Demand normalization method and system between demand documents and storage medium

Abstract

The invention discloses a demand normalization method and system between demand documents. The method comprises the following steps of atomic demand extraction, atomic demand blood relationship analysis and description and demand normalization. The system comprises an atomic demand extraction module, an atomic demand blood relationship analysis and description module, a demand normalization module and the like. According to the method, atomic demands are extracted from an original demand document, inheritance and dependency relationships among the atomic demands are constructed through Chinese semantic understanding ability of a large model, so that a blood relationship network of all the atomic demands of the demand document is formed through convergence, and then the atomic demands are obtained according to the distance sequence of the blood relationship. The method comprises the following steps: fusing atomic demands with a blood relationship from far to near to form an atomic demand package, integrating and unifying the atomic demand package into a target container document to realize normalization processing of a demand document, summarizing demand contents with an association relationship with the current demand document to a current document step by step, and storing the demand contents in the target container document. Therefore, other documents do not need to be browsed back and forth when the current document is read, the manual workload is reduced, and the working efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of requirements document technology, and more specifically, to a method, system, and storage medium for requirements normalization among requirements documents. Background Technology

[0002] As businesses rapidly expand, the complexity of business requirements for enterprise-level software systems also increases, significantly complicating business requirements analysis. Specifically, this manifests in the following ways: 1. Due to the increase in business functions, and the increased complexity of business function rules, state transitions, input and output, exception handling, etc., the length of the requirements document has increased dramatically. 2. The relationships within the requirements document and between different documents (such as mobile requirements, PC requirements and backend requirements, or Phase 1 requirements and Phase 2 requirements) form a network structure. Any modification to a single requirement may trigger a large number of related points to be manually traced and updated, making requirement change analysis extremely difficult.

[0003] This increased complexity of business requirements brings a huge workload to manual analysis and is prone to chain errors due to oversights. Ultimately, it makes it difficult to maintain the accuracy and consistency of requirement specifications. Even if completed in the short term, tracking the impact of requirement changes in the long term still requires a lot of work, which seriously reduces the quality and efficiency of requirement analysis.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] To address the aforementioned technical problems in related technologies, this invention proposes a method, system, and storage medium for requirement normalization among requirement documents. By leveraging the Chinese semantic understanding and text generation capabilities of Large Model LLM, and through engineering methods, the requirement analysis work can be completed automatically and intelligently, thus overcoming the aforementioned shortcomings of existing technologies.

[0006] To achieve the above-mentioned technical objectives, the technical solution of the present invention is implemented as follows: On the one hand, a method for requirement normalization across requirement documents is provided, including the following steps: S1 uses engineering methods to convert the original docx format requirement documents uploaded by users into html format documents while maintaining the consistency of the chapter structure. Based on the semantic understanding capabilities of the large model, the chapters of the html format document are merged into chapters suitable for atomic requirements. The text content of the atomic requirements and the merged and adjusted chapter titles are parsed out and stored in JSON format. At the same time, the unstructured requirement document data is converted into structured JSON format data. S2 summarizes the atomic requirements in the requirements document, selects any two pairs of atomic requirements and prompt words together to call the large model to analyze the lineage between atomic requirements. The large model outputs the lineage relationship between the two in the format of source-target lineage pair, as well as the confidence, lineage description and lineage evidence of the lineage relationship between the two. This process is repeated until all atomic requirements have been analyzed pairwise. The final data is stored in the form of graph data structure to form a lineage relationship map. S3 merges related atomic requirements into dependent source nodes in reverse order of hierarchical dependency, based on the semantic understanding capability of the large model. Then, it performs text deduplication, sentence fluency, and logical integrity consistency operations on the merged source node content to achieve normalization.

