Search

CN-121996782-A - Method, system and storage medium for normalizing requirements among requirement documents

CN121996782ACN 121996782 ACN121996782 ACN 121996782ACN-121996782-A

Abstract

The invention discloses a method and a system for normalizing requirements among requirement documents, wherein the method comprises the following steps of atomic requirement extraction, atomic requirement blood edge analysis and depiction and requirement normalization; the system comprises atomic demand extraction, atomic demand blood margin analysis and depiction, demand normalization and other modules. According to the invention, atomic demands are extracted from an original demand document, inheritance and dependency relations among the atomic demands are constructed through Chinese semantic understanding capability of a large model, so that a blood-margin relation network of all the atomic demands of the demand document is formed in a converging manner, then the atomic demands with the blood-margin relation are fused from far to near according to the far-near sequence of the blood-margin relation to form an atomic demand package, then the atomic demand package is integrated and unified into a target container document, the normalization processing of the demand document is realized, the demand contents with the association relation with the current demand document are summarized to the current document step by step, and therefore, the reading of the current document is unnecessary to read other documents back and forth, the manual workload is reduced, and the working efficiency is improved.

Inventors

  • CHANG JIANG
  • LIN GUANFENG

Assignees

  • 中国民生银行股份有限公司

Dates

Publication Date
20260508
Application Date
20251215

Claims (10)

  1. 1. A demand normalization method between demand documents is characterized by comprising the following steps: S1, converting a docx-format demand document into an html-format document by using an engineering means aiming at the docx-format original demand document uploaded by a user, simultaneously keeping consistency of chapter structures, merging chapters of the html-format document into chapters suitable for atomic demands based on semantic understanding capability of a large model, analyzing atomic demand text content and merging adjusted chapter titles, storing by using a json format, and simultaneously converting unstructured demand document data into structured json format data; S2, summarizing the atomic demands of the demand document, calling a large model from the atomic demands together with two optional combinations and prompt words to analyze the blood edges between the atomic demands, outputting the blood edge relation between the two large models according to the blood edge pair format of source-target, and carrying out the process circularly until all the atomic demands are analyzed pairwise according to the confidence level, the blood edge description and the blood edge evidence of the blood edge relation between the two large models, and storing final data according to a graph data structure to form a context graph of the blood edge relation; S3, merging the dependent target demand content to the dependent source node according to the reverse order of step-by-step dependency based on the semantic understanding capability of the large model, performing text deduplication on the merged source node content, and maintaining statement smoothness and logic integrity consistency operation to achieve normalization.
  2. 2. The method for normalizing requirements between requirement documents according to claim 1, wherein the specific processing procedure of the large model in step S2 is as follows: S201, analyzing blood relationship between two atomic demands according to character setting, skill requirements, task targets, task steps, input and output data formats and limit requirements given by prompt words; S202, outputting the atomic requirements and the dependent atomic requirements, and confidence level, blood margin description and blood margin evidence about the blood margin relation between the atomic requirements and the dependent atomic requirements according to the blood margin pair format of source-target; s203, caching current blood edge pair data, and continuously analyzing the blood edge relation of the next pair of atomic demands until any two atomic demands complete blood edge analysis; s204, traversing all blood edge pair data, finding out the upstream and downstream atomic requirements of each atomic requirement, storing the upstream and downstream atomic requirements as a graph data structure of nodes and edges, and describing the blood edge relation among the atomic requirements.
  3. 3. The method for normalizing requirements among requirement documents according to claim 1, wherein the blood relationship comprises a dependency relationship and an inheritance relationship, the dependency relationship refers to an atomic requirement for one atomic requirement, or the completion of one atomic requirement is a precondition for the starting of another atomic requirement, or one atomic requirement and another atomic requirement share the same business process or business resource, and the inheritance relationship refers to the update and optimization of document version and related atomic requirement content, including the update of semantical version number, the requirement refinement and the content evolution caused by constraint condition change.
  4. 4. The method for normalizing requirements between requirement documents according to claim 2, wherein the specific steps of step S3 are as follows: S301, taking a requirement document uploaded by a user as a target container, traversing all atomic requirement blood relationship graph data of the requirement document, finding out the atomic requirements of all leaf nodes of the requirement document, creating a stack for each node, and pushing an original text of each node into the stack; S302, finding out the latest atomic demand text content in stacks corresponding to the current node and the parent node thereof, calling a large model together with a prompt word to fuse, and pushing the combined atomic demand content of the current node into the stacks; S303, judging whether the current node has a parent node or not, if yes, returning to the step S301, and if not, executing the step S304; S304, judging whether the leaf nodes are not merged, if yes, executing step S204, and if not, completing the merging and normalization process from the leaf nodes to the root nodes step by step.
  5. 5. The method for normalizing requirements between requirement documents according to claim 4, wherein the specific fusion process of the large model in step S302 is to merge leaf node atomic requirement content into parent node atomic requirement according to role setting, skill requirement, task target, task step, input and output data format and limit requirement given by prompt words, and output the atomic requirement content of the parent node after merging.
  6. 6. The method according to claim 1, wherein the original requirement document in step S1 includes a requirement and an atomic requirement included in the sub-requirement, and the atomic requirement includes a business description, an input/output, and a business process.
  7. 7. The method for normalizing requirements among requirement documents according to claim 1, wherein the large model training process is characterized in that web publications, books, literary works and discussions on social media are used as Chinese text training data in the training process, so that models are learned to grammar structures, vocabulary collocations and semantic features of Chinese, local and global dependency relations in texts are captured based on a Transformer framework, semantics in contexts are understood, fine adjustment is carried out on the models through a supervised learning method, the intention of users is accurately understood by the models, and replies conforming to human language habits and value observations are generated through introducing human feedback learning.
  8. 8. A demand normalization system between demand documents, comprising: The atomic demand extraction module is used for converting the docx-format demand document into an html-format document by using an engineering means aiming at the docx-format original demand document uploaded by a user, simultaneously keeping consistency of chapter structures, merging the chapters of the html-format document into chapters suitable for atomic demands based on semantic understanding capacity of a large model, analyzing out atomic demand text content and merging adjusted chapter titles, storing by using a json format, and simultaneously converting unstructured demand document data into structured json format data; The atomic demand blood margin analysis and depiction module is used for summarizing the atomic demands of the demand document, optionally calling two combinations and prompt words from the atomic demands together to analyze the blood margin between the atomic demands, outputting the blood margin relation between the two according to the blood margin pair format of source-target by the large model, and circularly executing the process until all the atomic demands are analyzed pairwise, and storing final data according to a graph data structure to form a context graph of the blood margin relation; The demand normalization module is used for merging the dependent target demand content into the dependent source node based on the semantic understanding capability of the large model according to the reverse order of step-by-step dependency, performing text deduplication on the merged source node content, maintaining statement sequence and complete consistency operation, and achieving normalization.
  9. 9. The method for normalizing requirements between requirement documents according to claim 8, wherein the large model training process is characterized in that web publications, books, literary works and discussions on social media are used as Chinese text training data in the training process, so that models are learned to grammar structures, vocabulary collocations and semantic features of Chinese, local and global dependency relations in texts are captured based on a Transformer framework, semantics in contexts are understood, fine adjustment is carried out on the models through a supervised learning method, the intention of users is accurately understood by the models, and replies conforming to human language habits and value views are generated through introducing human feedback learning.
  10. 10. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program when run by a processor performs the steps of the demand normalization method of any one of claims 1 to 7.

