Business demand analysis method and device and computer readable storage medium

Abstract

The invention provides a service demand analysis method and device and a computer readable storage medium, and the method comprises the steps: obtaining all dialogue voice data discussed by a service demand, and converting all dialogue voice data into corresponding dialogue text documents; generating a preliminary demand document based on all dialogue text documents; performing information extraction on the preliminary demand document by using a big and small language model according to a set demand template to obtain specific information, and filling the specific information to a position corresponding to the set demand template to obtain a demand document; and performing function point analysis on the demand document by using the big and small language models to generate a function demand specification. The problem that in the prior art, work such as collection and summarization, key point extraction, induction analysis and content checking of massive multi-source heterogeneous original documents consumes a large amount of time cost is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of requirements analysis. Specifically, it relates to a business requirements analysis method, a business requirements analysis device, a computer-readable storage medium, and a computer program product. Background Art

[0002] In the software development process, business requirements analysis and management is a key and core task and link. As a bridge between the business and technical teams, requirements analysts often need to communicate with the business team repeatedly to clarify business requirements, and then negotiate solutions with the technical team. Requirements will change repeatedly during the refinement of business functions. The traditional manual processing of requirement documents is not only inefficient but also prone to errors and omissions of requirements. With the development of natural language processing technology, especially the application of deep learning and large models, it has become possible to automatically process requirements and intelligently generate requirement documents. However, there are some deficiencies in the existing artificial intelligence automatic processing methods. For example, the processing effect of a single model on a large amount of multi-source heterogeneous documents is limited, the extraction accuracy of core knowledge is not high, the understanding of business requirements is not deep enough, the accuracy of the generated requirement documents is not high, and the ability to ask questions based on the understanding of business requirements is not available.

[0003] A language model-driven enterprise virtual office assistant system and implementation method in the technical field of virtual office assistants are disclosed in the prior art. The system includes a conversation collection module, a conversation record module, etc. However, in the case of overlapping voices generated by multiple people talking simultaneously, the speech recognition error rate is high, and the recognition result may be deviated. After retrieving the prior art, there is no automated, intelligent, and process-based requirements analysis and knowledge extraction and orchestration technology that introduces seamless cooperation between large and small models for business requirements collection and analysis to ensure the accuracy, comprehensiveness, and efficiency of requirements analysis. For requirements analysts, in the process of business requirements collection and analysis, they often rely on their own excellent professional skills, including excellent communication and expression skills, strong business knowledge learning ability, excellent logical analysis ability, and proficiency in using computer office software.

[0004] The existing technical solutions often solve problems in general fields, such as speech-to-text methods and devices, intelligent dialogue and question-answering systems based on large models, or provide AI assistants for specific fields, but there is no AI intelligent analysis assistant for the requirements analysis field.

[0005] Currently, in the field of software engineering, requirements analysts have the following problems in the process of requirements collection, analysis, requirements knowledge extraction, requirement document writing, etc.:

[0006] 1) Low efficiency: The work of collecting and summarizing, extracting key points, inductively analyzing, and checking the content of a large amount of multi-source heterogeneous original documents consumes a lot of time and energy;

[0007] 2) Uneven professionalism: Different people have different descriptions and understandings of professional terms;

[0008] 3) The template is not flexible enough: The template for demand analysis is too fixed, which makes it inconvenient for project stakeholders to view and track functional items according to different dimensions. Summary of the Invention

[0009] The main purpose of this application is to provide a business needs analysis method, a business needs analysis device, a computer-readable storage medium and a computer program product, so as to at least solve the problem in the prior art that the collection, aggregation, key point extraction, inductive analysis, content verification and other tasks of massive multi-source heterogeneous original documents consume a lot of time and cost.

[0010] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a business demand analysis method is provided, including: obtaining all conversation voice data of business demand discussions, and converting all of the conversation voice data into corresponding conversation text documents; generating a preliminary demand document based on all of the conversation text documents; extracting information from the preliminary demand document using a large and small language model according to a set demand template to obtain specific information, and filling the specific information into the position corresponding to the set demand template to obtain a demand document, wherein the large and small language model is a model that uses a large model and a small model to work together, the large model is used to understand the overall semantics of the preliminary demand document, and the small model is used to accurately analyze the preliminary demand document to assist the large model in filling the gaps in semantic understanding; using the large and small language models to perform functional point analysis on the demand document to generate a functional requirement specification, so as to carry out business development according to the functional requirement specification, so that the developed product has the functional points of the functional requirement specification.

[0011] Optionally, all of the conversation voice data are converted into corresponding conversation text documents, including: using a large speech model to perform text translation on the conversation voice data to generate the corresponding conversation text documents, the large speech model is a convergent model obtained by iteratively training a predetermined model using multiple groups of sample data, and each group of sample data includes sample speech data and sample text data corresponding to the sample speech data.

[0012] Optionally, generate a preliminary requirements document based on all the conversation text documents, including: performing preliminary verification and secondary verification on each of the conversation text documents in sequence to obtain initial requirements. The preliminary verification is a verification operation to correct mis-identified requirement information, and the requirement information at least includes requirement priority and detailed requirement description. The secondary verification is to supplement missing valid requirement content, and the valid requirement content at least includes key requirement content not mentioned during the conversation process and developer task allocation; performing preliminary analysis and filtering on the content of the initial requirements to remove information irrelevant to the business requirements or duplicate and redundant information, thereby forming the preliminary requirements document.

[0013] Optionally, the specific information at least includes requirement name, detailed requirement description, requirement priority, associated system, and expected upper limit date / launch date. Use a large language model to extract information from the preliminary requirements document according to a set requirements template to obtain the specific information, and fill the specific information into the corresponding positions of the set requirements template to obtain the requirements document, including: setting the filling area for the specific information to obtain the set requirements template; using the attention mechanism of the large language model to filter redundant information from the preliminary requirements document to form a requirements analysis text; extracting information from the requirements analysis text according to preset keywords to screen out statement information related to the preset keywords, thereby obtaining the requirement name, detailed requirement description, associated system, and expected upper limit date / launch date in the specific information; determining the requirement priority in the specific information according to keyword frequency, where the keyword frequency includes high-frequency keywords and low-frequency keywords, and the requirement priority includes high priority and low priority; filling the specific information into the corresponding filling area of the set requirements template to obtain the requirements document.

[0014] Optionally, after determining the requirement priority in the specific information according to the keyword frequency, the method further includes: calculating the word frequency score corresponding to each word in the dialogue text document according to a first formula, where the first formula is tfidf(t, d, D) = tf(t, d) × idf(t, D), where tfidf(t, d, D) represents the word frequency score, tf(t, d) is the term frequency, tf(t, d) = log(1 + freq(t, d)), freq(t, d) represents the number of times the candidate word t appears in the current dialogue text document d, the candidate word is the word that appears in the dialogue text document, idf(t, D) = log(N / count(d ∈ D: t ∈ D)) is the inverse document frequency, count(d ∈ D: t ∈ D) is the number of dialogue text documents containing the candidate word t, and N represents the total number of dialogue text documents; determining the words with the word frequency score greater than or equal to a set score as high-frequency keywords; determining the words with the word frequency score less than the set score as low-frequency keywords, and D represents the set of all dialogue text documents.

[0015] Optionally, use the large and small language model to perform functional point analysis on the requirement document to generate a functional requirement specification, including: defining the document format of the functional requirement specification, where the document format at least includes functional points and functional descriptions; using the large model in the large and small language model to perform semantic analysis on the requirement document to determine multiple functional points, where the functional points at least include the trigger conditions, execution processes, expected outputs, exception handling, and mutual dependencies of the functions, and the mutual dependencies include the execution order between functions and the priorities between functions; using the small model in the large and small language model to perform semantic refinement on each functional point to determine the functional description corresponding to each functional point; using the text generation ability of the large model to convert the functional points and the functional descriptions corresponding to the functional points into the form of the document format, and at the same time using the small model to check whether the conversion result meets the document requirements during the conversion process, where the document requirements at least include conforming to the document format and the correct use of professional terms; in the case where the conversion result meets the document requirements, determining the conversion result as the functional requirement specification; in the case where the conversion result does not meet the document requirements, regenerating the functional requirement specification.

[0016] Optionally, after performing functional point analysis on the requirements document using the large language model to generate a functional requirements specification, the method further includes: determining various dimension information for requirement transposition, where the dimension information at least includes a list of personnel, system components, business process stages, requirement priorities, development stages, requirement functional points, and requirement progress; automatically identifying and extracting field information and requirement descriptions of each dimension information in the requirements document; determining corresponding transposition rules according to each dimension information, so as to perform a transposition operation according to the transposition rules and form a list of function items for different dimension information, where the transposition rule is a rule that will be displayed in different forms when different dimension information is selected, and the transposition operation is to automatically display the field information and requirement description corresponding to the dimension information in a corresponding predetermined form when the corresponding dimension information is selected, and the predetermined form at least includes a list form, a Gantt chart form, and a table form.

[0017] According to another aspect of the present application, there is provided a business requirements analysis device, where the device includes: an acquisition unit, configured to acquire all conversation voice data of business requirements discussions and convert all the above conversation voice data into corresponding conversation text documents; a first generation unit, configured to generate a preliminary requirements document based on all the above conversation text documents; an extraction and filling unit, configured to use a large language model to extract information from the above preliminary requirements document according to a set requirements template to obtain specific information, and fill the above specific information into the corresponding position of the above set requirements template to obtain a requirements document, where the large language model is a model that uses a large model and a small model to work together, the large model is used for overall semantic understanding of the above preliminary requirements document, and the small model is used for precise analysis of the above preliminary requirements document to assist the large model in filling in the missing parts of semantic understanding; a second generation unit, configured to use the above large language model to perform functional point analysis on the above requirements document to generate a functional requirements specification, so as to perform business development according to the above functional requirements specification, so that the developed product has the functional points of the above functional requirements specification.

[0018] According to still another aspect of the present application, there is provided a computer-readable storage medium, where the computer-readable storage medium includes a stored program, and when the above program runs, it controls the device where the computer-readable storage medium is located to execute any one of the above methods.

[0019] According to yet another aspect of the present application, there is provided a computer program product, including computer instructions, where when the above computer instructions are executed by a processor, they implement any one of the above methods.

