An ai-based natural language requirement analysis method and system

Abstract

This application relates to the field of requirements analysis technology, specifically providing an AI-based natural language requirements analysis method and system. The method includes the following steps: acquiring natural language requirements text, then using AI-based feature extraction technology to extract features from the natural language requirements text to obtain text domain features, user style features, and task features; determining the combination of language analysis modules and the parameters corresponding to each language analysis module in the combination based on the text domain features, user style features, and task features; determining the execution order of each language analysis module in the combination based on the text domain features and task features, then constructing a requirements analysis process based on the execution order and the combination of language analysis modules; and using the requirements analysis process to perform requirements analysis on the natural language requirements text to obtain requirements information. This method enables the analysis process to adapt to users' personalized expressions and accurately analyze different analysis tasks.

Description

Technical Field

[0001] This application relates to the field of requirements analysis technology, and more specifically, to an AI-based natural language requirements analysis method and system. Background Technology

[0002] Existing technologies typically utilize requirements analysis methods to analyze natural language requirements text in order to extract user requirements from the natural language requirements text.

[0003] However, different users may exhibit very different writing styles and expression habits. Some users prefer to use concise and direct language with relatively simple sentence structures, while others may be accustomed to using more complex sentence structures with more modifiers and clauses. For example, when describing the same mobile phone, some users may say "this mobile phone has a long battery life", while others may say "this mobile phone is equipped with a high-capacity battery that can provide up to two days of battery life under normal use". Furthermore, different analytical tasks require attention to different linguistic phenomena and information types in the text. For example, when analyzing product reviews, the focus is on identifying consumers' evaluations of the product, their emotional tendencies, and the specific attributes or functions they care about. In this case, the linguistic phenomena to pay attention to include words expressing emotional coloring (such as "easy to use," "bad," "surprising"), words describing product attributes (such as "screen," "battery," "performance"), and evaluative phrases (such as "high cost-performance ratio," "not worth buying"). However, when conducting requirements analysis for technical documents, the focus shifts to the description of system functions, performance indicators, and technical specifications. In this case, the linguistic phenomena to pay attention to include verbs indicating system behavior (such as "implement," "support," "process"), words describing performance (such as "fast," "stable," "efficient"), and quantitative indicators (such as "response time less than 1 second," "supports 1000 concurrent users"). Because existing requirements analysis methods typically use a fixed and general analysis process to analyze natural language requirements text, this fixed and general analysis process cannot adapt to users' personalized expressions and accurately analyze different analysis tasks. Therefore, existing technologies suffer from the problem that the fixed and general analysis process is not adaptable enough to the differences in user styles and the diversity of analysis tasks, resulting in the requirements analysis results failing to accurately reflect the actual needs of users, thus leading to low accuracy and reliability of requirements analysis.

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

[0005] The purpose of this application is to provide an AI-based natural language requirements analysis method and system that can effectively solve the problem that the requirements analysis results cannot accurately reflect the actual needs of users due to the insufficient adaptability of fixed and general analysis processes when facing differences in user styles and the diversity of analysis tasks.

[0006] Firstly, this application provides an AI-based natural language requirements analysis method, which includes the following steps:

[0007] S1. Obtain the natural language requirement text, and then use AI-based feature extraction technology to extract features from the natural language requirement text to obtain text domain features, user style features and task features;

[0008] S2. Determine the combination of language analysis modules and the parameters corresponding to each language analysis module in the combination of language analysis modules based on text domain characteristics, user style characteristics and task characteristics;

[0009] S3. Determine the execution order of each language analysis module in the language analysis module combination based on the text domain characteristics and task characteristics, and then construct the requirements analysis process based on the execution order and the language analysis module combination.

[0010] S4. Utilize the requirements analysis process to perform requirements analysis on the natural language requirements text in order to obtain requirements information.

[0011] This application provides an AI-based natural language requirements analysis method. It first utilizes AI-based feature extraction technology to obtain text domain features, user style features, and task features. Then, based on these features, it dynamically determines the combination of language analysis modules, the parameters corresponding to each language analysis module in the combination, and the execution order of each language analysis module. This enables the construction of a requirements analysis process adapted to specific domains, user-personalized expressions, and analysis tasks. In other words, this application enables the analysis process to adapt to user-personalized expressions and accurately analyze different analysis tasks. Therefore, this application effectively solves the problem that fixed and universal analysis processes lack adaptability when facing differences in user style and the diversity of analysis tasks, resulting in requirements analysis results that cannot accurately reflect users' actual needs. This effectively improves the accuracy and reliability of requirements analysis.

[0012] Secondly, this application also provides an AI-based natural language requirements analysis system, which includes:

[0013] The feature extraction module is used to acquire natural language requirement text, and then use AI-based feature extraction technology to extract features from the natural language requirement text to obtain text domain features, user style features and task features.

[0014] The parameter confirmation module is used to determine the combination of language analysis modules and the parameters corresponding to each language analysis module in the combination of language analysis modules based on text domain features, user style features and task features.

[0015] The analysis process confirmation module is used to determine the execution order of each language analysis module in the language analysis module combination based on the text domain characteristics and task characteristics, and then construct the requirements analysis process based on the execution order and the language analysis module combination.

[0016] The requirements analysis module is used to perform requirements analysis on natural language requirements text using the requirements analysis process to obtain requirements information.

[0017] This application provides an AI-based natural language requirements analysis system. It first utilizes AI-based feature extraction technology to obtain text domain features, user style features, and task features. Then, based on these features, it dynamically determines the combination of language analysis modules, the parameters corresponding to each language analysis module in the combination, and the execution order of each language analysis module. This enables the construction of a requirements analysis process adapted to specific domains, personalized user expressions, and analysis tasks. In other words, this application enables the analysis process to adapt to personalized user expressions and accurately analyze different analysis tasks. Therefore, this application effectively solves the problem that fixed and universal analysis processes lack adaptability when facing differences in user style and the diversity of analysis tasks, resulting in requirements analysis results that cannot accurately reflect users' actual needs. This effectively improves the accuracy and reliability of requirements analysis.