Description

Method, system and storage medium for normalizing requirements among requirement documents Technical Field The invention relates to the technical field of demand documents, in particular to a demand normalization method, a system and a storage medium between demand documents. Background Along with the rapid development of business of enterprises, the business scale is enlarged, the complexity of business requirements of an enterprise-level software system is improved, and remarkable difficulties are brought to business requirement analysis work, which are specifically shown in the following steps: 1. the increase of business functions and the increase of the complexity of the scenes such as rules, state circulation, input and output, exception processing and the like of the business functions lead to the drastic increase of the space of the required documents; 2. The association relationship between the inside of the requirement document and different documents (such as the mobile phone requirement, the PC requirement and the back requirement, and the first requirement and the second requirement) is in a net structure, and any single requirement modification can trigger manual tracing and updating of a large number of association points like pushing to dominoes, so that the requirement change analysis is extremely difficult. The improvement of the complexity of the service requirement brings huge workload to manual analysis, and is extremely easy to generate linkage errors due to omission, so that the accuracy and consistency of the requirement specification are difficult to maintain finally, even after the short-term completion, more workload is still required for long-term tracking of the requirement change influence, and the quality and efficiency of the requirement analysis are seriously reduced. There is currently no effective solution to the above problems. Disclosure of Invention Aiming at the technical problems in the related art, the invention provides a method, a system and a storage medium for normalizing requirements among requirement documents, which realize the work of automatically and intelligently completing the requirement analysis by engineering means by means of Chinese semantic understanding and text generating capability of a large model LLM and can overcome the defects in the prior art. In order to achieve the technical purpose, the technical scheme of the invention is realized as follows: in one aspect, a method for normalizing requirements between requirement documents is provided, including the following steps: S1, converting a docx-format demand document into an html-format document by using an engineering means aiming at the docx-format original demand document uploaded by a user, simultaneously keeping consistency of chapter structures, merging chapters of the html-format document into chapters suitable for atomic demands based on semantic understanding capability of a large model, analyzing atomic demand text content and merging adjusted chapter titles, storing by using a json format, and simultaneously converting unstructured demand document data into structured json format data; S2, summarizing the atomic demands of the demand document, calling a large model from the atomic demands together with two optional combinations and prompt words to analyze the blood edges between the atomic demands, outputting the blood edge relation between the two large models according to the blood edge pair format of source-target, and carrying out the process circularly until all the atomic demands are analyzed pairwise according to the confidence level, the blood edge description and the blood edge evidence of the blood edge relation between the two large models, and storing final data according to a graph data structure to form a context graph of the blood edge relation; S3, merging the dependent target demand content to the dependent source node according to the reverse order of step-by-step dependency based on the semantic understanding capability of the large model, performing text deduplication on the merged source node content, and maintaining statement smoothness and logic integrity consistency operation to achieve normalization. Further, the specific processing procedure of the large model in step S2 is as follows: S201, analyzing blood relationship between two atomic demands according to character setting, skill requirements, task targets, task steps, input and output data formats and limit requirements given by prompt words; S202, outputting the atomic requirements and the dependent atomic requirements, and confidence level, blood margin description and blood margin evidence about the blood margin relation between the atomic requirements and the dependent atomic requirements according to the blood margin pair format of source-target; s203, caching current blood edge pair data, and continuously analyzing the blood edge relation of the next pair of atomic demands until any two atomic de