[0020] Applying the technical solution of the present application in the business requirement analysis method, first, all the conversation voice data of the business requirement discussion is obtained, and all the above-mentioned conversation voice data is converted into corresponding conversation text documents; then, a preliminary requirement document is generated based on all the above-mentioned conversation text documents; after that, according to the set requirement template, the large and small language models are used to extract information from the above-mentioned preliminary requirement document to obtain specific information, and the above-mentioned specific information is filled into the corresponding position of the above-mentioned set requirement template to obtain a requirement document. The above-mentioned large and small language models are models that work collaboratively using a large model and a small model. The above-mentioned large model is used for the overall semantic understanding of the above-mentioned preliminary requirement document, and the above-mentioned small model is used for the precise analysis of the above-mentioned preliminary requirement document to assist the above-mentioned large model in filling the semantic understanding gaps; finally, the above-mentioned large and small language models are used to analyze the function points of the above-mentioned requirement document to generate a functional requirement specification, so as to carry out business development according to the above-mentioned functional requirement specification, so that the developed product has the function points of the above-mentioned functional requirement specification. The present application obtains all the conversation voice data of the business requirement, identifies the voice content and converts it into the corresponding text document, and generates a preliminary requirement document based on the text document for requirement analysis; according to the specified requirement template, the large and small language models are used to extract effective information, and the extracted specified information is filled into the corresponding position of the template; according to the business requirements, the large and small language models are used to cooperate with each other to complete the function point analysis of the filled business requirement document, generate a functional requirement specification, and provide guidance for system design and development according to the generated detailed functional requirement specification. There is no need for developers to collect, summarize, extract key points, inductively analyze, and check the content of a large number of multi-source heterogeneous original documents to obtain a functional requirement specification, which can improve the efficiency of requirement analysis in software project development for banking business and greatly reduce the time cost of developers. The present application solves the problem that a large amount of time cost is consumed in the prior art for the collection, summary, key point extraction, inductive analysis, and content checking of a large number of multi-source heterogeneous original documents. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 FIG. shows a hardware structure block diagram of a mobile terminal for implementing a business requirement analysis method provided in an embodiment of the present application;

[0022] Figure 2 FIG. shows a flowchart of a business requirement analysis method provided in an embodiment of the present application;

[0023] Figure 3 FIG. shows an architecture diagram of a business requirement analysis system based on the collaboration of large and small models provided in an embodiment of the present application;

[0024] Figure 4 FIG. shows a flowchart of information extraction provided in an embodiment of the present application;

[0025] Figure 5 shows a flowchart of an intelligent transposition provided according to an embodiment of the present application;

[0026] Figure 6 shows a schematic flowchart of a specific business requirement analysis method provided according to an embodiment of the present application;

[0027] Figure 7 shows a structural block diagram of a business requirement analysis device provided according to an embodiment of the present application.

[0028] Among them, the above-mentioned drawings include the following reference numerals:

[0029] 102, processor; 104, memory; 106, transmission device; 108, input / output device. Detailed implementation manners

[0030] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0031] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present application.

[0032] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such used data may be interchanged under appropriate circumstances so as to describe the embodiments of the present application here. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily need to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these process, method, product or device.

[0033] For the convenience of description, some nouns or terms related to the embodiments of the present application are described below:

[0034] Small model: Usually refers to a model using a neural network architecture, including shallow neural networks, lightweight convolutional neural networks (such as MobileNet), or small recurrent neural networks (such as LSTM or GRU), etc.

[0035] Large model: Usually refers to an artificial intelligence model with a large number of parameters. These models are typically trained on large-scale datasets, can perform complex tasks, and have high performance. In the field of artificial intelligence, especially in deep learning, "large models" often refer to pre-trained models, which are characterized by a large number of parameters, strong generalization ability, wide adaptability, and high resource consumption.

[0036] Large speech model: Refers to a large deep learning model used in tasks such as speech recognition or speech synthesis. These models usually have a large number of parameters and are trained on large-scale audio datasets to achieve high-quality speech processing functions.

[0037] Large language model: Refers to a language model with a large number of parameters. These models are usually constructed based on deep learning techniques and can perform well in natural language processing (NLP) tasks. Such models are typically pre-trained on large-scale text data, can understand and generate human language, and demonstrate powerful capabilities in various language-related tasks.

[0038] Transpose: A basic concept in linear algebra, referring to the operation of interchanging the rows and columns of a matrix.

[0039] As introduced in the background art, the prior art is inefficient: tasks such as the collection and summary, key point extraction, inductive analysis, and content verification of a large amount of multi-source heterogeneous original documents consume a large amount of time and effort. To solve the problem of the large time cost consumed by tasks such as the collection and summary, key point extraction, inductive analysis, and content verification of a large amount of multi-source heterogeneous original documents, embodiments of the present application provide a business requirement analysis method, a business requirement analysis device, a computer-readable storage medium, and a computer program product.

[0040] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention.

[0041] The method embodiments provided in the embodiments of the present application can be executed on a mobile terminal, a computer terminal, or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 is a hardware structure block diagram of a mobile terminal of a business requirement analysis method according to an embodiment of the present invention. As Figure 1 shown, the mobile terminal may include one or more ( Figure 1Only one processor 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a field-programmable gate array FPGA) and a memory 104 for storing data are shown. Among them, the above mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 The structure shown is only schematic and does not limit the structure of the above mobile terminal. For example, the mobile terminal may further include more or fewer components than Figure 1 shown therein, or have a different configuration from Figure 1 shown therein.

[0042] The memory 104 can be used to store computer programs. For example, software programs and modules of application software, such as the computer program corresponding to the business requirement analysis method in the embodiments of the present invention. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, that is, implements the above method. The memory 104 may include a high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories may be connected to the mobile terminal through a network. Examples of the above network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the above network may include a wireless network provided by a communication provider of the mobile terminal. In one instance, the transmission device 106 includes a network adapter (abbreviated as NIC), which can be connected to other network devices through a base station and thus communicate with the Internet. In one instance, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0043] In this embodiment, a business requirement analysis method running on a mobile terminal, a computer terminal, or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order.

[0044] Figure 2 is a flowchart diagram of the business requirement analysis method according to an embodiment of the present application. As Figure 2 shown, the method includes the following steps:

[0045] Step S201: Obtain all the conversation voice data of the business requirement discussion, and convert all the above conversation voice data into corresponding conversation text documents.

[0046] Specifically, use a recording device or software to capture the voice communication of all participants during the requirement discussion meeting to obtain all the conversation voice data. These meetings can be face-to-face meetings, conference calls or video conferences to ensure comprehensive collection of the requirement discussion content.

[0047] After obtaining all the conversation voice data of the business requirement discussion and before converting all the above conversation voice data into corresponding conversation text documents, preprocess the collected audio files, including noise cancellation, format standardization (such as converting to WAV or MP3 format), and segment them into small segments suitable for processing to facilitate subsequent speech recognition. Speech-to-text: Use a speech large model (such as an optimized model based on Whisper) to perform speech recognition on each preprocessed audio segment and transcribe it into a text document. Automatically converting oral discussions into written documents saves a large amount of time and effort in manual recording. The text document format is convenient for further information extraction and analysis using natural language processing techniques.

[0048] Step S202: Generate a preliminary requirement document based on all the above conversation text documents.

[0049] Specifically, aggregate all the conversation text documents together, perform preliminary merging and sorting to construct a document containing all the discussion points. Through text analysis, remove non-critical information and duplicate content, and retain the core requirements and discussion points. Perform preliminary structuring, classification and annotation on the filtered document to form a rough requirement document framework, including requirement titles, descriptions, requirement priorities and requirement categories. This provides a starting point and basic framework for subsequent detailed requirement analysis. Aggregating scattered conversation information into relevant requirement descriptions facilitates understanding and further analysis.

[0050] Step S203: Extract information from the above preliminary requirements document using a large and small language model according to the set requirements template to obtain specific information, and fill the above specific information into the corresponding positions of the above set requirements template to obtain a requirements document. The above large and small language model is a model that uses a large model and a small model to work together. The above large model is used for the overall semantic understanding of the above preliminary requirements document, and the above small model is used for precise analysis of the above preliminary requirements document to assist the large model in filling the gaps in semantic understanding. The large model is trained to understand the overall semantics, and the small model is fine-tuned through a dataset in a specific domain to accurately identify specific business vocabulary and terms. Based on the set requirements template, a series of questions are designed. The large model is used to conduct a high-level analysis of the preliminary requirements document, and the small model is used to answer specific questions in detail, such as identifying specific functions, parameter settings, business processes, etc. The answers provided by the large model and the small model are integrated and filled into the corresponding positions according to the structure of the requirements template, such as function description, operation guide, expected results, etc. On the basis of information filling, the document is improved, including format adjustment, consistency check, and necessary supplementary explanations. The collaborative work of the large and small models ensures the accuracy of professional terms and comprehensive details in the requirements document. The use of the template ensures the unity and comparability of the document, facilitating communication and understanding among team members.

[0051] Step S204: Use the above large and small language model to conduct a function point analysis on the above requirements document to generate a functional requirements specification, so as to carry out business development according to the above functional requirements specification, so that the developed product has the function points of the above functional requirements specification.

[0052] Specifically, based on the overall semantic understanding, the large model identifies the function points and business processes in the document, and the small model deeply analyzes the specific requirements of each function point. The identified and analyzed function points are formatted into the standard format of the functional requirements specification, including function description, input / output definition, operation process, expected behavior, exception handling, etc. Through multiple rounds of questions and answers, the description of the function points is continuously refined, including specific data fields, user interface elements, algorithm logic, etc. Finally, the requirements analyst or domain expert can review the document to ensure that all key function points are covered and the description is accurate. The functional requirements specification provides a clear function development guide for the development team, reducing uncertainty and rework in the development process. The detailed function point analysis enables developers to directly code specific functions without the need for further requirement clarification. The entire process realizes the automated understanding and documentation of business requirements through the collaborative work of the large and small language models, not only improving the efficiency and quality of requirement analysis, but also providing a solid foundation for subsequent system development, ensuring a high degree of consistency between the developed product and the expected functional requirements.

[0053] In this embodiment, first, all the dialogue voice data of the business requirement discussion is obtained, and all the above-mentioned dialogue voice data is converted into corresponding dialogue text documents; then, a preliminary requirement document is generated based on all the above-mentioned dialogue text documents; after that, according to the set requirement template, the large and small language models are used to extract information from the above-mentioned preliminary requirement document to obtain specific information, and the above-mentioned specific information is filled into the corresponding positions of the above-mentioned set requirement template to obtain a requirement document. The above-mentioned large and small language models are models that work collaboratively using a large model and a small model. The above-mentioned large model is used for the overall semantic understanding of the above-mentioned preliminary requirement document, and the above-mentioned small model is used for the precise analysis of the above-mentioned preliminary requirement document to assist the above-mentioned large model in filling the gaps in semantic understanding; finally, the above-mentioned large and small language models are used to analyze the function points of the above-mentioned requirement document to generate a functional requirement specification, so as to carry out business development according to the above-mentioned functional requirement specification, so that the developed product has the function points of the above-mentioned functional requirement specification. This application obtains all the dialogue voice data of the business requirement, identifies the voice content and converts it into the corresponding text document, and generates a preliminary requirement document based on the text document for requirement analysis; according to the specified requirement template, the large and small language models are used to extract effective information, and the extracted specified information is filled into the corresponding positions of the template; according to the business requirements, the large and small language models are used to cooperate with each other to complete the function point analysis of the filled business requirement document, generate a functional requirement specification, and provide guidance for system design and development according to the generated detailed functional requirement specification. There is no need for developers to collect, summarize, extract key points, inductively analyze, and check the content of a large number of multi-source heterogeneous original documents to obtain a functional requirement specification, which can improve the efficiency of requirement analysis in software project development for banking business and greatly reduce the time cost of developers. This application solves the problem that the work of collecting, summarizing, extracting key points, inductively analyzing, and checking the content of a large number of multi-source heterogeneous original documents in the prior art consumes a large amount of time cost.

