A method for estimating the original function points of software based on multi-agent cooperation

By employing a multi-agent collaborative approach, the problem of function point estimation in the pre-requirement stage of large and complex software was solved, achieving fully automatic, stable, and complete quantitative measurement. This approach overcomes document context overruns and constraint omissions, improving the accuracy and efficiency of estimation.

CN123114539APending Publication Date: 2026-11-17BEIJING HUARU TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611213564.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-11
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies cannot perform fully automatic, stable, and complete quantitative measurement of the original functional points of large and complex software before the software code is written. Furthermore, they suffer from problems such as excessively long document context, missing constraints, and untraceable intermediate results.

Method used

A multi-agent collaborative approach is adopted, which coordinates the work of the main agent and the special task agent through global orchestration and scheduling. The system is divided, technical requirements are separated, functions are decomposed and use cases are modeled. The original software technical documents are processed step by step to realize function point calculation. The data processing stability and traceability are ensured by isolating the independent large language model session context window from the local cache.

Benefits of technology

It achieves end-to-end fully automated estimation from original software technical documents to original software function points, improving the completeness, consistency and processing efficiency of the estimation, avoiding text truncation and constraint omissions, and ensuring the traceability of intermediate results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123114539A_ABST
    Figure CN123114539A_ABST
Patent Text Reader

Abstract

This invention discloses a method for estimating the original function points of software based on multi-agent collaboration, including: acquiring original software technical document information; parsing the original software technical document information to obtain a functional use case design document information set; and performing function point calculation processing on the functional use case design document information set to obtain the original function point calculation result information of the software.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of software engineering software scale measurement and large language model application technology, and in particular to a method for estimating the original function points of software based on multi-agent collaboration. Background Technology

[0002] In the pre-requirement stages of custom software projects, such as project initiation, feasibility studies, and bidding, it is necessary to quantify the software's functional scope even before the software code is written, using only the original software technical requirements document. This is crucial for workload assessment, cost budgeting, and schedule estimation. For large and complex software projects in specific professional fields containing multiple subsystems, subsystems, and functional modules, the original technical requirements document is typically an unstructured, extremely long document such as a tender document, technical agreement, or overall plan, ranging from tens of thousands to hundreds of thousands of words. Performance and security constraints are often marked with specific tags within the document. Therefore, how to automatically, stably, and completely identify and quantify the original functional points of the software, starting from the original software technical requirements document, has become a pressing technical problem in the field of software scope measurement.

[0003] Existing software scale measurement technologies mainly fall into the following categories: The first category is line-of-code measurement technology, which can only be carried out after the software code is written. It is a post-measurement method and cannot output scale data in the pre-requirement stage. Furthermore, the measurement process is affected by the programming language and has weak cross-language comparability. The second category is constructive cost model measurement technology, which requires manually pre-calculated line-of-code counts or function scale data as input. It cannot directly parse the scale data from the original technical requirements document and lacks data processing capabilities that allow direct scale measurement from the requirements document. The third category is manual function point analysis technology, which, although it can be carried out in the requirements stage, requires a large amount of manpower and is time-consuming, and is limited by system boundaries and functions. The division of functional units lacks unified quantitative criteria, leading to significant deviations in processing results for the same document by different operators. The fourth category involves single-model-assisted measurement techniques, which load extremely long documents into a single large language model session window for processing. This easily triggers context window capacity exceeding limits, resulting in data processing failures such as text truncation, module loss, constraint omissions, and inconsistencies. Furthermore, the output fluctuates significantly when processing the same document multiple times. The fifth category is general-purpose large-model multi-agent orchestration techniques. While these techniques break down tasks into sub-tasks and distribute them to multiple agents, they lack fixed procedures and constraint criteria for measuring software function points, and lack hierarchical persistence and reverse traceability of intermediate results. In summary, existing technologies lack a measurement method that can automatically, stably, and completely estimate the original software function points from the original software technical requirement document end-to-end, and possesses features such as context isolation processing for extremely long documents, mandatory constraint full-coverage verification, and reverse traceability of intermediate results. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a software original function point estimation method based on multi-agent collaboration, which can automatically, stably and completely estimate the software original function points from the original software technical requirements document end to end, avoid the context limit of excessively long documents, realize full coverage verification of mandatory constraints and reverse traceability of intermediate results, and improve the completeness, consistency and processing efficiency of software original function point estimation.

[0005] To address the aforementioned technical problems, a first aspect of this invention discloses a method for estimating the original function points of software based on multi-agent collaboration, the method comprising: S1, Obtain original software technical documentation information; S2, the original software technical document information is parsed and processed to obtain a functional use case design document information set; S3, perform function point calculation processing on the functional use case design document information set to obtain the original function point calculation result information of the software.

[0006] As an optional implementation, in the first aspect of the present invention, the step of parsing the original software technical document information to obtain a functional use case design document information set includes: S21, Perform subsystem division processing on the original software technical document information to obtain subsystem division list information; S22, Perform technical requirement separation processing on the original software technical document information and the subsystem partitioning list information to obtain a subsystem technical requirement document information set; S23, Perform functional decomposition processing on the subsystem technical requirements document information set to obtain a functional decomposition document information set; S24, Perform use case modeling processing on the functional decomposition document information set to obtain the functional use case design document information set.

[0007] As an optional implementation, in the first aspect of the present invention, the step of performing subsystem partitioning processing on the original software technical document information to obtain subsystem partitioning list information includes: S211, Perform module identification processing on the original software technical document information to obtain a candidate module information set; S212, default s=1; S213, perform feature extraction processing on the s-th candidate module information in the candidate module information set to obtain the s-th module cohesion value and the s-th cross-module coupling value; S214, using the subsystem partitioning confidence calculation model, calculate and process the cohesion value of the s-th module and the cross-module coupling value to obtain the s-th subsystem partitioning confidence value; The confidence calculation model for the subsystem partitioning is as follows: ; ; In the formula, Let be the cohesion value of the s-th module. , For the s-th cross-module coupling value, , The cohesive saturation coefficient. For the s-th cohesive saturation, , This is the coupling sharpening coefficient. Assign confidence values ​​to the s-th subsystem. ; S215, determine whether the confidence value of the s-th subsystem division is not less than the preset confidence threshold, and obtain the first judgment result; When the first judgment result is yes, the s-th candidate module information is determined to be the s-th subsystem identification information; When the first judgment result is negative, the s-th candidate module information is subjected to boundary adjustment processing to obtain a boundary-adjusted candidate module information set, and the boundary-adjusted candidate module information set is determined as the candidate module information set. S212 is executed. S216, determine whether s is equal to the number of candidate module information in the candidate module information set, and obtain a second judgment result; When the second judgment result is negative, increment s by 1 and execute S213; When the second judgment result is yes, all the subsystem identification information is collected and processed to obtain the subsystem division list information.

[0008] As an optional implementation, in the first aspect of the present invention, the step of performing feature extraction processing on the s-th candidate module information in the candidate module information set to obtain the s-th module cohesion value and the s-th cross-module coupling value includes: S2131, Perform correlation degree statistical processing on the s-th candidate module information to obtain the correlation degree value of the s-th internal item and the correlation degree value of the s-th cross-module item; S2132, normalize the correlation value of the s-th internal entry and the correlation value of the s-th cross-module entry to obtain the cohesion value of the s-th module and the coupling value of the s-th cross-module.

[0009] As an optional implementation, in the first aspect of the present invention, the step of separating the original software technical document information and the subsystem partitioning list information into a subsystem technical requirement document information set includes: S221, traverse each subsystem identifier information in the subsystem partitioning list information, extract the exclusive technical clause information corresponding to each subsystem identifier information from the original software technical document information, and obtain the subsystem exclusive technical clause information set; S222, the subsystem-specific technical clause information set is marked and retained to obtain the subsystem technical requirement document information set.

[0010] As an optional implementation, in the first aspect of the present invention, the step of marking and retaining the subsystem-specific technical clause information set to obtain the subsystem technical requirements document information set includes: S2221, Traverse each subsystem-specific technical clause information in the subsystem-specific technical clause information set, and perform mark retention processing on the performance mandatory constraint mark in each subsystem-specific technical clause information to obtain the technical clause information set after performance mark retention; S2222, Perform mark retention processing on the security mandatory constraint marks in the technical clause information set after the performance mark retention is obtained to obtain the subsystem technical requirement document information set.