[0007] Furthermore, the specific processing procedure for the large model in step S2 is as follows: S201 analyzes the lineage relationship between two atomic requirements based on the role settings, skill requirements, task objectives, task steps, input / output data formats, and constraints given in the prompts. S202 outputs the atomic requirements and dependent atomic requirements in the source-target lineage pair format, along with the confidence level, lineage description, and lineage evidence regarding the lineage relationship between them. S203 caches the current lineage pair data and continues to analyze the lineage relationship of the next pair of atomic requirements until any two atomic requirements have completed lineage analysis; S204 iterates through all lineage pairs, finds the upstream and downstream atomic requirements of each atomic requirement, and saves them as a graph data structure of nodes and edges to depict the lineage relationship between atomic requirements.

[0008] Furthermore, the lineage relationship includes dependency and inheritance; the dependency relationship is that one atomic requirement refers to another atomic requirement, or the completion of one atomic requirement is a prerequisite for the initiation of another atomic requirement, or one atomic requirement and another atomic requirement share the same business process or business resource; the inheritance relationship refers to the updating and optimization of document versions and related atomic requirement content, including the updating of semantic version numbers, requirement refinement, and content evolution caused by changes in constraints.

[0009] Furthermore, the specific steps of step S3 are as follows: S301 uses the user-uploaded requirement document as the target container, traverses all its atomic requirement lineage relationship graph data, finds the atomic requirements of all its leaf nodes, and creates a stack for each node, pushing the original text of each node onto the stack. S302 finds the latest atomic requirement text content in the stack corresponding to the current node and its parent node, calls the large model to merge it together with the prompt word, and pushes the merged atomic requirement content of the current node onto the stack. S303 Determine if the current node has a parent node. If yes, return to step S301; otherwise, proceed to step S304. S304 Determine if there are any leaf nodes that have not been merged. If so, proceed to step S204. Otherwise, complete the merging and unification process from the leaf nodes to the root node.

[0010] Furthermore, the specific fusion process of the large model in step S302 is as follows: according to the role setting, skill requirements, task objectives, task steps, input and output data formats, and restrictions given by the prompt words, the atomic requirements of the leaf nodes are merged into the atomic requirements of their parent nodes, and the atomic requirements of the merged parent nodes are output.

[0011] Furthermore, the original requirements document in step S1 includes the requirements and the atomic requirements contained in the sub-requirements. The atomic requirements include business descriptions, inputs and outputs, and business processes.

[0012] Furthermore, the large model training process is as follows: during training, publicly available online texts, books, literary works, and discussions on social media are used as Chinese text training data to enable the model to learn Chinese grammatical structures, vocabulary collocations, and semantic features; based on the Transformer architecture, the model captures local and global dependencies in the text to understand the semantics in the context; the model is fine-tuned through supervised learning to accurately understand the user's intent; and responses that conform to human language habits and values ​​are generated by introducing human feedback.

[0013] On the other hand, a requirements normalization system is provided among requirements documents, including: The atomic requirement extraction module is used to convert the original docx format requirement documents uploaded by users into html format documents using engineering methods, while maintaining the consistency of the chapter structure. Based on the semantic understanding capabilities of the large model, the chapters of the html format document are merged into chapters suitable for atomic requirements, the atomic requirement text content and the merged and adjusted chapter titles are parsed out and stored in JSON format, while converting unstructured requirement document data into structured JSON format data. The atomic requirement lineage analysis module is used to summarize the atomic requirements in the requirement document. It selects any two pairs of atomic requirements and prompt words to call the large model to analyze the lineage between atomic requirements. The large model outputs the lineage relationship between the two in the format of source-target lineage pair, as well as the confidence, lineage description and lineage evidence of the lineage relationship between the two. This process is repeated until all atomic requirements have been analyzed pairwise. The final data is stored in the form of graph data structure to form a lineage relationship map. The requirement normalization module is used to merge the content of dependent target requirements into dependent source nodes in reverse order of hierarchical dependencies, based on the semantic understanding capability of the large model. The module then performs text deduplication, sentence fluency, and logical integrity operations on the merged source node content to achieve normalization.