[0054] In order to enable those skilled in the art to more clearly understand the technical solution of this application, the implementation process of the business requirement analysis method of this application will be described in detail below in combination with specific embodiments.

[0055] In order to achieve the efficient conversion of dialogue voice data into text documents and improve the efficiency of requirement analysis, in an optional implementation manner, the above step S201 includes:

[0056] Step S2011, using a speech large model to perform text translation on the above-mentioned dialogue voice data to generate the corresponding above-mentioned dialogue text document. The above-mentioned speech large model is a convergent model obtained by iteratively training a predetermined model using multiple groups of sample data. Each group of the above-mentioned sample data includes sample voice data and the sample text data corresponding to the above-mentioned sample voice data.

[0057] In the above embodiment, the collected audio data is translated through a speech large model to automatically generate a text document in the form of conversation content. The speech large model is based on an open-source large model and is intensively trained using a large number of professional terms in the banking business knowledge base. For the problem of overlapping voices caused by multiple people speaking simultaneously during a meeting, a voice overlapping detection and segmentation algorithm is used for preprocessing, effectively improving the speech recognition accuracy. The iterative optimization during the training process of the speech large model ensures that the model has a high accuracy in recognizing professional terms in a specific field, reducing the probability of misrecognition caused by professional terms. Automatically converting speech into text significantly reduces the large amount of time and labor required for manual transcription, improving the efficiency of requirement analysis. The automated processing ensures that all the content discussed in the meeting can be accurately recorded, avoiding the omissions and inconsistencies that may occur in manual recording. The generated text document is convenient for further extracting requirement points, text analysis, and information integration using natural language processing technology, thus accelerating the entire requirement analysis process. Using the speech large model after tuning training to efficiently complete the processing of converting speech to a text document, the conversion accuracy reaches the level of human translation. As Figure 3 shown, the system realizes the translation of the above-mentioned speech data into a text document through a speech-to-text module. The above-mentioned speech-to-text module uses advanced speech recognition technology to accurately convert the speech data into text form and ensures that the generated text retains the integrity and naturalness of the original speech data. The module sends the speech data to the speech large model for speech-to-text processing, identifies the voice timbre to generate a unique identity identifier for processing, and then stores the converted text data in the text document in the format of {identity identifier: speech content}. Each segment of speech is saved independently with a timestamp as the file name, and the converted text document is saved in another directory with the same name. Through the above embodiment, the speech large model is used to realize the efficient conversion of dialogue speech data into a text document. The function of this conversion process is not only to greatly improve the efficiency of requirement analysis, but also to ensure the accuracy and professionalism of the requirement document through intensive training for professional terms, which is a crucial first step in the entire intelligent analysis process of business requirements.

[0058] To ensure the accuracy and reasonableness of requirement information, in an optional implementation manner, the above step S202 includes:

[0059] Step S2021, perform a preliminary verification and a secondary verification on each of the above-mentioned dialogue text documents in sequence to obtain the initial requirements. The above-mentioned preliminary verification is a verification operation to correct misrecognized requirement information. The above-mentioned requirement information includes at least the requirement priority and the detailed requirement description. The above-mentioned secondary verification is to supplement the missing valid requirement content. The above-mentioned valid requirement content includes at least the key requirement content not mentioned during the conversation and the task assignment for developers;

[0060] In step S2022, a preliminary analysis and filtering of the content of the above initial requirements are performed to remove information irrelevant to the above business requirements or duplicate and redundant information, thereby forming the above preliminary requirements document.

[0061] In the above embodiment, for the preliminary verification, natural language processing technologies such as semantic similarity comparison and entity recognition are used to automatically check whether there are common misrecognition phenomena in the document, such as errors in requirement priorities and detailed descriptions. The requirement analysts or domain experts manually check the document, paying particular attention to the coherence, logic, and rationality of the requirement information, correcting possible deviations in model recognition, rectifying any misrecognition of requirement priorities to ensure an accurate reflection of the priorities; at the same time, clarifying the requirements with unclear descriptions to make their expressions more precise. For the secondary verification: key requirement points that are omitted are supplemented by listening to the recording again or referring to the meeting minutes, such as certain technical details or requirements in specific scenarios. The list of developers involved in each requirement point is clarified. If not mentioned during the conversation, the requirement analysts should mark it in the document for subsequent development task allocation. The verification operations ensure the accuracy of the requirement information and reduce misrecognition caused by reasons such as unclear voice and accent differences. The secondary verification supplements important omitted information, such as key requirement points and developer allocation, making the requirement document more complete. The manual verification helps to maintain the consistency of the expression of requirement information and avoid differences in the expression of the same requirement in different documents. As Figure 3 shown, the data completion module performs the verification of the document data, mainly including verifying the extracted information to ensure its accuracy and integrity, and supplementing and improving the extracted information according to the business requirements to ensure that all key requirement points are covered. At the same time, it also provides a function for displaying the requirement analysis results, editing and modifying the requirement content items. For example, if it is found that the requirement priority recognition is incorrect, the correct priority can be manually entered, and the subsystems associated with the requirement, the developers involved, etc. can also be selected from the dropdown. These selected data need to be pre-entered into the data completion module.

[0062] Next, use text analysis techniques to identify the themes of the requirements document, ensuring that each requirement point revolves around the business objectives. Analyze whether there are relationships between requirement points and identify possible dependencies. Remove duplicate requirement descriptions to avoid confusion and duplicate work in subsequent development, and delete content unrelated to the current business requirements, such as chitchat, personal opinions, or non-decision-making comments, to ensure that the document focuses on the business core. Structurally organize the remaining requirement information to form a clear requirement list or matrix for subsequent analysis and task allocation. Filter out irrelevant and redundant information so that the project team can focus on the truly critical requirements, accelerating the requirement analysis and development process. By analyzing, ensure the accuracy and reasonableness of the requirement information, reducing possible errors in subsequent design and development. The structured preliminary requirements document highlights the key points and directions of project development, facilitating all team members to quickly understand the core of the requirements.

[0063] The preliminary analysis and filtering of the content of the above initial requirements are implemented by Figure 3 the information extraction module in it. The information extraction module focuses on the preliminary rough screening and structured processing in the early stage. Its main responsibilities are: identifying key information related to business requirements from multiple conversation documents through natural language processing technology. The information extraction at this stage is relatively extensive, aiming to capture all possible relevant points, and then performing preliminary structured processing on this information to make it more suitable for subsequent refined analysis. Based on the set requirements analysis template, the module will parse the requirements analysis dimensions in the template, such as function description, operation guide, expected results, system integration requirements, etc. Then, the module designs and generates a series of questions, which are designed to guide the large model to focus on information extraction in specific dimensions. The large model answers based on the questions, and the information extraction module sorts out and records these answers in the requirements analysis Excel sheet for further sorting and analysis. Therefore, in the process of generating the above preliminary requirements document, it also includes: inputting all the above conversation text documents into the above large and small language model to identify information from the conversation text documents through the large and small language model, obtaining key information related to the above business requirements, and recording all the key information in the requirements analysis table; integrating the above requirements analysis table with the above initial requirements to form the above preliminary requirements document.

[0064] To reduce the time and labor costs of manually filling in documents, in an optional implementation manner, the above step S203 includes:

[0065] Step S2031, set the filling area for the above specific information to obtain the above set requirements template;

[0066] Step S2032, use the attention mechanism of the above large and small language model to filter redundant information from the above preliminary requirements document to form a requirements analysis text;

[0067] Step S2033: Extract information from the above requirement analysis text according to the preset keywords to filter out the statement information related to the above preset keywords, and obtain the above requirement name, the above detailed requirement description, the above associated system, and the above expected online date in the above specific information;

[0068] Step S2034: Determine the requirement priority in the above specific information according to the keyword frequency. The above keyword frequency includes high-frequency keywords and low-frequency keywords, and the above requirement priority includes high priority and low priority;

[0069] Step S2035: Fill the above specific information into the above filling area corresponding to the above set requirement template to obtain the above requirement document.

[0070] In the above embodiment, the above specific information at least includes a requirement name, a detailed requirement description, a requirement priority, an associated system, and an expected online date. For example Figure 4As shown, create a table or document template that includes keyword fields such as requirement name, detailed requirement description, associated system, expected go-live date, etc. In the template, reserve specific filling areas for each keyword field, such as cells, text boxes, etc., to present requirement information in a structured manner. By setting the template and filling areas, ensure that requirement information is presented in a consistent format and structure, facilitating subsequent analysis, understanding, and tracking. The use of the template helps ensure that all key requirement information is collected and recorded, reducing omissions. The attention mechanism allows large and small language models to focus on the most important parts when processing long texts, thereby identifying and filtering out redundant information in the preliminary requirement document. Based on the overall semantic understanding and the attention mechanism, the large model generates a summary or key point summary of the preliminary requirement document, removing unimportant details. The small model then further analyzes the summary to identify information related to specific requirements, eliminating duplicate or irrelevant content. Filtering redundant information enables the model to understand and process key requirements more quickly. The attention mechanism helps the model focus on the truly important requirement descriptions, improving the accuracy of information extraction. Use natural language processing techniques, such as TF-IDF, part-of-speech tagging, etc., to identify preset keywords in the text. From the summary or refined text, select the sentences that contain the preset keywords, which are directly related to the requirement name, detailed requirement description, associated system, and expected go-live date, etc. The keyword identification and sentence screening process are automated, reducing the workload of manual screening. Ensure that the extracted information is closely related to the key elements of the requirements, improving the practicality of the requirement document. Count the frequency of keyword occurrences. High-frequency keywords may point to more general and important requirements. Set a threshold based on the keyword frequency. Requirements corresponding to keywords above the threshold are marked as high-priority, and vice versa for low-priority. High-priority requirements can be prioritized for resource development, which helps project management. Correlate the obtained specific requirement information with the fields in the requirement template. Use programming scripts or specialized document generation tools to automatically fill the extracted information into the filling areas of the set requirement template to form a complete and structured document. Greatly reduces the time and labor costs of manually filling out the document. Ensures the consistency and standardization of the generated requirement document, facilitating management and sharing.

[0071] It should also be noted that in this embodiment, it is through Figure 3It is implemented by the intelligent extraction module and the intelligent organization module. The intelligent extraction module, based on the work of the information extraction module, conducts deeper and more refined information processing. Its main responsibilities include: The intelligent extraction module receives the results and questions from the information extraction module, and uses the collaborative work of large and small models (the large model provides macroscopic understanding, and the small model supplements details and domain knowledge) to deeply analyze and optimize the initially extracted information, ensuring that each requirement point is accurately described and all necessary details are covered. The intelligent extraction module fills the deeply processed information into the corresponding positions according to the structure of the requirement template to form a requirement document. Here, "information filling" is the last step of structuring on the basis of ensuring information accuracy and integrity, that is, placing key information in the correct positions of the template for easy viewing and understanding by project stakeholders. The intelligent assembly module is responsible for intelligently assembling the extracted information to generate a complete requirement document. On the one hand, it is information integration, integrating the extracted information to ensure the integrity and consistency of the requirement document. On the other hand, it is document generation, that is, generating a complete requirement document to provide a basis for subsequent analysis and processing.