[0011] As an optional implementation, in the first aspect of the present invention, the step of performing functional decomposition processing on the subsystem technical requirements document information set to obtain a functional decomposed document information set includes: S231, preset k=1; S232, perform structured decomposition processing on the kth subsystem technical requirement document information in the subsystem technical requirement document information set to obtain the kth initial functional decomposition document information; S233, perform constraint binding processing on the kth initial functional decomposition document information to obtain the kth bound functional decomposition document information, the kth performance constraint coverage value, and the kth security constraint coverage value; S234, calculate and process the kth performance constraint coverage value and the kth security constraint coverage value to obtain the kth constraint binding completeness value; S235, determine whether the binding completeness value of the kth constraint is not less than the preset binding completeness threshold, and obtain the third judgment result; When the third judgment result is yes, the kth bound function decomposition document information is determined to be the kth function decomposition document information; When the third judgment result is negative, the initial functional decomposition document information of the kth initial functional decomposition document information is recursively re-decomposed to obtain the re-decomposed initial functional decomposition document information. The re-decomposed initial functional decomposition document information is determined to be the initial functional decomposition document information of the kth initial functional decomposition document information, and S233 is executed. S236, determine whether k is equal to the number of subsystem technical requirement document information in the subsystem technical requirement document information set, and obtain the fourth judgment result; When the fourth judgment result is negative, increment k by 1 and execute S232; When the fourth judgment result is yes, all functional decomposition document information is collected and processed to obtain a functional decomposition document information set.

[0012] As an optional implementation, in the first aspect of the present invention, the step of performing use case modeling processing on the functional decomposition document information set to obtain a functional use case design document information set includes: S241, perform element completion processing on each minimum functional item information in the functional decomposition document information set to obtain a functional use case information set after element completion. S242, After the elements are completed, the functional use case information set is standardized to obtain a standardized functional use case information set; S243, Perform constraint embedding comparison processing on the standardized functional use case information set to obtain the functional use case design document information set.

[0013] As an optional implementation, in the first aspect of the present invention, the step of performing function point calculation processing on the functional use case design document information set to obtain the original function point calculation result information of the software includes: S31, preset t=1; S32, perform functional unit identification processing on the t-th functional use case design document information in the functional use case design document information set to obtain the t-th functional unit identification result information; S33, Perform function point matching processing on the identification result information of the tth functional unit to obtain the tth function point splitting table information; S34, determine whether t is equal to the number of functional use case design document information in the functional use case design document information set, and obtain the fifth judgment result; When the fifth judgment result is negative, increment t by 1 and execute S32; When the fifth judgment result is yes, all the function point split table information is summarized and processed to obtain the initial function point calculation result information; S35, perform consistency evaluation processing on the initial function point calculation result information to obtain the original function point calculation result information of the software.

[0014] As an optional implementation, in the first aspect of the present invention, the step of performing functional unit identification processing on the t-th functional use case design document information in the functional use case design document information set to obtain the t-th functional unit identification result information includes: S321, Perform data function identification processing on the t-th functional use case design document information to obtain data function unit information; S322, Perform transaction function identification processing on the t-th functional use case design document information to obtain transaction function unit information; S323, integrate the data function unit information and the transaction function unit information to obtain the identification result information of the t-th function unit.

[0015] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: In this embodiment of the invention, original software technical document information is obtained; the original software technical document information is parsed to obtain a functional use case design document information set; the functional use case design document information set is processed for function point calculation to obtain the original function point calculation result information of the software. It can be seen that this application avoids the problems of context window capacity exceeding limits, text truncation, and constraint omissions triggered by loading excessively long documents into a single large model session from the bottom layer by configuring physically isolated independent large language model session context windows and local caches for the global orchestration and scheduling master agent and each special task agent, and by splitting excessively long documents into lightweight sub-documents according to subsystems for step-by-step processing. Furthermore, by persistently storing the intermediate results of each process in five independent directories—system partitioning list, subsystem technical requirements, functional decomposition, functional use case design, and function point calculation—and by ensuring that agents only transfer files between them, this application addresses these issues. The path and quantity summary constructs an evidence chain that can be traced back layer by layer from the function point calculation results to the source of constraints, overcoming the deficiency of general orchestration technology in lacking business evidence that can be traced back. By performing a full-coverage comparison of performance-forced constraint markers and security-forced constraint markers, and triggering the recursive regeneration of the corresponding intelligent agent when the completeness, quantity and consistency of constraint binding are not up to standard, the defects of function omission and constraint omission that are common in manual and single large model processing are overcome. Thus, the completeness, consistency and processing efficiency of the software's original function point estimation are improved without relying on manual preprocessing. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0017] Figure 1 This is a flowchart illustrating a software original function point estimation method based on multi-agent collaboration disclosed in an embodiment of the present invention. Figure 2 This is a schematic diagram of a software raw function point estimation device based on multi-agent collaboration disclosed in an embodiment of the present invention; Figure 3 This is a schematic diagram of another software raw function point estimation device based on multi-agent collaboration disclosed in an embodiment of the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0020] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0022] It should be noted that since the method in this application embodiment is executed in a computer device, the processing objects of each computer device exist in the form of data or information, such as time, which is essentially time information. It is understood that if size, quantity, position, etc. are mentioned in subsequent embodiments, they are all corresponding data that exist so that the computer device can process them. Specific details will not be elaborated here.

[0023] It should be noted that the artificial intelligence-related technologies that may be involved in this application will be briefly described. Artificial intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. Artificial intelligence is the study of the design principles and implementation methods of various intelligent machines, enabling machines to have the functions of perception, reasoning, and decision-making.

[0024] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0025] This application provides a method for estimating the original function points of software based on multi-agent collaboration, which will be described in detail below.

[0026] Example 1 Please see Figure 1 , Figure 1This is a flowchart illustrating a method for estimating the original function points of software based on multi-agent collaboration, as disclosed in an embodiment of the present invention. Wherein, Figure 1 The described method for estimating the original function points (RFP) of software based on multi-agent collaboration is applied to software RFP estimation systems, such as local servers or cloud servers used for software size measurement. This invention does not limit the application of this method. Figure 1 As shown, the method includes the following steps: S1, Obtain original software technical documentation information.

[0027] It should be noted that the software's original function point estimation system includes a global orchestration and scheduling main agent and several specialized task agents. These specialized task agents are, in order, a subsystem identification agent, a technical requirement separation agent, a function decomposition agent, a use case modeling agent, and a function point splitting agent. Each agent is a software entity running on a computer system, consisting of a large language model interface, preset task configurations, an independent large language model session context window, local caching, and processing modules. It generates output information by combining the preset task configurations with input information and then calling the large language model interface. The global orchestration and scheduling main agent and the specialized task agents are similar software entities with different preset task configurations and performing different data processing tasks. Specifically, the subsystem identification agent, the technical requirement separation agent, the function decomposition agent, the use case modeling agent, and the function point splitting agent each call the large language model interface through their respective preset task configurations to sequentially execute the data processing tasks of subsystem partitioning, technical requirement separation, function decomposition, use case modeling, and function point splitting. Each of the aforementioned task-specific intelligent agents is configured with an independent large language model session context window and local cache. The session data of each task-specific intelligent agent is physically isolated from and does not share data with the global orchestration and scheduling master intelligent agent and other task-specific intelligent agents. The global orchestration and scheduling master intelligent agent is a software entity that receives user-inputted estimation task instructions and schedules the execution of each task-specific intelligent agent sequentially according to a preset unidirectional serial dependency logic.

[0028] It should be noted that the large language model called by the above-mentioned large language model interface can be a Qwen series, DeepSeek series, or Kimi series large language model. Specifically, the embodiments of the present invention do not limit it. The preset task configuration includes task prompt templates and input / output format constraints, which are used to constrain the processing tasks and output structures when the corresponding agent calls the large language model interface.