[0014] Furthermore, the large model training process is as follows: during training, publicly available online texts, books, literary works, and discussions on social media are used as Chinese text training data to enable the model to learn Chinese grammatical structures, vocabulary collocations, and semantic features; based on the Transformer architecture, the model captures local and global dependencies in the text to understand the semantics in the context; the model is fine-tuned through supervised learning to accurately understand the user's intent; and responses that conform to human language habits and values ​​are generated by introducing human feedback.

[0015] On the other hand, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, performs the steps of the demand normalization method according to any one of claims 1 to 7.

[0016] The beneficial effects of this invention are as follows: This invention extracts atomic requirements from the original requirements document. Through the Chinese semantic understanding capabilities of the large model, it constructs the inheritance and dependency relationships between atomic requirements, thereby converging to form a bloodline network of all atomic requirements in the requirements document. Then, according to the closeness of the bloodline relationship, atomic requirements with bloodline relationships are merged from farthest to closest to form atomic requirement packages. These atomic requirement packages are then integrated and unified into the target container document, achieving normalization of the requirements document. Finally, a "single authoritative requirements baseline" is formed, which summarizes the requirements content related to the current requirements document level by level into the current document. Thus, when reading the current document, there is no need to flip back and forth between other documents, reducing manual workload and improving work efficiency. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the requirement normalization method between requirement documents according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the atomic requirement extraction stage of the requirement normalization method between requirement documents according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the requirement normalization stage of the requirement normalization method between requirement documents according to an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0020] like Figure 1-3 As shown in the embodiment of the present invention, a method for normalizing requirements among requirements documents is provided, comprising the following steps: S1 uses engineering methods to convert the original docx format requirement documents uploaded by users into html format documents while maintaining the consistency of the chapter structure. Based on the semantic understanding capabilities of the large model, the chapters of the html format document are merged into chapters suitable for atomic requirements. The text content of the atomic requirements and the merged and adjusted chapter titles are parsed out and stored in JSON format. At the same time, the unstructured requirement document data is converted into structured JSON format data. S2 summarizes the atomic requirements in the requirements document, selects any two pairs of atomic requirements and prompt words together to call the large model to analyze the lineage between atomic requirements. The large model outputs the lineage relationship between the two in the format of source-target lineage pair, as well as the confidence, lineage description and lineage evidence of the lineage relationship between the two. This process is repeated until all atomic requirements have been analyzed pairwise. The final data is stored in the form of graph data structure to form a lineage relationship map. S3 merges related atomic requirements into dependent source nodes in reverse order of hierarchical dependency, based on the semantic understanding capability of the large model. Then, it performs text deduplication, sentence fluency, and logical integrity consistency operations on the merged source node content to achieve normalization.

[0021] Preferably, the specific processing procedure for the large model in step S2 is as follows: S201 analyzes the lineage relationship between two atomic requirements based on the role settings, skill requirements, task objectives, task steps, input / output data formats, and constraints given in the prompts. S202 outputs the atomic requirements and dependent atomic requirements in the source-target lineage pair format, along with the confidence level, lineage description, and lineage evidence regarding the lineage relationship between them. S203 caches the current lineage pair data and continues to analyze the lineage relationship of the next pair of atomic requirements until any two atomic requirements have completed lineage analysis; S204 iterates through all lineage pairs, finds the upstream and downstream atomic requirements of each atomic requirement, and saves them as a graph data structure of nodes and edges to depict the lineage relationship between atomic requirements.

[0022] Preferably, the lineage relationship includes dependency and inheritance; the dependency relationship is that one atomic requirement refers to another atomic requirement, or the completion of one atomic requirement is a prerequisite for the initiation of another atomic requirement, or one atomic requirement and another atomic requirement share the same business process or business resource; the inheritance relationship refers to the updating and optimization of document version and related atomic requirement content, including the updating of semantic version numbers, requirement refinement, and content evolution caused by changes in constraints.