[0072] To improve the extraction and recognition accuracy of high-quality phrases by the intelligent extraction module, based on the above step S2033, the method further includes: obtaining high-quality phrases from the historical knowledge base as the Positive Pool, and other phrases as negative examples. 10% of the high-quality phrases in the Negative Pool are misclassified into the negative examples because they are not in the knowledge base. Then, a random forest ensemble classifier is used to reduce the impact of noise on classification, and through a formulaic method, the recognition accuracy of high-quality phrases is improved. The specific calculation method steps are as follows:

[0073] The first step is data definition. Let the total phrase set be D, where: 1. Positive Pool: The set of high-quality phrases extracted from the historical knowledge base, denoted as: P = {X i |X i is the high-quality phrase annotated by the knowledge base}, i = 1,..., K, where P is the positive pool, and X i is the i-th high-quality phrase, and K is the total number of phrases in the positive pool. 2. Negative Pool: The remaining phrase set, but it contains 10% of the high-quality phrases (noise) not covered by the knowledge base, denoted as: N = D\P, where |N + | = 0.1|N|, N + is the true positive example, and N is the negative pool.

[0074] The second step is random forest modeling. Among them, 1. Base classifier training, for each decision tree s ∈ {1,..., S}: 1). Bootstrap sampling: Sample n samples with replacement from P ∪ N to obtain the training set D s2) Node splitting rule: On a random subset of features (size m), select the split point to minimize the Gini impurity: in, Represents the subset of samples with label y in the node. 2. Ensemble prediction. Random forest aggregates the prediction results of S trees through majority voting:

[0075] Or based on probability averaging: h s (x) represents the classification result of the sth tree after using the CART algorithm.

[0076] The third step is noise robustness analysis. Random Forest reduces the impact of 10% noise in the negative pool through the following mechanisms: 1. Bagging variance reduction: Bootstrap sampling ensures that noise samples only appear in approximately 63.2% of the base classifiers, reducing the impact of a single noise sample. 2. Feature randomness: Each tree uses only m random features, weakening the spurious correlation between noise features and labels. 3. Majority voting correction: Incorrect predictions caused by noise are overwritten by the majority of correct predictions in the voting, and the final classification error is bounded by: Among them, ∈ Tree is the error of a single tree on noisy data.

[0077] The fourth step is to evaluate the classification effect. The accuracy and recall of the final classifier on the test set can be expressed as accuracy: D test Represents the total test set corresponding to the positive example; recall rate (for positive examples): P test represents the total test set corresponding to negative examples.

[0078] In order to quickly screen and understand key demand information, in an optional implementation, after the above step S2034, the method further includes:

[0079] Step S301, calculating the word frequency score corresponding to each word in the above-mentioned conversation text document according to the first formula, the above-mentioned first formula is tfidf(t, d, D)=tf(t, d)×idf(t, D), wherein fidf(t, d, D) represents the above-mentioned word frequency score, tf(t, d) is the word frequency, tf(t, d)=log(1+freq(t, d)), freq(t, d) represents the number of times the candidate word t appears in the current conversation text document d, the above-mentioned candidate word is a word that appears in the above-mentioned conversation text document, idf(t, D)=log(N / count(d∈D:t∈D)) is the inverse document frequency, count(d∈D:t∈D) is the number of the above-mentioned conversation text documents containing the above-mentioned candidate word t, N represents the total number of the above-mentioned conversation text documents, and D represents the set of all the above-mentioned conversation text documents;

[0080] Step S302: Determine the words with the word frequency scores greater than or equal to the set score as high-frequency keywords.

[0081] Step S303: Determine the words with the word frequency scores less than the set score as the low-frequency keywords.

[0082] In the above embodiment, first, a set including all dialogue text documents is established, and this set constitutes the corpus for calculating TF-IDF (Term Frequency-Inverse Document Frequency). Traverse each document, count the number of occurrences of the candidate word t in the document d, denoted as freq(t, d). Calculate the inverse document frequency of the candidate word t in the entire document set, and the formula is idf(t, D) = log(N / count(d ∈ D: t ∈ d)), where N is the total number of documents, and count(d ∈ D: t ∈ d) represents the number of documents in the document set D that contain the candidate word t. For each candidate word t, calculate its word frequency score tfidf(t, d) in each document d, using the formula tfidf(t, d) = tf(t, d) * idf(t, D), where tf(t, d) = log(1 + freq(t, d)), to ensure that common words will not lose their discrimination due to excessive frequency.

[0083] TF-IDF is an effective method for measuring the importance of terms in a document. It combines term frequency (how frequently a term appears in a single document) and inverse document frequency (how rare a term is in the document set), making the importance scoring of each word more accurate. By calculating the TF-IDF value of each word, those words that are frequent in the document and rare in the document set can be identified. These words often carry more valuable information and are the basis for keyword extraction in subsequent steps. Preset a word frequency score threshold T to distinguish important words from general words. Determine all words with word frequency scores greater than or equal to the threshold T as high-frequency keywords. High-frequency keywords reflect the most prominent and representative content in the document. They are usually closely related to the core requirements and are therefore key references when generating the functional requirements specification. By screening high-frequency keywords, the core information in the document can be quickly focused, providing more direct clues for subsequent intelligent analysis. Determine all words with word frequency scores less than the set score T as low-frequency keywords. Although low-frequency keywords are not the main part of the document, they may contain detailed information in specific situations and help to fully understand the business requirements. When generating the functional requirements specification, low-frequency keywords can be used as auxiliary information to help identify the descriptions of certain specific functions or abnormal situations, enhancing the comprehensiveness and accuracy of the document.

[0084] In order to improve the accuracy of generating the functional requirement specification, in an alternative implementation, step S204 described above includes:

[0085] Step S2041, define the document format of the above-mentioned functional requirement specification, and the above-mentioned document format includes at least function points and function descriptions;

[0086] Step S2042, perform semantic analysis on the above-mentioned requirement document using the above-mentioned large model in the above-mentioned large and small language models to determine multiple function points. The above-mentioned function points include at least the triggering conditions, execution processes, expected outputs, exception handling, and mutual dependencies of the functions. The above-mentioned mutual dependencies include the execution order between functions and the priorities between functions;

[0087] Step S2043, perform semantic refinement on each of the above-mentioned function points using the above-mentioned small model in the above-mentioned large and small language models to determine the function descriptions corresponding to each of the above-mentioned function points;

[0088] Step S2044, use the text generation ability of the above-mentioned large model to convert the above-mentioned function points and the above-mentioned function descriptions corresponding to the above-mentioned function points into the form of the above-mentioned document format, and at the same time use the above-mentioned small model to check whether the conversion result meets the document requirements during the conversion process. The above-mentioned document requirements include at least conforming to the document format and the correct use of professional terms;

[0089] Step S2045, when the above-mentioned conversion result meets the above-mentioned document requirements, determine the above-mentioned conversion result as the above-mentioned functional requirement specification;

[0090] Step S2046, when the above-mentioned conversion result does not meet the above-mentioned document requirements, regenerate the above-mentioned functional requirement specification.

[0091] In the above embodiments, the format of the functional requirements specification should include key parts such as function points and function descriptions, e.g., triggering conditions, execution processes, expected outputs, exception handling, and function dependencies. Clearly define the writing requirements for each part, such as the use of professional terms, the level of detail in the description, and format specifications. Ensure that the format of the functional requirements specification is unified to facilitate understanding and execution by project team members. Specify the requirements as function points and clearly describe their triggering conditions, execution processes, etc., to make the requirements clearer. The large model, based on the overall semantics of the requirements document, identifies various descriptions and instructions related to functions. Extract key elements such as the triggering conditions, execution processes, and expected outputs of functions from the requirements document to identify specific function points. Through semantic analysis, identify the execution order and priorities among functions and understand the mutual dependencies of functions. Through the semantic understanding and natural language processing capabilities of the large model, automatically identify function points and reduce manual work. The large model helps to identify all relevant function points, including those implicit and complex dependency relationships. The small model conducts in-depth analysis for each function point, supplementing specific parameters, operation details, etc., to make the function description more complete and accurate. The small model applies domain knowledge to ensure the correct use of professional terms in the function description and improve the professionalism of the document. The refinement work of the small model ensures the accuracy and completeness of the function description and improves the practical value of the specification. Through the domain expertise of the small model, standardize the use of terms and maintain the consistency of the document. The large model automatically generates a functional requirements specification that conforms to the document format based on function points and function descriptions. During the generation process, the small model checks in real time whether the generated text meets the document requirements, including format correctness, term accuracy, etc., to ensure the quality of the document. The large and small models collaborate to automatically complete the document writing, greatly improving efficiency. The check by the small model ensures that the document format is correct and the term usage is standardized, improving the professionalism and readability of the document. When the document jointly generated by the large and small models meets the preset document requirements, confirm that the document is the final functional requirements specification. Document correction: If all document requirements are not met during the document generation process, such as format errors or improper term usage, then initiate the correction process and regenerate until the requirements are met. Through repeated confirmation and correction, ensure that the quality of the functional requirements specification reaches the expected standard. Allow iterations during the document generation process until the document fully meets the requirements, improving the flexibility and accuracy of document generation. Through the collaborative work of the large and small language models, the generation process from the requirements document to the functional requirements specification is automated and standardized, not only improving the efficiency of document generation, but also ensuring the comprehensiveness, accuracy, and professionalism of the document, providing precise and clear requirement guidance for software development projects, reducing misunderstandings and rework during the development process, and promoting the efficient progress of the project.

[0092] It should be noted that the Functional Requirements Specification (FRS) is one of the crucial documents in a software development project, which is used to describe in detail the functional characteristics that the software or system should possess, ensuring that developers, testers, and other project stakeholders have a common understanding of the project requirements. The document format of a typical functional requirements specification may include the following parts: Cover and Title Page: Include information such as project name, document title, version number, author, date, approver, etc. Revision History: Record the date, version, modifier, and an overview of the modified content for each modification of the document, facilitating the tracking of document changes. Declaration: Include content such as copyright information, confidentiality statement, disclaimer, etc., to ensure the legal use and protection of the document. Table of Contents: An automatically or manually generated list of document chapters for convenient reader navigation. Introduction: A brief description of the project background, objectives, scope, and the purpose and goals of the document. System Environment and Dependencies: Describe the basic hardware and software environments required for the system or software to run, as well as the dependencies on other systems or services. User Roles and Permissions: Define the types of users in the system, the roles and permissions of each user, and the relationships between users. List of Functional Requirements: List in detail all the functional requirements of the system or software. Each requirement item should include number, title, description, input, output, triggering conditions, preconditions, postconditions, exception handling, priority, etc. Data Requirements: Describe the data types, data structures, data flows, and data storage requirements processed by the system. Business Rules and Workflows: List the business rules that support the functional requirements and describe the business processes of the system. Interface and Interaction Requirements: Describe the design requirements of the user interface, including screen layout, control styles, response time, interaction logic, etc. Performance and Non-functional Requirements: Include detailed descriptions of non-functional requirements such as performance metrics, security, reliability, compatibility, maintainability, etc. Compatibility and Internationalization Requirements: Describe the compatibility of the system with different operating systems, browsers, devices, and the support requirements for multiple languages and regions. Data Dictionary: List the definitions, data types, data lengths, default values, validity ranges, etc. of all important data items. Glossary: Define all the terms and abbreviations used in the document to ensure the consistency of terms and the accuracy of understanding. References and Appendices: List the cited documents, standards, protocols, etc., as well as additional tables, charts, design sketches, etc. Signature Page: After the document is completed, it needs to be signed and confirmed by all relevant project stakeholders, including the author, reviewers, project manager, and customer representative, to formally approve the content of the document.