[0029] It should be noted that the global orchestration and scheduling master agent and the various task agents complete the estimation of the original function points of the software through a three-layer collaboration: First, orchestration and scheduling collaboration, where the global orchestration and scheduling master agent triggers the subsystem identification agent, the technical requirement separation agent, the function decomposition agent, the use case modeling agent, and the function point splitting agent to execute sequentially according to a preset unidirectional serial dependency logic. After each task agent completes its execution, it sends a file path and quantity digest back to the global orchestration and scheduling master agent. The global orchestration and scheduling master agent then compares the results before triggering the next task agent. The system employs three main mechanisms: First, it integrates multiple intelligent agents. The first is file-driven data collaboration. Instead of directly transmitting complete document text, each agent persists its output to its corresponding directory. The second agent reads the previous process's output from the corresponding directory according to the file path, allowing each agent to process lightweight sub-documents of a single subsystem within its own independent large language model session context window. The third is verification feedback collaboration. When the global orchestration and scheduling master agent fails the consistency evaluation, it calls back the corresponding agent to regenerate and overwrite any missing or mismatched documents in the corresponding directory. This collaboration enables the subsystem division, technical requirement separation, functional decomposition, use case modeling, and functional point splitting of ultra-long documents to be handled by multiple agents with isolated contexts, sequentially connected via file-driven mechanisms, and closed-loop callbacks when verification fails. This avoids the context window capacity exceeding limits and constraint omissions caused by loading ultra-long documents into a single large language model session.

[0030] It should be noted that the original software technical document information refers to the original technical requirements document of the software to be estimated, including unstructured documents such as tender documents, technical agreements, or overall plans, using a marked text format with unified character encoding. The original software technical document information includes performance-mandated constraint markers and security-mandated constraint markers. The performance-mandated constraint markers are text markers in the original software technical document information used to identify performance-mandated indicator clauses, and the security-mandated constraint markers are text markers in the original software technical document information used to identify security-mandated indicator clauses. The original software technical document information is read after the global orchestration and scheduling main intelligent agent receives the estimation task instruction input by the user, and is used as input for subsequent processing after comparing document readability and completeness. The above-mentioned original software technical document information can be input via keyboard, mouse, or file upload interface; specifically, this embodiment of the invention does not limit the input.

[0031] S2, the original software technical document information is parsed and processed to obtain a functional use case design document information set.

[0032] It should be noted that the parsing process refers to the document structuring process executed sequentially by the subsystem identification agent, the technical requirement separation agent, the function decomposition agent, and the use case modeling agent. The parsing process includes subsystem partitioning, technical requirement separation, function decomposition, and use case modeling. During the parsing process, subsystem partitioning yields a subsystem partitioning list and stores it in the system partitioning list directory; technical requirement separation yields a subsystem technical requirement document information set and stores it in the subsystem technical requirement directory; function decomposition yields a function decomposition document information set and stores it in the function decomposition directory; and use case modeling yields a functional use case design document information set and stores it in the functional use case design directory. Only file paths and quantity summaries are transmitted between the agents; the complete document text content is not transmitted. The system partitioning list directory refers to a logical storage partition used to store the subsystem partitioning list information; the subsystem technical requirements directory refers to a logical storage partition indexed by subsystem identification information and used to store the corresponding subsystem technical requirements document information; the function decomposition directory refers to a logical storage partition indexed by subsystem identification information and used to store the corresponding function decomposition document information; the function use case design directory refers to a logical storage partition indexed by subsystem identification information and minimum functional item information and used to store the corresponding function use case design document information. The function use case design document information set includes several function use case design document information, each of which contains function function information, execution precondition information, input data set information, step-by-step operation process information, normal output result information, exception handling logic information, and multi-branch business path information.

[0033] S3, perform function point calculation processing on the functional use case design document information set to obtain the original function point calculation result information of the software.

[0034] It should be noted that the function point calculation process is jointly executed by the function point splitting agent and the global orchestration and scheduling main agent. The function point calculation process includes function unit identification, function point matching, result summarization, and consistency evaluation. Function unit identification identifies five types of function units: internal logic files, external interface files, external inputs, external outputs, and external queries. Function point matching queries a preset function point value mapping table to match corresponding function point values ​​for each function unit. Consistency evaluation performs cross-process consistency verification and recursive completion on documents from each process step to confirm the accuracy of the calculation results. The original function point calculation result information includes function point splitting table information for each subsystem and the total original function point value of the software. Each function point splitting table information includes functional requirements, sequence number, function unit type, number of function points, and calculation instructions. The original function point calculation result information is stored in a function point calculation directory. The function point calculation directory is a logical storage partition indexed by subsystem identification information and functional use case design document information, used to store function point splitting table information, initial function point calculation result information, and original function point calculation result information. The five directories are linked by a path according to the processing order of subsystem division, technical requirement separation, functional decomposition, use case modeling, and function point calculation, so that information in the later directory can be traced back to the source information in the previous directory through the corresponding index.

[0035] As can be seen, the software original function point estimation method based on multi-agent collaboration described in the embodiments of the present invention achieves end-to-end fully automatic estimation from original software technical document information to software original function point calculation result information through multi-agent context physical isolation and file-driven serial collaboration.

[0036] In an optional embodiment, the step of parsing the original software technical document information to obtain a functional use case design document information set includes: S21, Perform subsystem division processing on the original software technical document information to obtain subsystem division list information; It should be noted that the subsystem identification agent is used to identify several candidate module information from the original software technical document information, confirm the candidate boundary description information in each candidate module information, and adjust the boundaries of candidate module information whose confidence level does not reach a preset division confidence threshold, so as to generate subsystem identification information. The candidate module information refers to a structured data object formed by aggregating interrelated technical items in the original software technical document information, used to characterize the boundaries of the subsystem to be confirmed; each candidate module information includes a candidate module name, a set of technical item identification information, and the candidate boundary description information. The subsystem division process includes module identification, feature extraction, confidence calculation, and boundary confirmation. The subsystem division list information includes several subsystem identification information, and each subsystem identification information includes a subsystem name and a subsystem boundary description.

[0037] S22, Perform technical requirement separation processing on the original software technical document information and the subsystem partitioning list information to obtain a subsystem technical requirement document information set; It should be noted that the technical requirements separation agent is used to extract the functional clause information, performance clause information, interface clause information, and constraint clause information corresponding to each subsystem identifier from the original software technical document information, and retains the performance mandatory constraint marker and the security mandatory constraint marker during the separation process. The technical requirements separation process includes clause extraction, marker retention, and quantity matching. The subsystem technical requirements document information set includes several subsystem technical requirements document information, and each subsystem technical requirements document information corresponds one-to-one with a subsystem identifier in the subsystem partitioning list information.

[0038] S23, Perform functional decomposition processing on the subsystem technical requirements document information set to obtain a functional decomposition document information set; It should be noted that the functional decomposition agent is used to decompose the technical requirements document information of each subsystem into the minimum functional item information level by level, and bind the performance mandatory constraint marker and the security mandatory constraint marker to the corresponding minimum functional item information. The functional decomposition process includes decomposition, binding, and evaluation. The functional decomposition document information set includes several functional decomposition document information, each functional decomposition document information contains several minimum functional item information, and each minimum functional item information is bound to the corresponding performance mandatory constraint marker and the security mandatory constraint marker.

[0039] S24, Perform use case modeling processing on the functional decomposition document information set to obtain the functional use case design document information set.

[0040] It should be noted that the use case modeling agent is used to convert each of the minimum functional item information into corresponding functional use case design document information, and to complete the functional role information, execution precondition information, input data set information, step-by-step operation process information, normal output result information, exception handling logic information, and multi-branch business path information. The use case modeling process includes element completion, normalization, and embedding comparison. Each of the functional use case design document information in the functional use case design document information set corresponds one-to-one with one of the minimum functional item information in the functional decomposition document information set.

[0041] As can be seen, the software original function point estimation method based on multi-agent collaboration described in the embodiments of the present invention achieves structured parsing of ultra-long documents by subsystem lightweight splitting through step-by-step serial processing of subsystem division, technical requirement separation, functional decomposition and use case modeling.

[0042] In another optional embodiment, the step of performing subsystem segmentation processing on the original software technical document information to obtain subsystem segmentation list information includes: S211, Perform module identification processing on the original software technical document information to obtain a candidate module information set; It should be noted that the module identification processing refers to the subsystem identification agent aggregating interrelated technical items in the original software technical document information to form several candidate module information sets, which include several candidate module information sets. The above module identification processing can be performed using text clustering, topic models, or graph community discovery; specifically, this embodiment of the invention does not limit the specific methods used.

[0043] S212, default s=1; It should be noted that s is a counter variable for traversing the candidate module information set, with an initial value of 1.