[0023] Preferably, the specific steps of step S3 are as follows: S301 uses the user-uploaded requirement document as the target container, traverses all its atomic requirement lineage relationship graph data, finds the atomic requirements of all its leaf nodes, and creates a stack for each node, pushing the original text of each node onto the stack. S302 finds the latest atomic requirement text content in the stack corresponding to the current node and its parent node, calls the large model to merge it together with the prompt word, and pushes the merged atomic requirement content of the current node onto the stack. S303 Determine if the current node has a parent node. If yes, return to step S301; otherwise, proceed to step S304. S304 Determine if there are any leaf nodes that have not been merged. If so, proceed to step S204. Otherwise, complete the merging and unification process from the leaf nodes to the root node.

[0024] Preferably, the specific fusion process of the large model in step S302 is as follows: according to the role setting, skill requirements, task objectives, task steps, input and output data formats, and restriction requirements given by the prompt words, the atomic requirements of the leaf nodes are merged into the atomic requirements of their parent nodes, and the atomic requirements of the merged parent nodes are output.

[0025] Preferably, the original requirement document in step S1 includes requirements and atomic requirements contained in sub-requirements. Atomic requirements include business descriptions, inputs and outputs, and business processes.

[0026] Preferably, the large model training process is as follows: during training, publicly available online texts, books, literary works, and discussions on social media are used as Chinese text training data to enable the model to learn Chinese grammatical structures, vocabulary collocations, and semantic features; based on the Transformer architecture, the model captures local and global dependencies in the text to understand the semantics in the context; the model is fine-tuned through supervised learning to accurately understand the user's intent; and responses that conform to human language habits and values ​​are generated by introducing human feedback.

[0027] A requirements normalization system among requirements documents includes: The atomic requirement extraction module is used to convert the original docx format requirement documents uploaded by users into html format documents using engineering methods, while maintaining the consistency of the chapter structure. Based on the semantic understanding capabilities of the large model, the chapters of the html format document are merged into chapters suitable for atomic requirements, the atomic requirement text content and the merged and adjusted chapter titles are parsed out and stored in JSON format, while converting unstructured requirement document data into structured JSON format data. The atomic requirement lineage analysis module is used to summarize the atomic requirements in the requirement document. It selects any two pairs of atomic requirements and prompt words to call the large model to analyze the lineage between atomic requirements. The large model outputs the lineage relationship between the two in the format of source-target lineage pair, as well as the confidence, lineage description and lineage evidence of the lineage relationship between the two. This process is repeated until all atomic requirements have been analyzed pairwise. The final data is stored in the form of graph data structure to form a lineage relationship map. The requirement normalization module is used to merge the content of dependent target requirements into dependent source nodes in reverse order of hierarchical dependencies, based on the semantic understanding capability of the large model. The module then performs text deduplication, sentence fluency, and logical integrity operations on the merged source node content to achieve normalization.

[0028] Preferably, the large model training process is as follows: during training, publicly available online texts, books, literary works, and discussions on social media are used as Chinese text training data to enable the model to learn Chinese grammatical structures, vocabulary collocations, and semantic features; based on the Transformer architecture, the model captures local and global dependencies in the text to understand the semantics in the context; the model is fine-tuned through supervised learning to accurately understand the user's intent; and responses that conform to human language habits and values ​​are generated by introducing human feedback.

[0029] A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to perform the steps of the demand normalization method according to any one of claims 1 to 7.

[0030] To facilitate understanding of the above-mentioned technical solutions of this utility model, the following detailed description of the above-mentioned technical solutions of this utility model is provided through specific usage methods.