[0093] To enhance the efficiency and accuracy of project management, in an optional implementation, after the above step S204, the method further includes:

[0094] Step S401: Determine various dimensional information for requirement transposition. The above dimensional information includes at least a list of personnel, system components, business process stages, requirement priorities, development stages, requirement function points, and requirement progress.

[0095] Step S402: Automatically identify and extract the field information and requirement descriptions of each of the above dimensional information in the above requirement document.

[0096] Step S403: Determine the corresponding transposition rules according to each of the above dimensional information, perform transposition operations according to the above transposition rules, and form a list of function items for different above dimensional information. The above transposition rules are rules that will be displayed in different forms when different above dimensional information is selected. The above transposition operation is to automatically display the above field information and the above requirement description corresponding to the above dimensional information in the corresponding predetermined form when the corresponding above dimensional information is selected. The above predetermined form includes at least a list form, a Gantt chart form, and a table form.

[0097] In the above embodiment, as Figure 5As shown, the system administrator or requirements analyst needs to define transposable dimension information, including personnel lists, system components, business process stages, requirement priorities, development stages, requirement function points, and requirement progress, etc. These dimension information reflect the considerations from different perspectives in the requirements document. In the system configuration interface, set these dimension information as transposable options for subsequent operation selection. The definition of dimension information needs to consider the needs of project management to ensure that each dimension has practical application value. The defined dimension information provides the possibility for project stakeholders to view requirements from multiple perspectives, making the requirements analysis more comprehensive. Different dimensions can be selected for transposition operations according to the needs of different stages and roles, enhancing the flexibility of system use. Using natural language processing technologies such as entity recognition and syntactic analysis, automatically identify the relevant fields of each dimension information in the requirements document. For example, identify the part about "personnel list" in the requirement description. Extract the specific requirement descriptions related to each dimension information to ensure that the field information corresponding to each dimension has detailed requirement backgrounds and contexts. Structurally process the extracted field information and requirement descriptions and store them in a specific format for subsequent transposition operations. Automatically identifying and extracting dimension information reduces manual operations and improves efficiency. Information association: Associate the field information with the specific requirement description to ensure that the details of the requirements can be accurately displayed during the transposition operation. Set transposition rules for each dimension information, specifying how the field information and requirement description should be displayed when this dimension is selected. For example, when "personnel list" is selected, display the task list assigned to each developer; when "requirement progress" is selected, display the completion status of each requirement in the form of a Gantt chart. When the user selects a certain dimension information, the system automatically performs the transposition operation according to the preset transposition rules and displays the function item list or view in a predefined form. The transposition rules enable the requirement information to be displayed according to the preferences of individuals or teams, improving the readability and comprehensibility of the information. Display the requirement progress or task assignment in the form of a Gantt chart, table, etc., making the information intuitive and easy to understand. The dimension transposition function facilitates project managers and developers to track the implementation of requirements and ensure that the project progresses as planned. This embodiment is implemented through Figure 3 the intelligent transposition module in Figure 3 . The intelligent transposition module provides a requirement transposition operation process. Project stakeholders can select different views and can enter certain query filtering conditions for requirement tracking and viewing. Project managers can select a Gantt chart to view the requirement progress. Developers can select the employee list as the transposition condition and filter the employee names specified by themselves for requirement development function tracking and viewing.

[0098] Through the above embodiments, the intelligent transposition of the requirement list can be performed, and providing different perspectives to view the requirement view is an important part of improving the efficiency of business requirement development. It can not only provide various requirement display methods, but also reduce the progress risk and requirement omission risk of project development to a certain extent. The intelligent transposition process not only realizes the multi-dimensional viewing of requirement document information, but also provides an intuitive and customized display method, greatly enhancing the efficiency and accuracy of project management, enabling team members to quickly understand the requirements from their respective concerned perspectives, and promoting the smooth execution of the project. It can also improve the efficiency of software project development requirement analysis for the business, track the implementation of business projects, understand the project progress risk, and promote the project to be delivered as soon as possible.

[0099] It should also be noted that after determining the corresponding transposition rules according to the above-mentioned dimension information, performing the transposition operation according to the above-mentioned transposition rules, and forming a list of function items with different above-mentioned dimension information, the above method further includes: analyzing the functional requirement specification, and determining each of the above-mentioned function points or the requirement function items in the above-mentioned list of function items as a high-level task, where one function point or one requirement function item corresponds to one of the above-mentioned high-level tasks; according to the complexity and execution details of the above-mentioned function points and the above-mentioned requirement function items, dividing the above-mentioned high-level tasks into multiple subtasks, and each of the above-mentioned subtasks includes at least execution steps, expected results, and required resources; based on the requirement priority and functional dependency relationship, assigning priorities to each of the above-mentioned subtasks to obtain the corresponding subtask priorities; determining the task execution order according to all the above-mentioned subtask priorities, and determining the corresponding developers according to the task nature of each of the above-mentioned subtasks, where the task nature includes at least the technical difficulty and required professional skills of the subtask; generating a task list according to the above-mentioned task execution order, the above-mentioned subtask priorities, and the developers of each of the above-mentioned subtasks, so as to perform development according to the above-mentioned task list. Figure 3The task decomposition module in is responsible for decomposing the function item list into specific tasks. It mainly focuses on task decomposition, breaking down the function item list into specific and executable tasks. Then, based on the nature and priority of the tasks, task allocation is carried out to provide clear guidance for subsequent execution. Specifically: The task decomposition module first analyzes the functional requirement specification, regarding each function point or required function item as a high-level task. Then, according to the complexity and execution details of the function points, the module further breaks down the high-level tasks into a series of subtasks. Each subtask should include information such as specific execution steps, expected results, and required resources to ensure the executability and clarity of the subtasks. After that, based on the requirement priority and functional dependency relationship, the module assigns priorities to each subtask. Tasks with high priority or on the critical path are arranged first to ensure the smooth progress of the project as planned. Subsequently, according to the nature of the subtasks, such as technical difficulty and required professional skills, the module assigns tasks to appropriate team members and considers resource availability to ensure the feasibility of task execution. Finally, a task list containing all subtasks, priorities, assigned personnel, and resources is generated, which serves as a direct guide for project execution. During the project execution process, the task decomposition module is also responsible for monitoring the task completion status, and timely adjusting the task priorities and assignments according to the project progress and resource changes to ensure the achievement of project goals. By transforming functional requirements into specific and executable tasks, the work content and responsibilities are clarified, the uncertainty during the execution process is reduced, and the project progress is accelerated. Through the reasonable allocation of personnel and resources according to the nature and priority of the tasks, resource waste is avoided, and the work efficiency and satisfaction of team members are improved. Task decomposition ensures that each function point has a clear execution plan, reducing the risk of project out-of-control caused by unclear requirements or unreasonable task division.

[0100] This embodiment relates to a specific business requirement analysis method, such as Figure 6 shown, which includes the following steps:

[0101] Step1: First, obtain the demand dialogue voice data, use multimedia devices for audio collection, and record the communication content during the meeting discussion between the demand analysts and business requirement personnel;

[0102] Step2: Use the speech large model to identify the speech content and generate a dialogue text document. Translate the collected audio data through the speech large model to automatically generate a text document in the form of dialogue content;

[0103] Step3: Manually verify, correct the mis-identified requirement information, such as incorrect identification of requirement priority, poor detailed description, etc., conduct secondary verification, supplement the requirement content that has not been effectively included, and supplement and proofread the key content not mentioned during the dialogue process, developer task assignment, etc.;

[0104] Step 4: Conduct a preliminary analysis and filtering of the requirements document and requirement content for the formed requirements to remove irrelevant or redundant information, and form a more refined and focused preliminary requirements document.

[0105] Step 5: According to the specified requirements template, use large and small language models to extract effective information, and fill specific information into the corresponding positions of the template. Utilize the semantic understanding and context awareness capabilities of the large model to better understand complex concepts and implicit meanings in the text, extract high-level concepts and information from the preliminary requirements document, identify key requirement points and their relationships, and use the information integration and logical reasoning of the large model to comprehensively analyze different requirement descriptions, and infer necessary requirement information that may not be directly expressed through logical reasoning. At the same time, cooperate with the small model's professional knowledge and detail processing capabilities in a specific field to fill in the details of the information extracted by the large model, ensure that each requirement point is accurately described, and use the refined information extraction ability of the small model to help identify specific terms, functional requirements and other detailed information, and accurately fill this information into the corresponding positions of the requirements template;

[0106] Step 6: According to the business requirements, use the cooperation of large and small models to complete the function point analysis of the filled business requirements document, and generate a functional requirements specification;

[0107] Step 7: Through an intelligent transposition process, generate a list of functional requirements in different dimensions, which is convenient for different project stakeholders to review and understand the business requirements from multiple perspectives, and ensure that all key function points are fully considered and covered.

[0108] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0109] The embodiment of the present application also provides a business requirements analysis device. It should be noted that the business requirements analysis device of the embodiment of the present application can be used to execute the business requirements analysis method provided by the embodiment of the present application. The device is used to implement the above embodiments and preferred implementation manners, and those that have been described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that realizes a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0110] The following introduces the business requirements analysis device provided by the embodiment of the present application.

[0111] Figure 7It is a structural block diagram of a business requirement analysis device according to an embodiment of the present application. As Figure 7 shown, the device includes:

[0112] An acquisition unit 10, configured to acquire all conversation voice data of business requirement discussions, and convert all the above-mentioned conversation voice data into corresponding conversation text documents.

[0113] Specifically, a recording device or software is used to capture the voice communications of all participants during the requirement discussion meeting to obtain all conversation voice data. These meetings can be face-to-face meetings, telephone conferences or video conferences to ensure that comprehensive requirement discussion content is collected.

[0114] After acquiring all conversation voice data of business requirement discussions and before converting all the above-mentioned conversation voice data into corresponding conversation text documents, the collected audio files are preprocessed, including noise cancellation, format standardization (such as converting to WAV or MP3 format), and segmented into small segments suitable for processing to facilitate subsequent speech recognition. Speech-to-text: A speech large model (such as an optimized model based on Whisper) is used to perform speech recognition on each preprocessed audio segment and transcribe it into a text document. Automatically converting oral discussions into written documents saves a large amount of time and effort in manual recording. The text document format facilitates further information extraction and analysis using natural language processing techniques.