[0044] S213, perform feature extraction processing on the s-th candidate module information in the candidate module information set to obtain the s-th module cohesion value and the s-th cross-module coupling value; It should be noted that the feature extraction process refers to calculating the correlation between the technical item identifiers within the s-th candidate module information as the internal item correlation, and calculating the correlation between its technical item identifiers and the technical item identifiers in other candidate module information as the cross-module item correlation, and then normalizing them separately. The s-th module cohesion value is used to characterize the degree of correlation concentration among the technical item identifiers within the s-th candidate module information. The larger the s-th module cohesion value, the tighter the correlation between the technical items within the s-th candidate module information. The s-th cross-module coupling value is used to characterize the degree of cross-correlation between the technical item identifiers in the s-th candidate module information and the technical item identifiers in other candidate module information. The larger the s-th cross-module coupling value, the higher the overlap of candidate boundaries between the s-th candidate module information and other candidate module information. The above feature extraction process can be performed using text clustering, keyword co-occurrence statistics, or graph connectivity statistics. Specifically, this embodiment of the invention does not limit the specific methods used.

[0045] S214, using the subsystem partitioning confidence calculation model, calculate and process the cohesion value of the s-th module and the cross-module coupling value to obtain the s-th subsystem partitioning confidence value; The confidence calculation model for the subsystem partitioning is as follows: ; ; In the formula, Let be the cohesion value of the s-th module. , For the s-th cross-module coupling value, , The cohesive saturation coefficient. For the s-th cohesive saturation, , This is the coupling sharpening coefficient. Assign confidence values ​​to the s-th subsystem. .

[0046] It should be noted that the cohesive saturation coefficient has a value range of [3,8], with a preferred value of 5; the coupling sharpening coefficient has a value range of [1,4], with a preferred value of 2.

[0047] It should be noted that the confidence value of the s-th subsystem partitioning is based on the s-th cohesion saturation and uses the coupling sharpening index as the exponent: as the cohesion value of the s-th module increases, the s-th cohesion saturation approaches 1 through the saturation curve; as the coupling value of the s-th cross-module increases, the exponent increases, which exponentially suppresses the low and medium cohesion saturation; when the coupling value of the s-th cross-module approaches zero, it degenerates into the s-th cohesion saturation itself. The confidence value of the s-th subsystem partitioning serves as the boundary confirmation criterion for the subsystem partitioning process, used to determine the generation of the subsystem partitioning list information and candidate modules. Boundary adjustment enables this solution to stably confirm subsystem partitioning in scenarios with high cohesion and low coupling in ultra-long documents and suppress erroneous partitioning when content is ambiguous. This serves the stable splitting goal of using subsystems as physical isolation units for subsequent technical requirement separation and functional decomposition. Compared with the shortcomings of existing technologies that use fixed threshold linear determination of cohesion and coupling, resulting in subsystem boundary jitter and the number of subsystems drifting in multiple processing, the saturated nested power modulation achieves a balance between penalizing high coupling and avoiding excessive collapse of low coupling areas, making subsystem partitioning more stable and consistent.

[0048] S215, determine whether the confidence value of the s-th subsystem division is not less than the preset confidence threshold, and obtain the first judgment result; When the first judgment result is yes, the s-th candidate module information is determined to be the s-th subsystem identification information; When the first judgment result is negative, the s-th candidate module information is subjected to boundary adjustment processing to obtain a boundary-adjusted candidate module information set, and the boundary-adjusted candidate module information set is determined as the candidate module information set. S212 is executed. It should be noted that the boundary adjustment process refers to the subsystem identification agent re-aggregating or splitting the technical item identification information of the s-th candidate module information and synchronously updating the candidate boundary description information of related candidate module information, so as to improve the degree of correlation and concentration of technical items within the candidate module information and reduce the degree of cross-correlation between different candidate module information; when executing S212, s is restored to 1 and the candidate module information set is re-traversed. The above boundary adjustment process can be carried out by re-dividing text clustering, merging and splitting hierarchical clustering, or re-dividing graph communities. Specifically, the embodiments of the present invention are not limited. The division confidence threshold ranges from [0.6, 0.9], with a preferred value of 0.75.

[0049] S216, determine whether s is equal to the number of candidate module information in the candidate module information set, and obtain a second judgment result; When the second judgment result is negative, increment s by 1 and execute S213; When the second judgment result is yes, all the subsystem identification information is collected and processed to obtain the subsystem division list information.

[0050] As can be seen, the software original function point estimation method based on multi-agent collaboration described in the embodiments of the present invention achieves stable division of subsystem boundaries and boundary adjustment of low-confidence candidate module information by extracting cohesion coupling features of candidate module information one by one, calculating confidence and comparing boundaries.

[0051] In another optional embodiment, the step of performing feature extraction processing on the s-th candidate module information in the candidate module information set to obtain the s-th module cohesion value and the s-th cross-module coupling value includes: S2131, Perform correlation degree statistical processing on the s-th candidate module information to obtain the correlation degree value of the s-th internal item and the correlation degree value of the s-th cross-module item; It should be noted that the correlation statistics processing refers to calculating the correlation strength between the internal technical item identifiers of the s-th candidate module information as the correlation value of the s-th internal item, and calculating the correlation strength between its technical item identifiers and the technical item identifiers of other candidate module information as the correlation value of the s-th cross-module item. The above correlation statistics processing can be performed using keyword co-occurrence statistics or semantic similarity statistics. Specifically, the embodiments of the present invention do not limit this.

[0052] S2132, normalize the correlation value of the s-th internal entry and the correlation value of the s-th cross-module entry to obtain the cohesion value of the s-th module and the coupling value of the s-th cross-module.

[0053] It should be noted that the above normalization process can be performed using max-min normalization or saturated truncation normalization. Specifically, the embodiments of the present invention do not limit the specifics.

[0054] As can be seen, the software original function point estimation method based on multi-agent collaboration described in the embodiments of the present invention provides input for the calculation of subsystem partition confidence through correlation statistics and normalization.

[0055] In another optional embodiment, the step of separating the original software technical documentation information and the subsystem partitioning list information into a subsystem technical requirement document information set includes: S221, traverse each subsystem identifier information in the subsystem partitioning list information, extract the exclusive technical clause information corresponding to each subsystem identifier information from the original software technical document information, and obtain the subsystem exclusive technical clause information set; It should be noted that the proprietary technical terms information includes functional terms information, performance terms information, interface terms information, and constraint terms information. The subsystem proprietary technical terms information set includes several subsystem proprietary technical terms, and each subsystem proprietary technical terms corresponds one-to-one with a subsystem identifier. The above extraction process can be performed using text matching or semantic retrieval; specifically, this embodiment of the invention does not limit the specific methods used.

[0056] S222, the subsystem-specific technical clause information set is marked and retained to obtain the subsystem technical requirement document information set; It should be noted that the mark retention process refers to, when generating the technical requirements document information for each subsystem, sequentially performing mark character positioning, mark binding with the clause, mark writing as is, and mark quantity comparison on the corresponding subsystem-specific technical clause information, so that all performance mandatory constraint marks and security mandatory constraint marks are written together with the corresponding clauses into the subsystem technical requirements document information, avoiding the loss of mandatory constraint marks during the separation process.

[0057] As can be seen, the software original function point estimation method based on multi-agent collaboration described in the embodiments of the present invention achieves complete separation of the technical requirements of each subsystem through the extraction of exclusive technical clause information, retention of mandatory constraint markers and recursive completion of quantity matching.

[0058] In another optional embodiment, the step of marking and retaining the subsystem-specific technical clause information set to obtain the subsystem technical requirements document information set includes: S2221, Traverse each subsystem-specific technical clause information in the subsystem-specific technical clause information set, and perform mark retention processing on the performance mandatory constraint mark in each subsystem-specific technical clause information to obtain the technical clause information set after performance mark retention; It should be noted that the mark retention processing of the performance mandatory constraint mark refers to scanning according to the character order of the technical clause information specific to each subsystem, locating the performance mandatory constraint mark through precise matching of preset character encoding or regular expression matching, establishing a binding relationship between each performance mandatory constraint mark and the performance clause information it belongs to, and writing the original characters of the performance mandatory constraint mark, the corresponding performance clause information, and their arrangement order in the original text into the technical clause information set after performance mark retention. The above mark retention processing can be performed using Unicode character encoding matching, regular expression matching, or finite state automata; specifically, this embodiment of the invention does not limit the specific methods used.

[0059] S2222, Perform mark retention processing on the security mandatory constraint marks in the technical clause information set after the performance mark retention is obtained to obtain the subsystem technical requirement document information set.