[0031] First, since this invention is based on the Chinese semantic understanding and generation capabilities of a large-scale model, the principles behind this capability will be explained. Firstly, the large-scale model utilizes a large amount of Chinese text data during training, including publicly available online texts, books, literary works, and discussions on social media. This data covers various expressions in the Chinese language, including formal and informal styles, enabling the model to learn Chinese grammatical structures, lexical collocations, and semantic features. Secondly, it is based on the Transformer architecture, with self-attention as its core mechanism. This mechanism captures local and global dependencies in the text, thus more accurately understanding the semantics within the context. Thirdly, the model is fine-tuned through supervised learning (SFT) to more accurately understand the user's intent. Furthermore, by introducing human feedback (RLHF), the model can learn how to generate responses that better align with human language habits and values. Therefore, the large-scale model can effectively accomplish tasks involving in-depth understanding and generation of Chinese semantics.

[0032] Secondly, since this invention originates in an enterprise-level software development environment, its premise is that requirements management is well-standardized. Enterprise-level project requirements documents often use a unified template format. In a typical template, the requirements document can consist of sub-requirements and the atomic requirements contained within those sub-requirements. An atomic requirement is the finest-grained expression of a functional requirement, generally composed of indivisible parts such as "business description," "input / output," and "business process." Between atomic requirements, there are two types of relationships: dependency and inheritance. For example, if an atomic requirement description uses words like "see also" or "refer to," it indicates a direct dependency with another atomic requirement. Alternatively, if the completion of one atomic requirement is a prerequisite for the initiation of another, a logical dependency exists. Furthermore, if one atomic requirement shares the same business process or business resource with another, a resource dependency exists. Inheritance refers to document version updates and optimizations of related atomic requirement content, including semantic version number updates, requirement refinement, and content evolution due to changes in constraints. Therefore, enterprise-level requirement documents can be viewed as a combination of atomic requirements, and the relationships between requirement documents can be viewed as a bloodline network composed of dependencies and inheritance relationships between "source atomic requirements" and "target atomic requirements".

[0033] like Figure 2 As shown, by breaking down the original requirement document into sub-requirements and atomic requirements, and using the Chinese semantic analysis capabilities of the large model to analyze the relationships between any two atomic requirements, "source atomic requirements" and "target atomic requirements" lineage pairs are formed. All lineage pairs are then aggregated, and a lineage network is depicted according to the "source -> target" direction. At this point, for the target document, its contained atomic requirements, as well as other related atomic requirements, have been identified through lineage relationships. Then, following the reverse direction of the lineage relationships, atomic requirements are merged level by level from farthest to closest, ultimately achieving the normalization of the target document's atomic requirements. The merging process is also carried out using the Chinese semantic understanding and text generation capabilities of the large model. During merging, the original text of the atomic requirements should be preserved as much as possible, but the large model is required to perform deduplication and maintain language coherence. The overall process is as follows: Figure 3 As shown.

[0034] In this embodiment, the main process implemented through engineering methods based on the domestically produced Qwen series general-purpose large model consists of three major steps: 1. Atomic Requirements Extraction Stage. From the user-uploaded original docx format requirement document, engineering methods are used to convert the docx format requirement document into an HTML format document, while maintaining the consistency of the chapter structure. Then, based on the semantic understanding capabilities of the large model, the original document chapters are merged into chapters suitable for atomic requirements, thereby parsing out the atomic requirement text content and the merged and adjusted chapter titles, and storing them in JSON format, converting the unstructured requirement document data into structured JSON format data for use in subsequent steps.

[0035] 2. Atomic Requirement Lineage Analysis and Characterization Stage. The main task of this stage is to summarize the atomic requirements in the requirement document, randomly select two pairs from the atomic requirements, and call the large model along with the prompt words to analyze the lineage between the atomic requirements. The prompt words provide the role settings, skill requirements, task objectives, task steps, input and output data formats, and constraints of the large model for this analysis. This allows the large model to output, according to the source-target lineage pair format, which atomic requirement is the dependent, as well as the confidence level, lineage description, and lineage evidence between the two. This process is repeated until all atomic requirements have been analyzed pairwise. The final data is stored according to a graph data structure to form a lineage relationship diagram.

