrees; Respectively carrying out planning retrieval processing, guided entity extraction processing and audit processing on each capability problem subtree to generate an industry chain sub-map corresponding to each capability problem subtree; And based on a large language model, iteratively fusing all the industrial chain sub-maps to generate the industrial chain map.
5. The method of claim 4, wherein one of the industrial chain sub-maps is generated by the knowledge engineer agent based on the steps of: carrying out semantic analysis on the capability problem subtrees to generate a knowledge retrieval strategy; corpus retrieval is carried out based on the knowledge retrieval strategy, and a comprehensive corpus corresponding to the capability problem subtree is generated; based on the capability problem subtree, performing guided entity extraction processing on the comprehensive corpus to generate an entity extraction result, and generating an initial industry chain sub-map based on the entity extraction result; performing audit processing on the initial industrial chain map to generate an audit report; if the structure of the initial industrial chain sub-map is complete and a first map gap with a priority higher than a preset threshold value does not exist based on the audit report, the initial industrial chain sub-map is used as the industrial chain sub-map; If the first map gap exists in the initial industry chain sub-map based on the audit report, updating the knowledge retrieval strategy based on the audit report, and returning to the step of carrying out corpus retrieval based on the knowledge retrieval strategy to generate a comprehensive corpus corresponding to the capability problem subtree; And if the initial industrial chain sub-map is determined to have only the second map gap with the priority lower than or equal to the preset threshold value based on the audit report, the initial industrial chain sub-map is used as the industrial chain sub-map, and optimization information is generated for the industrial chain sub-map.
6. The method of claim 1, wherein the inputting the capability problem tree into a knowledge engineer agent, after obtaining the industry chain map of the target industry chain output by the knowledge engineer agent, further comprises: Based on an evaluation intelligent agent, carrying out quality evaluation on the industrial chain map, and determining structural alignment, hierarchical rationality, node accuracy and node redundancy of the industrial chain map, wherein the structural alignment is used for measuring logical consistency of the industrial chain map and the capability problem tree, the hierarchical rationality is used for measuring consistency of relationships of all entity nodes in the industrial chain map and real industrial chain logic, the node accuracy is used for measuring authenticity of all entity nodes and correlation of all entity nodes and the target industrial chain, and the node redundancy is used for measuring semantic repeatability of all entity nodes in the industrial chain map; If it is determined that the industry chain graph fails the quality assessment based on the structural alignment, the hierarchical rationality, the node accuracy, and the node redundancy, generating a correction report; And correcting the industrial chain map by the knowledge engineer agent based on the correction report.
7. The method according to claim 5, wherein the step of inputting the capability problem tree to a knowledge engineer agent and obtaining the industry chain map of the target industry chain output by the knowledge engineer agent further comprises: Performing generalization processing on the capability problem tree based on the evaluation agent to generate a first forward paradigm; Based on the audit report, extracting a positive entity extraction result and a negative entity extraction result, taking the positive entity extraction result as a second positive example, and taking the negative entity extraction result as a negative example; The first positive example, the second positive example, and the negative example are stored to an experience library.
8. An industrial chain map construction apparatus, comprising: The system comprises an acquisition module, a storage module and a processing module, wherein the acquisition module is used for acquiring analysis demand information of a user, and the analysis demand information comprises a target industrial chain and an industrial analysis demand of the user on the target industrial chain; The capability problem tree construction module is used for inputting the analysis requirement information to an industrial analyst intelligent body to obtain a capability problem tree output by the industrial analyst intelligent body, wherein the capability problem tree is obtained by decomposing the industrial analysis requirement by the industrial analyst intelligent body based on a recursive decomposition strategy, and comprises a plurality of capability problems, and one capability problem is used for defining a knowledge boundary of an entity related to the target industrial chain or determining the relationship of the entity; And the map construction module is used for inputting the capability problem tree into a knowledge engineer agent and obtaining an industrial chain map of the target industrial chain output by the knowledge engineer agent.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and running on the processor, wherein the processor implements the industrial chain map construction method of any one of claims 1 to 7 when the computer program is executed by the processor.
10. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the industrial chain map construction method according to any one of claims 1 to 7.

Description

Industrial chain map construction method, device, equipment and storage medium Technical Field The present invention relates to the field of artificial intelligence technologies, and in particular, to a method, an apparatus, a device, and a storage medium for constructing an industrial chain map. Background The industrial chain map is mainly used for organizing and describing knowledge elements and associations thereof in a specific industrial chain, and comprises a plurality of entity nodes related to the industrial chain and relations among the entity nodes, wherein the relations among the entity nodes generally comprise an upper-lower relation and an upper-lower relation, and the relations connect the entity nodes to represent a complete industrial chain logic architecture. At present, the construction methods of the industrial chain atlas can be divided into two types, namely an industrial chain atlas construction method based on artificial deconstruction and an industrial chain atlas construction method based on natural language processing. The industrial chain map construction method based on artificial deconstruction is characterized in that expert knowledge is taken as a core, and industrial chain maps of specific industrial chains are constructed from top to bottom, and the method has the core advantage of being capable of deeply fusing the experience of the field expert, so that the generated industrial chain map has higher guarantee on the accuracy of macroscopic logic and core links, but the method is seriously dependent on the cognitive boundaries of a few experts, and the construction period of the industrial chain map is long and the cost is high, so that the construction efficiency of the industrial chain map is low. In order to improve the construction efficiency of the industrial chain map, the industrial chain map construction method based on natural language processing is gradually popularized and applied. The method for constructing the industrial chain atlas based on natural language processing refers to that the natural language processing technology is utilized, constraint extraction of entities is carried out from a corpus according to manual rules constructed in advance, and then the industrial chain atlas is generated according to entity extraction results, but the method still needs to rely on manual rules predefined by experts to carry out entity extraction, so that the generated industrial chain atlas mode is solidified, the analysis requirements of users in various and the industrial chain structure of dynamic evolution are difficult to flexibly adapt, and the construction effect of the industrial chain atlas is poor. For example, a user located in a specific area may only pay attention to the information of an industrial chain located in the specific area, a user focusing on researching a specific technical branch of an industrial chain may only pay attention to the information of an industrial chain of the specific technical branch of the industrial chain, while the existing industrial chain map construction method based on natural language processing is difficult to consider the actual analysis requirements of different users, and only a general industrial chain map cured by a mode can be generated according to a predefined manual rule, which easily causes that the generated industrial chain map does not conform to the expectations of the user. In summary, the existing industrial chain map construction method is difficult to flexibly adapt to the diversified analysis requirements of users and the dynamically evolving industrial chain structure, and the construction effect of the industrial chain map is poor. Disclosure of Invention The invention provides an industrial chain map construction method, an industrial chain map construction device, industrial chain map construction equipment and a storage medium, which are used for solving the defects that the existing industrial chain map construction method is difficult to flexibly adapt to the diversified analysis requirements of users and the dynamically-evolved industrial chain structure and the construction effect of the industrial chain map is poor. The invention provides an industrial chain map construction method which comprises the steps of obtaining analysis requirement information of a user, inputting the analysis requirement information into an industrial analyzer intelligent body to obtain a capability problem tree output by the industrial analyzer intelligent body, wherein the capability problem tree is obtained by decomposing the industrial analysis requirement by the industrial analyzer intelligent body based on a recursive decomposition strategy and comprises a plurality of capability problems, one capability problem is used for defining a knowledge boundary of an entity related to the target industrial chain or determining the relation of the entity, and inputting the capability problem tree into the knowledge