[0060] It should be noted that the mark retention processing of the security mandatory constraint marks refers to scanning according to the character order of the technical clause information set after the performance marks are retained, locating the security mandatory constraint marks through precise matching of preset character encoding or regular expression matching, establishing a binding relationship between each security mandatory constraint mark and the constraint clause information it belongs to, and writing the original characters of the security mandatory constraint mark, the corresponding constraint clause information, and their arrangement order in the original text into the subsystem technical requirements document information set. The above mark retention processing can be performed using Unicode character encoding matching, regular expression matching, or finite state automata; specifically, this embodiment of the invention does not limit the specific methods. The subsystem technical requirements document information set completely retains all performance mandatory constraint marks and security mandatory constraint marks in the technical clause information specific to each subsystem.

[0061] As can be seen, the software original function point estimation method based on multi-agent collaboration described in the embodiments of the present invention achieves full coverage retention of mandatory constraint labels in the process of separating technical requirements by classifying and retaining performance mandatory constraint labels and security mandatory constraint labels.

[0062] In another optional embodiment, the step of performing functional decomposition processing on the subsystem technical requirements document information set to obtain a functional decomposed document information set includes: S231, preset k=1; It should be noted that k is a counter variable for traversing the technical requirements document information set of the subsystem, with an initial value of 1.

[0063] S232, perform structured decomposition processing on the kth subsystem technical requirement document information in the subsystem technical requirement document information set to obtain the kth initial functional decomposition document information; It should be noted that the structured decomposition process refers to the hierarchical breakdown of the functional clause information in the technical requirements document of the k-th subsystem, according to the subsystem, first-level module, second-level submodule, and minimum functional item levels, and the process of mining implicit functions that are related to the business and supplementing them as minimum functional item information. Implicit functions refer to supporting functions that are not explicitly described in the functional clause information but are necessary to complete the corresponding business, such as data storage, log retention, and exception rollback. The initial functional decomposition document information of the k-th subsystem contains several minimum functional item information. The above structured decomposition process can be performed using hierarchical hint parsing of large language models, syntactic dependency analysis, or rule template extraction; specifically, this embodiment of the invention does not limit the specific methods used.

[0064] S233, perform constraint binding processing on the kth initial functional decomposition document information to obtain the kth bound functional decomposition document information, the kth performance constraint coverage value, and the kth security constraint coverage value; It should be noted that the constraint binding process refers to taking each minimum functional item in the k-th initial functional decomposition document information, and writing the performance-mandated constraint markers and security-mandated constraint markers semantically pointing to that minimum functional item in the k-th subsystem technical requirement document information into the constraint field of that minimum functional item information, thus obtaining the k-th bound functional decomposition document information; the k-th performance constraint coverage value is the ratio of the number of bound performance-mandated constraint markers in the k-th bound functional decomposition document information to the total number of performance-mandated constraint markers in the k-th subsystem technical requirement document information, and the k-th security constraint coverage value is the ratio of the number of bound security-mandated constraint markers in the k-th bound functional decomposition document information to the total number of security-mandated constraint markers in the k-th subsystem technical requirement document information. The above constraint binding process can be performed using semantic similarity matching, keyword mapping, or large language model referencing resolution; specifically, this embodiment of the invention does not limit the specific methods used.

[0065] S234, calculate and process the kth performance constraint coverage value and the kth security constraint coverage value to obtain the kth constraint binding completeness value; It should be noted that the above calculation process can be performed using a forced constraint binding completeness evaluation calculation model, or using a weighted geometric mean algorithm or a harmonic mean algorithm. Specifically, the embodiments of the present invention do not limit the specifics.

[0066] The calculation model for evaluating the completeness of the mandatory constraint binding is as follows: ; ; ; In the formula, For the k-th performance constraint coverage value, , For the k-th security constraint coverage value, , As the baseline coefficient for performance loss, The exponent for performance loss. To cover the loss component for the k-th performance constraint, ; As a benchmark coefficient for safe losses, For safety saturation coefficient, To cover the loss component for the k-th safety constraint, ; For the k-th constraint binding completeness value, .

[0067] It should be noted that the performance loss benchmark coefficient ranges from [0.5, 1], with a preferred value of 1; the performance loss power exponent ranges from [0.7, 2], with a preferred value of 1.2; the safety loss benchmark coefficient ranges from [0.5, 1], with a preferred value of 1; and the safety saturation coefficient ranges from [3, 8], with a preferred value of 5.

[0068] It should be noted that the k-th constraint binding completeness value is first squared and summed for the performance constraint coverage loss component and the safety constraint coverage loss component, then the square root is taken and multiplied by the normalization factor. The result is obtained by squaring, which eliminates the mutual cancellation of the directional differences between the two types of losses, and the square root restores the comprehensive loss to the same order of magnitude as the single component. Therefore, any insufficient coverage of any type of mandatory constraint is fully included without being masked by the other type. The completeness value of the constraint binding mentioned in the k-th case serves as the completeness criterion for the functional decomposition stage, used to determine whether to trigger recursive re-decomposition of the k-th initial functional decomposition document information. This ensures that all performance mandatory constraints and safety mandatory constraints are bound to the minimum functional item information, with no loose constraints, serving the overall goal of full coverage of mandatory constraints and no omissions in the final functional point calculation. Compared with the existing technology that directly adds the two types of coverage rates, causing the directional differences to cancel each other out and resulting in missed coverage losses, takes the maximum value to ignore the smaller type of weak loss, or relies on the large model itself for decomposition without coverage criteria, the heterogeneous component Euclidean synthesis ensures that even the slight lack of any type of constraint is detected, and the constraint binding is more complete.

[0069] S235, determine whether the binding completeness value of the kth constraint is not less than the preset binding completeness threshold, and obtain the third judgment result; When the third judgment result is yes, the kth bound function decomposition document information is determined to be the kth function decomposition document information; When the third judgment result is negative, the initial functional decomposition document information of the kth initial functional decomposition document information is recursively re-decomposed to obtain the re-decomposed initial functional decomposition document information. The re-decomposed initial functional decomposition document information is determined to be the initial functional decomposition document information of the kth initial functional decomposition document information, and S233 is executed. It should be noted that the recursive re-decomposition process refers to re-executing the structured decomposition process guided by the unbound performance and security mandatory constraint tags in the technical requirements document information of the k-th subsystem. This process supplements and decomposes the functional clause information corresponding to the unbound mandatory constraint tags into new minimum functional item information, enabling the uncovered mandatory constraint tags to obtain bindable minimum functional item information. The binding completeness threshold ranges from [0.9, 0.99], with a preferred value of 0.95.

[0070] S236, determine whether k is equal to the number of subsystem technical requirement document information in the subsystem technical requirement document information set, and obtain the fourth judgment result; When the fourth judgment result is negative, increment k by 1 and execute S232; When the fourth judgment result is yes, all functional decomposition document information is collected and processed to obtain a functional decomposition document information set.

[0071] As can be seen, the software original function point estimation method based on multi-agent collaboration described in the embodiments of the present invention achieves full coverage binding of mandatory constraints and functional decomposition without free constraints by iteratively traversing, structurally decomposing, binding constraints, calculating completeness, and recursively re-decomposing the technical requirements document information of the subsystem.

[0072] In another optional embodiment, the step of performing use case modeling processing on the functional decomposition document information set to obtain a functional use case design document information set includes: S241, perform element completion processing on each minimum functional item information in the functional decomposition document information set to obtain a functional use case information set after element completion. It should be noted that the element completion processing refers to completing the functional role information, execution precondition information, input data set information, step-by-step operation process information, normal output result information, exception handling logic information, and multi-branch business path information for each minimum functional item. The above element completion processing can be performed using large language model slot filling, template matching, or sequence labeling algorithms; specifically, this embodiment of the invention does not limit the specific methods used.

[0073] S242, After the elements are completed, the functional use case information set is standardized to obtain a standardized functional use case information set; It should be noted that the standardization process refers to unifying the step-by-step operation process information in the functional use case information set after the elements are completed into verb-object phrase descriptions, and configuring result saving and data storage operations at the end of the business process. The above standardization process can be performed using dependency parsing, part-of-speech tagging, or preset text conversion templates. Specifically, this embodiment of the invention does not limit the specific methods used.

[0074] S243, Perform constraint embedding comparison processing on the standardized functional use case information set to obtain the functional use case design document information set.

[0075] It should be noted that the constraint embedding comparison processing refers to comparing whether the performance-mandated constraint markers and security-mandated constraint markers bound to each minimum functional item information have been embedded in the corresponding functional use case information content, and completing any unembedded constraints to ensure that all mandatory constraints are fully embedded in the functional use case design document information. The above constraint embedding comparison processing can use character exact matching, semantic similarity matching, or text implication recognition algorithms; specifically, this embodiment of the invention does not impose limitations.