[0036] 3. Requirement Normalization Phase. The main task of this phase is to merge the content of dependent target requirements into dependent source nodes in reverse order of hierarchical dependencies, and perform text deduplication, sentence fluency, and logical integrity consistency operations on the merged source node content. The merging process is based on the semantic understanding capabilities of the large model. First, from the graph data structure of the hierarchical relationships, the upstream and downstream nodes of each atomic requirement node are identified, and the text content of each node is pushed onto the corresponding stack. Then, the leaf nodes and their parent nodes are found, and together with the prompt words, the large model is called to merge the text content of the leaf nodes into the parent nodes. The result is stored in the stack corresponding to the parent node itself so that the latest text content of the current node can be retrieved in multiple mergings. This process is repeated until all leaf nodes have been merged upwards to the root node, thus achieving the normalization process.

[0037] In summary, through the above three steps, the lineage analysis between the original requirement documents is achieved, and the requirement documents are normalized. The flowchart is as follows: Figure 1 As shown, this implements lineage analysis between original requirement documents and normalizes the requirement documents.

[0038] In summary, by utilizing the above-mentioned technical solution of this invention, the invention extracts atomic requirements from the original requirement document, constructs inheritance and dependency relationships between atomic requirements through the Chinese semantic understanding capabilities of the large model, thereby converging to form a lineage network of all atomic requirements in the requirement document. Then, according to the closeness of the lineage relationship, atomic requirements with lineage relationships are merged from farthest to closest to form atomic requirement packages. Finally, the atomic requirement packages are integrated and unified into the target container document, realizing the normalization processing of the requirement document, and ultimately forming a "single authoritative requirement baseline". The requirement content related to the current requirement document is summarized level by level into the current document, so that when reading the current document, there is no need to flip back and forth between other documents, reducing manual workload and improving work efficiency.

[0039] 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, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for requirement normalization among requirement documents, characterized in that, Includes the following steps: S1 uses engineering methods to convert the original docx format requirement documents uploaded by users into html format documents while maintaining the consistency of the chapter structure. Based on the semantic understanding capabilities of the large model, the chapters of the html format document are merged into chapters suitable for atomic requirements. The text content of the atomic requirements and the merged and adjusted chapter titles are parsed out and stored in JSON format. At the same time, the unstructured requirement document data is converted into structured JSON format data. S2 summarizes the atomic requirements in the requirements document, selects any two pairs of atomic requirements and prompt words together to call the large model to analyze the lineage between atomic requirements. The large model outputs the lineage relationship between the two in the format of source-target lineage pair, as well as the confidence, lineage description and lineage evidence of the lineage relationship between the two. This process is repeated until all atomic requirements have been analyzed pairwise. The final data is stored in the form of graph data structure to form a lineage relationship map. S3 merges related atomic requirements into dependent source nodes in reverse order of hierarchical dependency, based on the semantic understanding capability of the large model. Then, it performs text deduplication, sentence fluency, and logical integrity consistency operations on the merged source node content to achieve normalization.

2. The requirement normalization method among requirement documents according to claim 1, characterized in that, The specific processing procedure for the large model in step S2 is as follows: S201 analyzes the lineage relationship between two atomic requirements based on the role settings, skill requirements, task objectives, task steps, input / output data formats, and constraints given in the prompts. S202 outputs the atomic requirements and dependent atomic requirements in the source-target lineage pair format, along with the confidence level, lineage description, and lineage evidence regarding the lineage relationship between them. S203 caches the current lineage pair data and continues to analyze the lineage relationship of the next pair of atomic requirements until any two atomic requirements have completed lineage analysis; S204 iterates through all lineage pairs, finds the upstream and downstream atomic requirements of each atomic requirement, and saves them as a graph data structure of nodes and edges to depict the lineage relationship between atomic requirements.

3. The requirement normalization method among requirement documents according to claim 1, characterized in that, The lineage relationship includes dependency and inheritance; the dependency relationship is that one atomic requirement refers to another atomic requirement, or the completion of one atomic requirement is a prerequisite for the start of another atomic requirement, or one atomic requirement and another atomic requirement share the same business process or business resource; the inheritance relationship refers to the update and optimization of document version and related atomic requirement content, including the update of semantic version number, requirement refinement and content evolution caused by changes in constraints.