[0115] A first generation unit 20, configured to generate a preliminary requirement document based on all the above-mentioned conversation text documents.

[0116] Specifically, all conversation text documents are aggregated together, and preliminary merging and sorting are performed to construct a document containing all discussion points. Through text analysis, non-critical information and duplicate content are removed, and core requirements and discussion points are retained. The screened document is preliminarily structured, classified and labeled to form a rough requirement document framework, including requirement titles, descriptions, requirement priorities and requirement categories. It provides a starting point and a basic framework for subsequent detailed requirement analysis. Aggregating scattered conversation information into relevant requirement descriptions facilitates understanding and further analysis.

[0117] An extraction and filling unit 30 is used to extract information from the above-mentioned preliminary requirements document using a large and small language model according to a set requirements template, obtain specific information, and fill the specific information into the corresponding positions of the set requirements template to obtain a requirements document. The large and small language model is a model that uses a large model and a small model to work together. The large model is used for the overall semantic understanding of the preliminary requirements document, and the small model is used for precise analysis of the preliminary requirements document to assist the large model in filling in the missing parts in semantic understanding. The large model is trained to understand the overall semantics, and the small model is fine-tuned through a dataset in a specific domain to accurately identify specific business vocabulary and terms. Based on the set requirements template, a series of questions are designed, and the large model is used to conduct a high-level analysis of the preliminary requirements document, while the small model answers specific questions in detail, such as identifying specific functions, parameter settings, business processes, etc. The answers provided by the large model and the small model are integrated and filled into the corresponding positions according to the structure of the requirements template, such as function description, operation guide, expected results, etc. On the basis of information filling, the document is improved, including format adjustment, consistency check, and necessary supplementary explanations. The collaborative work of the large and small models ensures the accuracy of professional terms and the comprehensiveness of detailed descriptions in the requirements document. The use of the template ensures the unity and comparability of the document, facilitating communication and understanding among team members.

[0118] A second generation unit 40 is used to analyze the function points of the requirements document using the above-mentioned large and small language model, generate a functional requirements specification, and conduct business development based on the functional requirements specification, so that the developed product has the function points of the functional requirements specification.

[0119] Specifically, based on the overall semantic understanding, the large model identifies the function points and business processes in the document, while the small model deeply analyzes the specific requirements of each function point. The identified and analyzed function points are formatted into the standard format of the functional requirements specification, including function description, input / output definition, operation process, expected behavior, exception handling, etc. Through multiple rounds of question and answer, the description of the function points is continuously refined, including specific data fields, user interface elements, algorithm logic, etc. Finally, the document can be reviewed by requirements analysts or domain experts to ensure that all key function points are covered and the description is accurate and error-free. The functional requirements specification provides a clear functional development guide for the development team, reducing uncertainty and rework in the development process. The detailed function point analysis enables developers to directly code specific functions without the need for further requirements clarification. The whole process realizes the automatic understanding and documentation of business requirements through the collaborative work of the large and small language models, not only improving the efficiency and quality of requirements analysis, but also providing a solid foundation for subsequent system development, ensuring a high degree of consistency between the developed product and the expected functional requirements.

[0120] In this embodiment, an acquisition unit is configured to acquire all the dialogue voice data of business requirement discussions and convert all the above-mentioned dialogue voice data into corresponding dialogue text documents; a first generation unit is configured to generate a preliminary requirement document based on all the above-mentioned dialogue text documents; an extraction and filling unit is configured to extract specific information from the above-mentioned preliminary requirement document by using a large and small language model according to a set requirement template, and fill the above-mentioned specific information into the corresponding positions of the above-mentioned set requirement template to obtain a requirement document. The above-mentioned large and small language model is a model that uses a large model and a small model to work together. The above-mentioned large model is used for the overall semantic understanding of the above-mentioned preliminary requirement document, and the above-mentioned small model is used for precise analysis of the above-mentioned preliminary requirement document to assist the above-mentioned large model in filling the missing semantic understanding; a second generation unit is configured to use the above-mentioned large and small language models to perform function point analysis on the above-mentioned requirement document to generate a functional requirement specification, so as to perform business development according to the above-mentioned functional requirement specification, so that the developed product has the function points of the above-mentioned functional requirement specification. This application acquires all the dialogue voice data of business requirements, identifies the voice content and converts it into corresponding text documents, and generates a preliminary requirement document based on the text documents for requirement analysis; according to the specified requirement template, uses a large and small language model to extract effective information, and fills the extracted specified information into the corresponding positions of the template; according to the business requirements, uses the large and small language models to cooperate to complete the function point analysis of the filled business requirement document, generates a functional requirement specification, and provides guidance for system design and development according to the generated detailed functional requirement specification. There is no need for developers to collect, summarize, extract key points, inductively analyze, and check the content of a large number of multi-source heterogeneous original documents to obtain a functional requirement specification, which can improve the efficiency of requirement analysis for software project development in the banking industry and greatly reduce the time cost of developers. This application solves the problem that a large amount of time cost is consumed in the prior art for collecting, summarizing, extracting key points, inductively analyzing, and checking the content of a large number of multi-source heterogeneous original documents.

[0121] In an optional implementation manner, to achieve the efficient conversion of dialogue voice data into text documents and improve the efficiency of requirement analysis, the above-mentioned acquisition unit includes:

[0122] A voice conversion module is configured to use a voice large model to perform text translation on the above-mentioned dialogue voice data to generate the corresponding above-mentioned dialogue text document. The above-mentioned voice large model is a convergent model obtained by iteratively training a predetermined model with multiple groups of sample data. Each group of the above-mentioned sample data includes sample voice data and the sample text data corresponding to the above-mentioned sample voice data.

[0123] In an optional implementation manner, to ensure the accuracy and reasonableness of requirement information, the above-mentioned first generation unit includes:

[0124] A verification module is used to perform preliminary verification and secondary verification on each of the above-mentioned dialogue text documents in sequence to obtain initial requirements. The above-mentioned preliminary verification is a verification operation to correct mis-identified requirement information. The above-mentioned requirement information includes at least requirement priority and detailed requirement description. The above-mentioned secondary verification is to supplement missing valid requirement content. The above-mentioned valid requirement content includes at least key requirement content not mentioned during the dialogue process and developer task allocation;

[0125] A first filtering module is used to perform preliminary analysis and filtering of requirement content on the above-mentioned initial requirements to remove information irrelevant to the above-mentioned business requirements or duplicate and redundant information, and form the above-mentioned preliminary requirement document.

[0126] In the above-mentioned embodiment, during the process of generating the above-mentioned preliminary requirement document, it further includes: an input module, which is used to input all the above-mentioned dialogue text documents into the above-mentioned large and small language model to identify information in the above-mentioned dialogue text documents through the large and small language model, obtain key information related to the above-mentioned business requirements, and record all the key information in a requirement analysis table; an integration module, which is used to integrate the above-mentioned requirement analysis table with the above-mentioned initial requirements to form the above-mentioned preliminary requirement document.

[0127] In order to reduce the time and labor costs of manually filling in documents, in an alternative implementation manner, the above-mentioned extraction and filling unit includes:

[0128] A setting module is used to set the filling area of the above-mentioned specific information to obtain the above-mentioned set requirement template;

[0129] A second filtering module is used to filter redundant information from the above-mentioned preliminary requirement document by using the attention mechanism of the above-mentioned large and small language model to form a requirement analysis text;

[0130] An extraction module is used to extract information from the above-mentioned requirement analysis text according to preset keywords to screen out statement information related to the above-mentioned preset keywords, and obtain the above-mentioned requirement name, the above-mentioned detailed requirement description, the above-mentioned associated system, and the above-mentioned expected online date in the above-mentioned specific information;

[0131] A first determination module is used to determine the requirement priority in the above-mentioned specific information according to keyword frequency. The above-mentioned keyword frequency includes high-frequency keywords and low-frequency keywords. The above-mentioned requirement priority includes high priority and low priority;

[0132] A filling module is used to fill the above-mentioned specific information into the corresponding filling area of the above-mentioned set requirement template to obtain the above-mentioned requirement document.

[0133] In the above embodiments, the specific information at least includes the requirement name, requirement detailed description, requirement priority, associated system, and expected online date. As Figure 4 shown, create a table or document template that includes keyword fields such as requirement name, requirement detailed description, associated system, and expected online date. In the template, reserve specific filling areas for each keyword field, such as cells, text boxes, etc., to present the requirement information in a structured manner. By setting the template and filling areas, ensure that the requirement information is presented in a consistent format and structure, facilitating subsequent analysis, understanding, and tracking. The use of the template helps ensure that all key requirement information is collected and recorded, reducing omissions. The attention mechanism allows large and small language models to focus on the most important parts when processing long texts, thereby identifying and filtering out redundant information in the preliminary requirement document. Based on the overall semantic understanding and attention mechanism, the large model generates a summary or key point summary of the preliminary requirement document, removing irrelevant details. The small model then further analyzes the summary to identify information related to specific requirements, eliminating duplicate or irrelevant content. Filtering redundant information enables the model to understand and process key requirements more quickly. The attention mechanism helps the model focus on the truly important requirement descriptions, improving the accuracy of information extraction. Use natural language processing techniques, such as TF-IDF, part-of-speech tagging, etc., to identify preset keywords in the text. From the summary or refined text, select the sentences that contain the preset keywords, which are directly related to the requirement name, requirement detailed description, associated system, and expected online date. The keyword identification and sentence screening process are automated, reducing the workload of manual screening. Ensure that the extracted information is closely related to the key elements of the requirements, improving the practicality of the requirement document. Count the frequency of keyword occurrences. High-frequency keywords may point to more general and important requirements. Set a threshold based on the keyword frequency. Requirements corresponding to keywords above the threshold are marked as high priority, and vice versa for low priority. High-priority requirements can be prioritized for resource development, which helps project management. Correlate the obtained specific requirement information with the fields in the requirement template. Use programming scripts or specialized document generation tools to automatically fill the extracted information into the filling areas of the set requirement template to form a complete and structured document. Greatly reduces the time and labor costs of manually filling out the document. Ensures the consistency and standardization of the generated requirement document, facilitating management and sharing.

[0134] To quickly screen and understand key requirement information, in an alternative embodiment, the device further includes:

[0135] A calculation unit is configured to calculate the word frequency score corresponding to each word in the dialogue text document according to a first formula after determining the requirement priority in the specific information based on the keyword frequency. The first formula is tfidf(t, d, D) = tf(t, d) × idf(t, D), where tfidf(t, d, D) represents the word frequency score, tf(t, d) is the term frequency, tf(t, d) = log(1 + freq(t, d)), freq(t, d) represents the number of times the candidate word t appears in the current dialogue text document d, the candidate word is the word that appears in the dialogue text document, idf(t, D) = log(N / count(d ∈ D: t ∈ D)) is the inverse document frequency, count(d ∈ D: t ∈ D) is the number of the dialogue text documents containing the candidate word t, N represents the total number of the dialogue text documents, and D represents the set of all the dialogue text documents;

[0136] A first determination unit is configured to determine the words with the word frequency score greater than or equal to a set score as high-frequency keywords;

[0137] A second determination unit is configured to determine the words with the word frequency score less than the set score as the low-frequency keywords.