[0076] As can be seen, the software original function point estimation method based on multi-agent collaboration described in the embodiments of the present invention achieves standardized functional use case modeling that conforms to engineering specifications through element completion, standardization and constraint embedding comparison.

[0077] In another optional embodiment, the step of performing function point calculation processing on the functional use case design document information set to obtain the original function point calculation result information of the software includes: S31, preset t=1; It should be noted that t is a counter variable for traversing the functional use case design document information set, with an initial value of 1.

[0078] S32, perform functional unit identification processing on the t-th functional use case design document information in the functional use case design document information set to obtain the t-th functional unit identification result information; It should be noted that the functional unit identification result information includes several functional unit information, and each functional unit information includes a functional unit type. The functional unit type is one of five types: internal logic file, external interface file, external input, external output, or external query.

[0079] S33, Perform function point matching processing on the identification result information of the tth functional unit to obtain the tth function point splitting table information; It should be noted that the function point matching process refers to querying a preset function point value mapping table and matching the corresponding function point value for each function unit information in the t-th function unit identification result information. For example, in the preset function point value mapping table, the function point value corresponding to the internal logic file is 7, the function point value corresponding to the external interface file is 5, the function point value corresponding to the external input is 4, the function point value corresponding to the external query is 4, and the function point value corresponding to the external output is 5. The above function point values ​​can be adjusted according to the software function point measurement standard. The t-th function point splitting table information includes functional requirements, serial number, function unit type, number of function points, and calculation instructions.

[0080] S34, determine whether t is equal to the number of functional use case design document information in the functional use case design document information set, and obtain the fifth judgment result; When the fifth judgment result is negative, increment t by 1 and execute S32; When the fifth judgment result is yes, all the function point split table information is summarized and processed to obtain the initial function point calculation result information; It should be noted that the summary processing refers to aggregating all function point breakdown table information by subsystem and summing all function point numbers to obtain the total original function point value of the software. The initial function point calculation result information includes the function point breakdown table information of each subsystem and the total original function point value of the software.

[0081] S35, perform consistency evaluation processing on the initial function point calculation result information to obtain the original function point calculation result information of the software.

[0082] It should be noted that the consistency assessment process refers to performing cross-process quantity and item consistency verification, multi-dimensional consistency calculation and recursive completion on the documents in the system partition list directory, subsystem technical requirements directory, function decomposition directory, function use case design directory and function point calculation directory, and generating full-process execution report information when consistency is achieved, and determining the initial function point calculation result information as the original function point calculation result information of the software.

[0083] As can be seen, the software original function point estimation method based on multi-agent collaboration described in the embodiments of the present invention realizes the automatic quantification and overall consistency verification of the software original function points by iteratively traversing the functional use case design document information, identifying functional units, matching function points and evaluating consistency.

[0084] In another optional embodiment, the step of performing functional unit identification processing on the t-th functional use case design document information in the functional use case design document information set to obtain the t-th functional unit identification result information includes: S321, Perform data function identification processing on the t-th functional use case design document information to obtain data function unit information; It should be noted that the data function identification processing refers to extracting data objects with persistent storage characteristics from the input data set information, step-by-step operation process information, and normal output result information of the t-th functional use case design document information; merging data objects pointing to the same logical data according to data object name, field composition, and association relationship; and comparing the maintenance location and access method of each data object. When a data object is added, modified, or deleted by the software subsystem corresponding to the t-th functional use case design document information, the type of the corresponding functional unit information is determined to be an internal logical file; when a data object is maintained by other software subsystems outside the software subsystem and is only read or referenced by the software subsystem, the type of the corresponding functional unit information is determined to be an external interface file, thus obtaining the data functional unit information. The data functional unit information includes functional unit information of type internal logical file or external interface file. The above data function identification processing can be performed using named entity recognition, semantic role labeling, entity referencing resolution, or data flow analysis algorithms; specifically, this embodiment of the invention does not limit the specific methods used.

[0085] S322, Perform transaction function identification processing on the t-th functional use case design document information to obtain transaction function unit information; It should be noted that the transaction function identification and processing refers to extracting independently complete business processing procedures from the execution preconditions, input data set information, step-by-step operation process information, and normal output result information of the t-th functional use case design document information, and identifying the data transmission direction, internal logic file maintenance operation, and data derivation operation of each business processing procedure; when external data enters the software subsystem and causes the addition, modification, or deletion of internal logic files, or changes the processing state of the software subsystem, the type of the corresponding functional unit information is determined as external input; when data leaves the software subsystem and includes data calculation, derived data generation, or internal logic file maintenance, the type of the corresponding functional unit information is determined as external output; when data leaves the software subsystem and only includes data retrieval, without including derived data generation or internal logic file maintenance, the type of the corresponding functional unit information is determined as external query, thus obtaining the transaction function unit information. The transaction function unit information includes functional unit information of type external input, external output, or external query. The above-mentioned transaction function identification and processing can be performed using event extraction, dependency parsing, data flow analysis, or large language model text classification algorithms. Specifically, the embodiments of the present invention are not limited to these methods.

[0086] S323, integrate the data function unit information and the transaction function unit information to obtain the identification result information of the t-th function unit.

[0087] As can be seen, the software original function point estimation method based on multi-agent collaboration described in the embodiments of the present invention achieves complete comparison and identification of five types of functional units through data function identification and transaction function identification.

[0088] In another optional embodiment, the consistency evaluation processing of the initial function point calculation result information to obtain the original function point calculation result information of the software includes: S351, perform consistency statistical processing on the documents in the system partitioning list directory, subsystem technical requirements directory, function decomposition directory, function use case design directory and function point calculation directory to obtain the quantity deviation rate value and the item mismatch rate value. It should be noted that the consistency statistical processing includes performing quantity deviation statistics on subsystem-level documents in each directory based on the subsystem partitioning list information, and performing item mismatch statistics on corresponding items in subsequent directories based on the minimum functional item information in the functional decomposition directory. The quantity deviation rate is used to characterize the proportion of missing, duplicate, or mismatched subsystem-level documents in the subsystem technical requirements directory, functional decomposition directory, and functional use case design directory; the item mismatch rate is used to characterize the proportion of items in the functional use case design directory and functional point calculation directory that cannot establish a one-to-one correspondence with the minimum functional item information.

[0089] S352, calculate and process the quantity deviation rate value and the item mismatch rate value to obtain a multidimensional consistency value; It should be noted that the above calculation process can be performed using a multidimensional consistency evaluation calculation model, or using a weighted arithmetic average algorithm or a fuzzy comprehensive evaluation algorithm. Specifically, the embodiments of the present invention do not limit the specifics.

[0090] The multidimensional consistency evaluation calculation model is as follows: ; ; ; In the formula, The quantity deviation rate value, , This is the quantity deviation attenuation coefficient. For components with consistent quantity, ; The mismatch rate value for the item. , For the consistent power exponent of the entries, For the consistency component of the entries, ; The multidimensional consistency value, .

[0091] It should be noted that the value range of the quantity deviation attenuation coefficient is [2,6], with a preferred value of 4; the value range of the item consistency power exponent is [1,3], with a preferred value of 2.

[0092] It should be noted that the multidimensional consistency value is obtained by the geometric mean of the quantity consistency component and the item consistency component. When both components are equally high, they are close to the upper bound; when either is low, the value is significantly lowered but does not collapse drastically. The multidimensional consistency value serves as a consistency criterion in the consistency assessment process, used to determine whether to trigger recursive completion of the corresponding process. This allows the scheme to uniformly measure the quantity and item consistency across processes with a single scalar, drive recursive completion, and serve the overall goal of a complete and hierarchical evidence chain that traces back from the function point calculation results to the source of constraints. Compared with the shortcomings of existing technologies that use simple addition or minimum values ​​to measure quantity and item consistency, resulting in quantity matching but item mismatch, or item matching but quantity missing values ​​being masked by a single dimension, the geometric mean strikes a balance between penalizing imbalance and avoiding excessive collapse, making the multidimensional consistency verification more robust.

[0093] S353, determine whether the multidimensional consistency value is not less than the preset consistency threshold, and obtain the sixth judgment result; When the sixth judgment result is yes, traceability report generation processing is performed on the intermediate results of each process to obtain the full process execution report information, and the initial function point calculation result information is determined to be the original function point calculation result information of the software. When the sixth judgment result is negative, the documents in the directory that do not meet the consistency standard are updated to obtain the updated directory documents, and S351 is executed.