4. The requirement normalization method among requirement documents according to claim 2, characterized in that, The specific steps of step S3 are as follows: S301 uses the user-uploaded requirement document as the target container, traverses all its atomic requirement lineage relationship graph data, finds the atomic requirements of all its leaf nodes, and creates a stack for each node, pushing the original text of each node onto the stack. S302 finds the latest atomic requirement text content in the stack corresponding to the current node and its parent node, calls the large model to merge it together with the prompt word, and pushes the merged atomic requirement content of the current node onto the stack. S303 Determine if the current node has a parent node. If yes, return to step S301; otherwise, proceed to step S304. S304 Determine if there are any leaf nodes that have not been merged. If so, proceed to step S204. Otherwise, complete the merging and unification process from the leaf nodes to the root node.

5. The requirement normalization method among requirement documents according to claim 4, characterized in that, The specific fusion process of the large model in step S302 is as follows: according to the role setting, skill requirements, task objectives, task steps, input and output data formats, and restrictions given by the prompt words, merge the atomic requirements of the leaf nodes into the atomic requirements of their parent nodes, and output the atomic requirements of the merged parent nodes.

6. The requirement normalization method among requirement documents according to claim 1, characterized in that, The original requirements document in step S1 includes the requirements and the atomic requirements contained in the sub-requirements. The atomic requirements include business descriptions, inputs and outputs, and business processes.

7. The requirement normalization method among requirement documents according to claim 1, characterized in that, The large model training process is as follows: during the training process, publicly available online texts, books, literary works, and discussions on social media are used as Chinese text training data to enable the model to learn Chinese grammatical structures, vocabulary collocations, and semantic features. Based on the Transformer architecture, it captures local and global dependencies in text and understands the semantics in the context; it fine-tunes the model through supervised learning to enable the model to accurately understand the user's intent. By incorporating human feedback, we learn to generate responses that align with human language habits and values.

8. A requirement normalization system among requirement documents, characterized in that, include: The atomic requirement extraction module is used to convert the original docx format requirement documents uploaded by users into html format documents using engineering methods, while maintaining the consistency of the chapter structure. Based on the semantic understanding capabilities of the large model, the chapters of the html format document are merged into chapters suitable for atomic requirements, the atomic requirement text content and the merged and adjusted chapter titles are parsed out and stored in JSON format, while converting unstructured requirement document data into structured JSON format data. The atomic requirement lineage analysis module is used to summarize the atomic requirements in the requirement document. It selects any two pairs of atomic requirements and prompt words to call the large model to analyze the lineage between atomic requirements. The large model outputs the lineage relationship between the two in the format of source-target lineage pair, as well as the confidence, lineage description and lineage evidence of the lineage relationship between the two. This process is repeated until all atomic requirements have been analyzed pairwise. The final data is stored in the form of graph data structure to form a lineage relationship map. The requirement normalization module is used to merge the content of dependent target requirements into dependent source nodes in reverse order of hierarchical dependencies, based on the semantic understanding capability of the large model. The module then performs text deduplication, sentence fluency, and logical integrity operations on the merged source node content to achieve normalization.

9. The requirement normalization method among requirement documents according to claim 8, characterized in that, The large model training process is as follows: during the training process, publicly available online texts, books, literary works, and discussions on social media are used as Chinese text training data to enable the model to learn Chinese grammatical structures, vocabulary collocations, and semantic features. Based on the Transformer architecture, it captures local and global dependencies in text and understands the semantics in the context; it fine-tunes the model through supervised learning to enable the model to accurately understand the user's intent. By incorporating human feedback, we learn to generate responses that align with human language habits and values.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program, when run by a processor, performs the steps of the demand normalization method according to any one of claims 1 to 7.

Images