[0138] In the above embodiment, first, a set including all dialogue text documents is established, and this set constitutes the corpus for calculating TF-IDF (Term Frequency-Inverse Document Frequency). Each document is traversed, and the number of times the candidate word t appears in the document d is counted, denoted as freq(t, d). The inverse document frequency of the candidate word t in the entire document set is calculated, and the formula is idf(t, D) = log(N / count(d ∈ D: t ∈ d)), where N is the total number of documents, and count(d ∈ D: t ∈ d) represents the number of documents containing the candidate word t in the document set D. For each candidate word t, its word frequency score tfidf(t, d) is calculated in each document d using the formula tfidf(t, d) = tf(t, d) * idf(t, D), where tf(t, d) = log(1 + freq(t, d)), ensuring that common words do not lose their distinctiveness due to excessive frequency. Through the above embodiment, high-frequency and low-frequency keywords can be effectively identified from the dialogue text document, providing strong support for subsequent intelligent requirement analysis. High-frequency keywords can capture the core of the document, while low-frequency keywords can provide additional details, making the finally generated requirement document highlight the key points and also take into account comprehensiveness. This method is particularly effective when dealing with a large amount of text data, helping to quickly screen and understand key requirement information, thereby improving the efficiency and accuracy of business requirement analysis.

[0139] To improve the accuracy of generating the functional requirements specification, in an optional implementation, the above-mentioned second generation unit includes:

[0140] A definition module for defining the document format of the above-mentioned functional requirements specification, where the document format at least includes function points and function descriptions;

[0141] A semantic analysis module for performing semantic analysis on the above-mentioned requirements document using the above-mentioned large model in the above-mentioned large and small language models to determine multiple function points. The function points at least include the trigger conditions, execution processes, expected outputs, exception handling, and mutual dependencies of the functions. The mutual dependencies include the execution order between functions and the priorities between functions;

[0142] A semantic refinement module for refining the semantics of each of the above-mentioned function points using the above-mentioned small model in the above-mentioned large and small language models to determine the function descriptions corresponding to each of the above-mentioned function points;

[0143] A conversion module for converting the above-mentioned function points and the above-mentioned function descriptions corresponding to the function points into the form of the above-mentioned document format using the text generation ability of the above-mentioned large model, and at the same time using the above-mentioned small model to check whether the conversion result meets the document requirements during the conversion process. The document requirements at least include conforming to the document format and the correct use of professional terms;

[0144] A second determination module for determining the above-mentioned conversion result as the above-mentioned functional requirements specification when the above-mentioned conversion result meets the above-mentioned document requirements;

[0145] A regeneration module for regenerating the above-mentioned functional requirements specification when the above-mentioned conversion result does not meet the above-mentioned document requirements.

[0146] In the above embodiments, the format of the functional requirements specification should include key parts such as function points and function descriptions, such as trigger conditions, execution processes, expected outputs, exception handling, and function dependencies. Clearly define the writing requirements for each part, such as the use of professional terms, the level of detail in the description, and format specifications. Ensure that the format of the functional requirements specification is unified to facilitate understanding and execution by project team members. Specify the requirements as function points and clearly describe their trigger conditions, execution processes, etc., to make the requirements clearer. The large model, based on the overall semantics of the requirements document, identifies various descriptions and instructions related to functions. Extract key elements such as the trigger conditions, execution processes, and expected outputs of functions from the requirements document and identify specific function points. Through semantic analysis, identify the execution order and priority between functions and understand the mutual dependencies of functions. Using the semantic understanding and natural language processing capabilities of the large model, automatically identify function points and reduce manual work. The large model helps to identify all relevant function points, including those implicit and complex dependency relationships. The small model conducts in-depth analysis for each function point, supplements specific parameters, operation details, etc., to make the function description more complete and accurate. The small model applies domain knowledge to ensure the correct use of professional terms in the function description and improve the professionalism of the document. The refinement work of the small model ensures the accuracy and completeness of the function description and improves the practical value of the specification. Through the domain expertise of the small model, standardize the use of terms and maintain the consistency of the document. The large model automatically generates a functional requirements specification that conforms to the document format based on function points and function descriptions. During the generation process, the small model checks in real time whether the generated text meets the document requirements, including format correctness, term accuracy, etc., to ensure the quality of the document. The large and small models collaborate to automatically complete the document writing, greatly improving efficiency. The check by the small model ensures that the document format is correct and the terms are used standardly, improving the professionalism and readability of the document. When the document jointly generated by the large and small models meets the preset document requirements, confirm the document as the final functional requirements specification. Document correction: If all document requirements are not met during the document generation process, such as format errors or improper use of terms, then initiate the correction process and regenerate until the requirements are met. Through repeated confirmation and correction, ensure that the quality of the functional requirements specification reaches the expected standard. Allow iteration during the document generation process until the document fully meets the requirements, improving the flexibility and accuracy of document generation. Through the collaborative work of the large and small language models, the generation process from the requirements document to the functional requirements specification is automated and standardized, not only improving the efficiency of document generation, but also ensuring the comprehensiveness, accuracy, and professionalism of the document, providing precise and clear requirements guidance for software development projects, reducing misunderstandings and rework during the development process, and promoting the efficient progress of the project.

[0147] To enhance the efficiency and accuracy of project management, in an optional implementation, the device further includes:

[0148] A third determination unit, configured to determine various dimension information of requirement transposition after generating a functional requirement specification by performing functional point analysis on the requirement document using the above large language model, where the dimension information at least includes a personnel list, system components, business process stages, requirement priorities, development stages, requirement function points, and requirement progress;

[0149] An identification and extraction unit, configured to automatically identify and extract field information and requirement descriptions of each of the above dimension information in the above requirement document;

[0150] A fourth determination unit, configured to determine corresponding transposition rules according to each of the above dimension information, so as to perform a transposition operation according to the transposition rules and form a list of function items for different above dimension information. The transposition rules are rules that will be displayed in different forms when different above dimension information is selected. The transposition operation is to automatically display the above field information and the above requirement description corresponding to the above dimension information in a corresponding predetermined form when the corresponding above dimension information is selected. The predetermined form at least includes a list form, a Gantt chart form, and a table form.

[0151] In the above embodiment, as Figure 5 shown, the system administrator or requirement analyst needs to define the dimension information that can be transposed, including a personnel list, system components, business process stages, requirement priorities, development stages, requirement function points, requirement progress, etc. These dimension information reflect considerations from different perspectives in the requirement document. In the configuration interface of the system, these dimension information are set as transposition options for subsequent operation selection. The definition of dimension information needs to consider the needs of project management to ensure that each dimension has practical application value. The defined dimension information provides the possibility for project stakeholders to view requirements from multiple perspectives, making the requirement analysis more comprehensive. Different dimensions can be selected for transposition operations according to the needs of different stages and roles, enhancing the flexibility of system use. Using natural language processing technologies, such as entity recognition and syntactic analysis, automatically identify the relevant fields of each dimension information in the requirement document. For example, identify the part about the "personnel list" in the requirement description. Extract the specific requirement descriptions related to each dimension information to ensure that the field information corresponding to each dimension has a detailed requirement background and context. Structurally process the extracted field information and requirement descriptions and store them in a specific format for subsequent transposition operations. Automatically identifying and extracting dimension information reduces manual operations and improves efficiency. Information association: Associate the field information with the specific requirement description to ensure that the details of the requirement can be accurately displayed during the transposition operation. Set transposition rules for each dimension information, specifying how the field information and requirement description should be displayed when this dimension is selected. This embodiment is through Figure 3It is implemented by the intelligent transposition module. The intelligent transposition module provides an operation process for requirement transposition. Project stakeholders can select views of different dimensions and can enter certain query filtering conditions to view requirement tracking. Project managers can select the Gantt chart to view the requirement progress. Developers can select the employee list as the transposition condition and filter the names of employees specified by themselves to track and view the requirement development function. Through the above embodiments, the requirement list can be intelligently transposed, and providing different perspectives to view the requirement view is an important part of improving the efficiency of business requirement development. It can not only provide diverse requirement display methods, but also reduce the progress risk and requirement omission risk of project development to a certain extent. The intelligent transposition process not only realizes the multi-dimensional view of requirement document information, but also provides an intuitive and customized display method, greatly enhancing the efficiency and accuracy of project management, enabling team members to quickly understand the requirements from the perspectives they care about, and promoting the smooth execution of the project. It can also improve the efficiency of software project development requirement analysis of the business, track the implementation of business projects, understand the project progress risk, and promote the project to be delivered as soon as possible.

[0152] It should also be noted that the above device further includes: an analysis unit, which is used to determine the corresponding transposition rules according to the above dimension information, perform transposition operations according to the above transposition rules, and after forming a list of function items of different above dimension information, analyze the functional requirement specification, and determine each of the above function points or the requirement function items in the above function item list as a high-level task, where one function point or one requirement function item corresponds to one of the above high-level tasks; a task division unit, which is used to divide the above high-level tasks into multiple subtasks according to the complexity and execution details of the above function points and the above requirement function items, and each of the above subtasks includes at least execution steps, expected results, and required resources; an allocation unit, which is used to assign priorities to each of the above subtasks based on requirement priorities and functional dependencies to obtain corresponding subtask priorities; a determination unit, which is used to determine the task execution order according to all the above subtask priorities, and determine the corresponding developers according to the task natures of each of the above subtasks, where the task natures include at least the technical difficulty and required professional skills of the subtasks; a third generation unit, which is used to generate a task list according to the above task execution order, the above subtask priorities, and the developers of each of the above subtasks, and perform development according to the above task list. Figure 3The task decomposition module in is responsible for decomposing the function item list into specific tasks. Mainly for task decomposition, it decomposes the function item list into specific and executable tasks. Then, according to the nature and priority of the tasks, it conducts task allocation to provide clear guidance for subsequent execution. Specifically: The task decomposition module first analyzes the functional requirement specification, regarding each function point or required function item as a high-level task. Then, based on the complexity and execution details of the function points, the module breaks down the high-level tasks into a series of subtasks. Each subtask should include information such as specific execution steps, expected results, and required resources to ensure the executability and clarity of the subtasks. After that, based on the requirement priority and functional dependency relationships, the module assigns priorities to each subtask. Tasks with high priorities or on the critical path are arranged first to ensure the smooth progress of the project as planned. Subsequently, according to the nature of the subtasks, such as technical difficulty and required professional skills, the module assigns the tasks to appropriate team members and considers resource availability to ensure the feasibility of task execution. Finally, it generates a task list containing all subtasks, priorities, assigned personnel, and resources, which serves as the direct guidance for project execution. During the project execution process, the task decomposition module is also responsible for monitoring the task completion status. According to the project progress and resource changes, it adjusts the task priorities and assignments in a timely manner to ensure the achievement of project goals. By transforming the functional requirements into specific and executable tasks, it clarifies the work content and responsibilities, reduces the uncertainty during the execution process, and speeds up the project progress. Through the reasonable allocation of personnel and resources according to the nature and priority of the tasks, it avoids resource waste and improves the work efficiency and satisfaction of team members. Task decomposition ensures that each function point has a clear execution plan, reducing the risk of project out-of-control caused by unclear requirements or unreasonable task division.