[0094] It should be noted that the intermediate deliverables of each process refer to the information stored in the system partitioning list, subsystem technical requirement document set, function decomposition document set, function use case design document set, and initial function point calculation result information of each process, including subsystem partitioning, technical requirement separation, function decomposition, use case modeling, and function point splitting. These deliverables are stored in the system partitioning list directory, subsystem technical requirement directory, function decomposition directory, function use case design directory, and function point calculation directory, respectively. The traceability report generation process refers to the aggregation of the input and output file paths, execution time, and five-level directory storage paths of the intermediate deliverables of each process, and the organization of these into a full-process execution report information based on the relationship between the initial function point calculation result information and the upstream process deliverables, layer by layer.

[0095] It should be noted that the process of updating documents in the consistency non-compliance directory refers to the following: First, the subsystem technical requirements directory, functional decomposition directory, or functional use case design directory pointed to when the quantity deviation rate value exceeds the upper limit of the consistency threshold, and the functional decomposition directory, functional use case design directory, or functional point calculation directory pointed to when the item mismatch rate value exceeds the corresponding mismatch upper limit, are identified as directories to be updated; then, the global orchestration and scheduling main agent calls the special task agent corresponding to the directory to be updated to regenerate the corresponding document with the missing or mismatched subsystem identification information and minimum functional item information in the directory to be updated as the target, and obtain the updated directory document; finally, the updated directory document is overwritten into the directory to be updated. For example, when the quantity deviation rate exceeds the limit, the technical requirement separation agent or the function decomposition agent is invoked to regenerate the missing subsystem-level document; when the item mismatch rate exceeds the limit, the function decomposition agent or the use case modeling agent is invoked to regenerate or align the document corresponding to the mismatched minimum functional item information; after overwriting, S351 is executed to re-perform consistency statistics, and this process is repeated until the sixth judgment result is yes. The above update processing can be performed using file path index backtracking, directed acyclic graph process arrangement, or large language model content completion. Specifically, this embodiment of the invention does not limit the specific processing. The consistency threshold range is [0.9, 0.99], with a preferred value of 0.95.

[0096] As can be seen, the software original function point estimation method based on multi-agent collaboration described in the embodiments of the present invention achieves overall consistency verification and verified output of function point calculation results through cross-process consistency statistics, multi-dimensional consistency calculation and recursive completion.

[0097] In another optional embodiment, the process of generating traceability reports for intermediate results of each process to obtain full-process execution report information includes: S3531, aggregate the input document paths, execution times, and five-level directory storage paths of each process to obtain process path aggregation information; It should be noted that the aforementioned processes refer sequentially to the subsystem partitioning process, technical requirement separation process, functional decomposition process, use case modeling process, and function point splitting process. The aggregation process refers to reading the input document path, start time, end time, output document path, and corresponding directory storage path of each process, and associating the paths of each process with their execution time according to the processing order by utilizing the correspondence between the output document path of the previous process and the input document path of the next process. The process path aggregation information includes the storage paths of the original input document path, the execution time of each process, the system partitioning list directory, the subsystem technical requirement directory, the functional decomposition directory, the functional use case design directory, and the function point calculation directory. The above aggregation process can be processed using file system metadata reading, runtime log parsing, hash mapping association, or directed acyclic graph path association algorithms. Specifically, this embodiment of the invention does not limit the specific methods used.

[0098] S3532, perform report generation processing on the process path aggregation information and the software original function point calculation result information to obtain the full process execution report information.

[0099] It should be noted that the report generation process refers to starting with the subsystem identification information and function point breakdown table information in the software's original function point calculation results, and sequentially back-associating the corresponding functional use case design document information, function decomposition document information, subsystem technical requirement document information, and subsystem partitioning list information according to the path association relationship of each process in the process path aggregation information. The resulting path information, execution time of each process, total original function point values ​​of each subsystem, and execution exception records are then written into a preset report data structure. The full-process execution report information includes the original input document path, execution time of each process, five-level directory storage path, total original function point values ​​of each subsystem, execution exception records, and full-link reverse tracing path. The above report generation process can be performed using template filling, structured data serialization, graph path backtracking, or a document generation engine; specifically, this embodiment of the invention does not limit the specific methods used.

[0100] As can be seen, by implementing the software original function point estimation method based on multi-agent collaboration described in the embodiments of the present invention, a full-link evidence chain that can be traced back layer by layer is constructed through process path aggregation and report generation.

[0101] Experimental Results Description To verify the technical effectiveness of the software original function point estimation method based on multi-agent collaboration described in this invention embodiment, a software original function point estimation system was tested using the overall technical document of a large-scale simulation software as the original software technical document information. The overall technical document contains approximately 92,000 words and includes 6 first-level subsystems, 21 performance mandatory constraint markers, and 16 security mandatory constraint markers, representing a typical multi-branch complex software project. During the test, the overall technical document in a tagged text format with unified character encoding was used as the sole input. The global orchestration and scheduling master agent completed the file readability and integrity comparison, created a five-level directory, and then sequentially scheduled each specialized task agent to perform subsystem division, technical requirement separation, functional decomposition, use case modeling, and function point splitting, and performed overall consistency assessment and traceability report generation. Test results show that: First, after adopting the multi-agent context physical isolation and file-driven processing method, the entire process stably processes the 92,000-word overall technical document without exceeding the context window capacity or text truncation, identifying 6 first-level subsystems and generating 6 subsystem technical requirement documents, with no missing items in the quantity comparison and matching; Second, after using the mandatory constraint binding completeness evaluation calculation model for functional decomposition verification, all performance mandatory constraint tags and security mandatory constraint tags are bound to the corresponding minimum functional item information, with no loose constraints, and the measured functional omission rate is controlled below 0.5%; Third, after using the multi-dimensional consistency evaluation calculation model for overall consistency evaluation, the names, quantities, and order of documents at all levels of functional decomposition, functional use case design, and functional point calculation are consistent, and the consistency of results from multiple processing reaches over 90%, with a calculation deviation of less than 3%; Fourth, in terms of processing efficiency, compared to the at least 20 person-days required for manual functional point estimation of a document of the same scale, this method can output the original functional point calculation results and the entire process execution report information in about 2 hours of fully automatic operation. The above results verify the technical effectiveness of this method in terms of completeness, consistency, and processing efficiency.

[0102] Example 2 Please see Figure 2 , Figure 2 This is a schematic diagram of a software original function point estimation device based on multi-agent collaboration disclosed in an embodiment of the present invention. Figure 2 The described software raw function point estimation device based on multi-agent collaboration is applied to software raw function point estimation systems, such as local servers or cloud servers used for software size measurement, etc., and the embodiments of the present invention are not limited thereto. Figure 2 As shown, the device includes: Module 201 is used to acquire original software technical document information; The parsing module 202 is used to parse and process the original software technical document information to obtain a functional use case design document information set; The function point calculation module 203 is used to perform function point calculation processing on the function use case design document information set to obtain the original function point calculation result information of the software.

[0103] As can be seen, the software original function point estimation device based on multi-agent collaboration described in the embodiments of the present invention achieves end-to-end fully automatic estimation from original software technical document information to software original function point calculation result information through multi-agent context physical isolation and file-driven serial collaboration.

[0104] Example 3 Please see Figure 3 , Figure 3 This is a schematic diagram of another software raw function point estimation device based on multi-agent collaboration disclosed in an embodiment of the present invention. Figure 3 The described software raw function point estimation device based on multi-agent collaboration is applied to software raw function point estimation systems, such as local servers or cloud servers used for software size measurement, etc., and the embodiments of the present invention are not limited thereto. Figure 3 As shown, the device includes: Processor 301; A memory 302 containing executable program code is coupled to the processor 301; The processor 301 calls the executable program code stored in the memory 302 to execute some or all of the steps of the software primitive function point estimation method based on multi-agent cooperation in Embodiment 1.

[0105] As can be seen, the software original function point estimation device based on multi-agent collaboration described in the embodiments of the present invention achieves end-to-end fully automatic estimation from original software technical document information to software original function point calculation result information through multi-agent context physical isolation and file-driven serial collaboration.

[0106] Example 4 This invention discloses a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform the steps in a software primitive function point estimation method based on multi-agent cooperation described in Embodiment 1.

[0107] Example 5 This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps in the software primitive function point estimation method based on multi-agent cooperation described in Embodiment 1.

[0108] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0109] It should be noted that all calculation expressions or mathematical functions in the embodiments of the present invention have undergone dimensionless processing of the variables involved before calculation.

[0110] It should be noted that in all the calculation expressions or mathematical functions in the embodiments of the present invention, the values ​​of the input independent variables all meet the reasonable requirements of the input value range of the calculation expression or mathematical function, and can ensure that the calculation expression or mathematical function can be calculated smoothly without violating physical laws or mathematical rules.

[0111] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), once programmable read-only memory (OTPROM), electronically erasable rewritable read-only memory (EEPROM), read-only optical disc (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0112] Finally, it should be noted that the software original function point estimation method based on multi-agent collaboration disclosed in the embodiments of the present invention is only a preferred embodiment of the present invention and is only used to illustrate the technical solution of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for estimating the original function points of software based on multi-agent cooperation, characterized in that, The method includes: S1, Obtain original software technical documentation information; S2, the original software technical document information is parsed and processed to obtain a functional use case design document information set; S3, perform function point calculation processing on the functional use case design document information set to obtain the original function point calculation result information of the software.

2. The method for estimating the original function points of software based on multi-agent collaboration according to claim 1, characterized in that, The process of parsing and processing the original software technical document information to obtain a functional use case design document information set includes: S21, Perform subsystem division processing on the original software technical document information to obtain subsystem division list information; S22, Perform technical requirement separation processing on the original software technical document information and the subsystem partitioning list information to obtain a subsystem technical requirement document information set; S23, Perform functional decomposition processing on the subsystem technical requirements document information set to obtain a functional decomposition document information set; S24, Perform use case modeling processing on the functional decomposition document information set to obtain the functional use case design document information set.

3. The method for estimating the original function points of software based on multi-agent collaboration according to claim 2, characterized in that, The process of dividing the original software technical document information into subsystems to obtain a subsystem division list includes: S211, Perform module identification processing on the original software technical document information to obtain a candidate module information set; S212, default s=1; S213, perform feature extraction processing on the s-th candidate module information in the candidate module information set to obtain the s-th module cohesion value and the s-th cross-module coupling value; S214, using the subsystem partitioning confidence calculation model, calculate and process the cohesion value of the s-th module and the cross-module coupling value to obtain the s-th subsystem partitioning confidence value; The confidence calculation model for the subsystem partitioning is as follows: ; ; In the formula, Let be the cohesion value of the s-th module. , For the s-th cross-module coupling value, , The cohesive saturation coefficient. For the s-th cohesive saturation, , This is the coupling sharpening coefficient. Assign confidence values ​​to the s-th subsystem. ; S215, determine whether the confidence value of the s-th subsystem division is not less than the preset confidence threshold, and obtain the first judgment result; When the first judgment result is yes, the s-th candidate module information is determined to be the s-th subsystem identification information; When the first judgment result is negative, the s-th candidate module information is subjected to boundary adjustment processing to obtain a boundary-adjusted candidate module information set, and the boundary-adjusted candidate module information set is determined as the candidate module information set. S212 is executed. S216, determine whether s is equal to the number of candidate module information in the candidate module information set, and obtain a second judgment result; When the second judgment result is negative, increment s by 1 and execute S213; When the second judgment result is yes, all the subsystem identification information is collected and processed to obtain the subsystem division list information.

4. The method for estimating the original function points of software based on multi-agent collaboration according to claim 3, characterized in that, The step of performing feature extraction processing on the s-th candidate module information in the candidate module information set to obtain the s-th module cohesion value and the s-th cross-module coupling value includes: S2131, Perform correlation degree statistical processing on the s-th candidate module information to obtain the correlation degree value of the s-th internal item and the correlation degree value of the s-th cross-module item; S2132, normalize the correlation value of the s-th internal entry and the correlation value of the s-th cross-module entry to obtain the cohesion value of the s-th module and the coupling value of the s-th cross-module.

5. The method for estimating the original function points of software based on multi-agent collaboration according to claim 2, characterized in that, The step of separating the original software technical document information and the subsystem partitioning list information into technical requirements results in a subsystem technical requirement document information set, including: S221, traverse each subsystem identifier information in the subsystem partitioning list information, extract the exclusive technical clause information corresponding to each subsystem identifier information from the original software technical document information, and obtain the subsystem exclusive technical clause information set; S222, the subsystem-specific technical clause information set is marked and retained to obtain the subsystem technical requirement document information set.

6. The method for estimating the original function points of software based on multi-agent cooperation according to claim 5, characterized in that, The process of marking and retaining the information set of specific technical clauses for the subsystem yields a set of technical requirement documents for the subsystem, including: S2221, Traverse each subsystem-specific technical clause information in the subsystem-specific technical clause information set, and perform mark retention processing on the performance mandatory constraint mark in each subsystem-specific technical clause information to obtain the technical clause information set after performance mark retention; S2222, Perform mark retention processing on the security mandatory constraint marks in the technical clause information set after the performance mark retention is obtained to obtain the subsystem technical requirement document information set.

7. The method for estimating the original function points of software based on multi-agent collaboration according to claim 2, characterized in that, The process of functionally decomposing the technical requirements document information set of the subsystem to obtain a functionally decomposed document information set includes: S231, preset k=1; S232, perform structured decomposition processing on the kth subsystem technical requirement document information in the subsystem technical requirement document information set to obtain the kth initial functional decomposition document information; S233, perform constraint binding processing on the kth initial functional decomposition document information to obtain the kth bound functional decomposition document information, the kth performance constraint coverage value, and the kth security constraint coverage value; S234, calculate and process the kth performance constraint coverage value and the kth security constraint coverage value to obtain the kth constraint binding completeness value; S235, determine whether the binding completeness value of the kth constraint is not less than the preset binding completeness threshold, and obtain the third judgment result; When the third judgment result is yes, the kth bound function decomposition document information is determined to be the kth function decomposition document information; When the third judgment result is negative, the initial functional decomposition document information of the kth initial functional decomposition document information is recursively re-decomposed to obtain the re-decomposed initial functional decomposition document information. The re-decomposed initial functional decomposition document information is determined to be the initial functional decomposition document information of the kth initial functional decomposition document information, and S233 is executed. S236, determine whether k is equal to the number of subsystem technical requirement document information in the subsystem technical requirement document information set, and obtain the fourth judgment result; When the fourth judgment result is negative, increment k by 1 and execute S232; When the fourth judgment result is yes, all functional decomposition document information is collected and processed to obtain a functional decomposition document information set.

8. The method for estimating the original function points of software based on multi-agent collaboration according to claim 2, characterized in that, The process of performing use case modeling on the functional decomposition document information set to obtain a functional use case design document information set includes: S241, perform element completion processing on each minimum functional item information in the functional decomposition document information set to obtain a functional use case information set after element completion. S242, After the elements are completed, the functional use case information set is standardized to obtain a standardized functional use case information set; S243, Perform constraint embedding comparison processing on the standardized functional use case information set to obtain the functional use case design document information set.

9. The method for estimating the original function points of software based on multi-agent cooperation according to claim 1, characterized in that, The process of performing function point calculation on the functional use case design document information set to obtain the original function point calculation results of the software includes: S31, preset t=1; S32, perform functional unit identification processing on the t-th functional use case design document information in the functional use case design document information set to obtain the t-th functional unit identification result information; S33, Perform function point matching processing on the identification result information of the tth functional unit to obtain the tth function point splitting table information; S34, determine whether t is equal to the number of functional use case design document information in the functional use case design document information set, and obtain the fifth judgment result; When the fifth judgment result is negative, increment t by 1 and execute S32; When the fifth judgment result is yes, all the function point split table information is summarized and processed to obtain the initial function point calculation result information; S35, perform consistency evaluation processing on the initial function point calculation result information to obtain the original function point calculation result information of the software.

10. The method for estimating the original function points of software based on multi-agent cooperation according to claim 9, characterized in that, The step of performing functional unit identification processing on the t-th functional use case design document information in the functional use case design document information set to obtain the t-th functional unit identification result information includes: S321, Perform data function identification processing on the t-th functional use case design document information to obtain data function unit information; S322, Perform transaction function identification processing on the t-th functional use case design document information to obtain transaction function unit information; S323, integrate the data function unit information and the transaction function unit information to obtain the identification result information of the t-th function unit.