[0153] The above-mentioned business requirement analysis device includes a processor and a memory. The above-mentioned acquisition unit, first generation unit, extraction and filling unit, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to implement corresponding functions. The above-mentioned modules are all located in the same processor; or, the above-mentioned each module is separately located in different processors in any combination form.

[0154] The processor contains a kernel, and the kernel retrieves the corresponding program units from the memory. One or more kernels can be set, and by adjusting the kernel parameters, it can solve the problem that the work of collecting and summarizing, extracting key points, inductively analyzing, and content checking of a large amount of multi-source heterogeneous original documents in the prior art consumes a large amount of time cost.

[0155] The memory may include non-permanent memory in computer-readable media, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM), and the memory includes at least one storage chip.

[0156] An embodiment of the present invention provides a computer-readable storage medium. The computer-readable storage medium includes a stored program. When the program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned business requirement analysis method.

[0157] An embodiment of the present invention provides a processor. The processor is used to run a program. When the program runs, it executes the above-mentioned business requirement analysis method.

[0158] An embodiment of the present invention provides a business requirement analysis system. The business requirement analysis system includes a processor, a memory, and a program stored on the memory and executable on the processor. When the processor executes the program, it implements at least the steps of the business requirement analysis method.

[0159] This application also provides a computer program product. When executed on a data processing device, it is adapted to execute a program initialized with at least the steps of the business requirement analysis method:

[0160] Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. They can be implemented by program codes executable by the computing device. Thus, they can be stored in a storage device and executed by the computing device. And in some cases, the steps shown or described can be executed in a different order from here, or they can be separately made into individual integrated circuit modules, or multiple modules or steps among them can be made into a single integrated circuit module to implement. In this way, the present invention is not limited to any specific combination of hardware and software.

[0161] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0162] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, as well as the combination of flows and / or blocks in the flowchart and / or block diagram. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0163] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0164] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0165] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0166] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory. The memory is an example of computer-readable media.

[0167] A computer-readable medium includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassette tapes, magnetic disk storage, or other magnetic storage devices, or any other non-transitory medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0168] It should also be noted that the term "comprising", "including", or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0169] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0170] 1), The business requirement analysis method of the present application obtains all the dialogue voice data of business requirements, identifies the voice content and converts it into the corresponding text document, conducts requirement analysis based on the text document to generate a preliminary requirement document; according to the specified requirement template, uses large and small language models to extract effective information, and fills the extracted specified information into the corresponding positions of the template; according to the business requirements, uses the large and small language models to cooperate with each other to complete the function point analysis of the filled business requirement document, generates a functional requirement specification, and provides guidance for system design and development according to the generated detailed functional requirement specification. It is not necessary for developers to collect, summarize, extract key points, inductively analyze, and check the content of a large amount of multi-source heterogeneous original documents to obtain a functional requirement specification, which can improve the efficiency of requirement analysis for software project development in the banking industry and greatly reduce the time cost of developers. The present application solves the problem of consuming a large amount of time cost in the prior art for collecting, summarizing, extracting key points, inductively analyzing, and checking the content of a large amount of multi-source heterogeneous original documents.

[0171] 2) The business requirement analysis device of the present application obtains all the dialogue voice data of business requirements, identifies the voice content and converts it into the corresponding text document, conducts requirement analysis based on the text document to generate a preliminary requirement document; according to the specified requirement template, uses large and small language models to extract effective information, and fills the extracted specified information into the corresponding positions of the template; according to the business requirements, uses the large and small language models to cooperate with each other to complete the function point analysis of the filled business requirement document, generates a functional requirement specification, and provides guidance for system design and development according to the generated detailed functional requirement specification. There is no need for developers to obtain the functional requirement specification through tasks such as collecting and summarizing a large amount of multi-source heterogeneous original documents, extracting key points, inductive analysis, and content verification, which can improve the efficiency of requirement analysis for software project development in the banking industry and greatly reduce the time cost of developers. The present application solves the problem in the prior art that a large amount of time cost is consumed in tasks such as collecting and summarizing a large amount of multi-source heterogeneous original documents, extracting key points, inductive analysis, and content verification.

[0172] The above are only the preferred embodiments of the present application and are not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A business requirement analysis method, characterized in that Including: Obtain all the dialogue voice data of the business requirement discussion, and convert all the dialogue voice data into corresponding dialogue text documents; Generate a preliminary requirement document based on all the dialogue text documents; Extract information from the preliminary requirement document using a large and small language model according to a set requirement template to obtain specific information, and fill the specific information into the corresponding positions of the set requirement template to obtain a requirement document. The large and small language model is a model that uses a large model and a small model to work together. The large model is used for the overall semantic understanding of the preliminary requirement document, and the small model is used for precise analysis of the preliminary requirement document to assist the large model in filling the missing semantic understanding; Use the large and small language model to perform function point analysis on the requirement document to generate a functional requirement specification, so as to carry out business development according to the functional requirement specification, so that the developed product has the function points of the functional requirement specification.

2. The method according to claim 1, wherein Converting all the dialogue voice data into corresponding dialogue text documents includes: Use a speech large model to perform text translation on the dialogue voice data to generate the corresponding dialogue text document. The speech large model is a convergent model obtained by iteratively training a predetermined model using multiple sets of sample data. Each set of sample data includes sample voice data and sample text data corresponding to the sample voice data.

3. The method according to claim 1, wherein Generating a preliminary requirement document based on all the dialogue text documents includes: Perform preliminary verification and secondary verification on each dialogue text document in turn to obtain an initial requirement. The preliminary verification is a verification operation to correct misrecognized requirement information. The requirement information at least includes requirement priority and requirement detailed description. The secondary verification is to supplement the missing valid requirement content. The valid requirement content at least includes key requirement content not mentioned during the dialogue process and developer task assignment; Perform preliminary analysis and filtering of the requirement content of the initial requirement to remove information irrelevant to the business requirement or duplicate and redundant information, and form the preliminary requirement document.

4. The method according to claim 1, wherein The specific information at least includes requirement name, requirement detailed description, requirement priority, associated system, and expected online date. Extract information from the preliminary requirement document using a large and small language model according to a set requirement template to obtain specific information, and fill the specific information into the corresponding positions of the set requirement template to obtain a requirement document, including: Set the filling area of the specific information to obtain the set requirement template; Use the attention mechanism of the large and small language model to filter redundant information from the preliminary requirement document to form a requirement analysis text; Extract information from the requirement analysis text according to preset keywords to screen out statement information related to the preset keywords, and obtain the requirement name, the requirement detailed description, the associated system, and the expected online date in the specific information; Determine the requirement priority in the specific information according to the keyword frequency, where the keyword frequency includes high-frequency keywords and low-frequency keywords, and the requirement priority includes high priority and low priority; Fill the specific information into the corresponding filling area of the set requirement template to obtain the requirement document.

5. The method according to claim 4, characterized in that After determining the requirement priority in the specific information according to the keyword frequency, the method further includes: Calculate the word frequency score corresponding to each word in the dialogue text document according to the first formula, where the first formula is tfidf(t, d, D) = tf(t, d) × idf(t, D), where fidf(t, d, D) represents the word frequency score, tf(t, d) is the term frequency, tf(t, d) = log(1 + freq(t, d)), freq(t, d) represents the number of times the candidate word t appears in the current dialogue text document d, the candidate word is the word that appears in the dialogue text document, and idf(t, D) = log(N / count(d ∈ D: t ∈ D)) is the inverse document frequency, count(d ∈ D: t ∈ D) is the number of dialogue text documents containing the candidate word t, N represents the total number of dialogue text documents, and D represents the set of all dialogue text documents; Determine the words with the word frequency score greater than or equal to the set score as high-frequency keywords; Determine the words with the word frequency score less than the set score as the low-frequency keywords.

6. The method according to claim 1, characterized in that, Use the large and small language models to perform function point analysis on the requirement document to generate a functional requirement specification, including: Define the document format of the functional requirement specification, where the document format at least includes function points and function descriptions; Use the large model in the large and small language models to perform semantic analysis on the requirement document to determine multiple function points, where the function points at least include the trigger conditions, execution processes, expected outputs, exception handling, and mutual dependencies of the functions, and the mutual dependencies include the execution order between functions and the priority between functions; Use the small model in the large and small language models to perform semantic refinement on each function point to determine the function description corresponding to each function point; Use the text generation ability of the large model to convert the function points and the function descriptions corresponding to the function points into the form of the document format, and at the same time use the small model to check whether the conversion result meets the document requirements during the conversion process, where the document requirements at least include compliance with the document format and the correct use of professional terms; In the case where the conversion result meets the document requirements, determine the conversion result as the functional requirement specification; In the case where the conversion result does not meet the document requirements, regenerate the functional requirement specification.

7. The method according to claim 1, characterized in that, After using the large and small language models to perform function point analysis on the requirement document to generate a functional requirement specification, the method further includes: Determine various dimension information for requirement transposition, where the dimension information at least includes a personnel list, system components, business process stages, requirement priorities, development stages, requirement function points, and requirement progress; Automatically identify and extract the field information and requirement descriptions of each dimension information in the requirement document; Determine corresponding transposition rules according to each dimension information, so as to perform a transposition operation according to the transposition rules and form a list of function items for different dimension information. The transposition rule is a rule that will be displayed in different forms when different dimension information is selected. The transposition operation is to automatically display the field information and requirement description corresponding to the dimension information in a corresponding predetermined form when the corresponding dimension information is selected. The predetermined form at least includes a list form, a Gantt chart form, and a table form.

8. A business requirement analysis device, characterized in that, The device includes: An acquisition unit, configured to acquire all the dialogue voice data of business requirement discussions and convert all the dialogue voice data into corresponding dialogue text documents; A first generation unit, configured to generate a preliminary requirement document based on all the dialogue text documents; An extraction and filling unit, configured to extract information from the preliminary requirement document by using a large and small language model according to a set requirement template, obtain specific information, and fill the specific information into the corresponding position of the set requirement template to obtain a requirement document. The large and small language model is a model that uses a large model and a small model to work together. The large model is used for the overall semantic understanding of the preliminary requirement document, and the small model is used for precise analysis of the preliminary requirement document to assist the large model in filling in the missing semantic understanding; A second generation unit, configured to perform function point analysis on the requirement document by using the large and small language model, generate a functional requirement specification, and perform business development according to the functional requirement specification, so that the developed product has the function points of the functional requirement specification.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein when the program runs, it controls the device where the computer-readable storage medium is located to execute the method according to any one of claims 1 to 7.

10. A computer program product, comprising computer instructions, characterized in that, The computer instructions, when executed by a processor, implement the method according to any one of claims 1 to 7.

Images