[0018] As can be seen from the above, the AI-based natural language requirements analysis method and system provided in this application can first obtain text domain features, user style features, and task features using AI-based feature extraction technology, and then dynamically determine the combination of language analysis modules, the parameters corresponding to each language analysis module in the combination, and the execution order of each language analysis module in the combination based on these features. This enables the construction of a requirements analysis process that is adapted to specific domains, user-personalized expressions, and analysis tasks. In other words, this application enables the analysis process to adapt to user-personalized expressions and accurately analyze different analysis tasks. Therefore, this application can effectively solve the problem that the requirements analysis results cannot accurately reflect the actual needs of users due to the insufficient adaptability of fixed and general analysis processes when facing differences in user style and diversity of analysis tasks, thereby effectively improving the accuracy and reliability of requirements analysis. Attached Figure Description

[0019] Figure 1 A flowchart illustrating an AI-based natural language requirements analysis method provided in this application embodiment.

[0020] Figure 2 This is a schematic diagram of the structure of an AI-based natural language requirements analysis system provided in an embodiment of this application.

[0021] Attached reference numerals: 1. Feature extraction module; 2. Parameter confirmation module; 3. Analysis process confirmation module; 4. Requirements analysis module. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0023] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0024] Firstly, such as Figure 1 As shown, this application provides an AI-based natural language requirements analysis method, which includes the following steps:

[0025] S1. Obtain the natural language requirement text, and then use AI-based feature extraction technology to extract features from the natural language requirement text to obtain text domain features, user style features and task features;

[0026] S2. Determine the combination of language analysis modules and the parameters corresponding to each language analysis module in the combination of language analysis modules based on text domain characteristics, user style characteristics and task characteristics;

[0027] S3. Determine the execution order of each language analysis module in the language analysis module combination based on the text domain characteristics and task characteristics, and then construct the requirements analysis process based on the execution order and the language analysis module combination.

[0028] S4. Utilize the requirements analysis process to perform requirements analysis on the natural language requirements text in order to obtain requirements information.

[0029] The natural language demand text in step S1 refers to the user's demand content expressed in everyday language. This embodiment can obtain the natural language demand text by receiving input from the user interface, reading text content from a file, or obtaining text data through a network interface. Utilizing AI-based feature extraction technology to extract features from the natural language demand text refers to using artificial intelligence models or algorithms to perform deep analysis on the natural language text to identify and extract information with specific meanings or attributes. Specifically, this AI-based feature extraction technology can be a deep learning model (such as a recurrent neural network (RNN), a long short-term memory network (LSTM), or a Transformer model) or a pre-trained language model (BERT or GPT). Step S1 can obtain structured or semi-structured feature information from the original natural language demand text by utilizing AI-based feature extraction technology to provide input for subsequent analysis steps. Domain features refer to the professional field or topic category (e.g., finance, healthcare, law, or technology) to which the natural language requirement text belongs. This embodiment can use text classification algorithms or keyword extraction methods to extract domain features (e.g., determining the domain by analyzing professional vocabulary or phrases in the natural language requirement text). These domain features reflect the professional attributes of the natural language requirement text content, guiding the selection of language analysis tools suitable for that domain. User style features refer to the language habits and preferences exhibited by users when expressing their needs (e.g., the formality of word choice, sentence complexity, presence of colloquial expressions or omissions, etc.). This embodiment can use style analysis models or statistical methods to extract user style features (e.g., quantifying style by calculating sentence length distribution, lexical diversity index, or frequency of specific syntactic structures). These user style features reflect the user's personalized expression style, guiding the selection of language analysis tools suitable for that user's expression style. Task features refer to the specific goals or focus of this requirements analysis (e.g., identifying functional requirements, performance requirements, or user interface requirements). This embodiment can use a task feature extraction model or a recognition method based on preset rules to extract task features (e.g., determining the task by analyzing the type of questions raised by the user or specific combinations of verbs and nouns appearing in the text). These task features can reflect the purpose of the task analysis, guiding the system to focus on extracting information related to the task.

[0030] The language analysis module combination in step S2 refers to a collection of functional modules used for processing and analyzing natural language text. This combination may include named entity recognition, syntactic analysis, semantic role labeling, and sentiment analysis modules. The parameters corresponding to each language analysis module in the combination refer to settings that configure and control the behavior of each module, such as model weights, thresholds, dictionary paths, and algorithm hyperparameters. Step S2 can determine the language analysis module combination and the parameters corresponding to each language analysis module within it by querying a pre-built mapping table based on text domain features, user style features, and task features. This embodiment can also... The system selects appropriate language analysis modules from a pre-set language analysis module library based on text domain features, user style features, and task features. It then determines the combination of language analysis modules and the parameters of each language analysis module within the combination by setting parameters for the language analysis modules. For example, if the text domain features indicate that the text belongs to the "medical" domain, the user style features indicate that the text has a "professional and rigorous" style, and the task features indicate that "disease entity recognition" is required, the system can select named entity recognition modules and medical terminology standardization modules specifically for the medical field from the pre-set language analysis module library, and adjust the confidence threshold of the named entity recognition model to meet the high accuracy requirements of the medical field. This embodiment enables the selection of appropriate language analysis module combinations based on the professional attributes of the natural language requirement text content, the user's personalized expression style, and the purpose of the analysis task. This allows for the extraction of structured language information by determining the combination of language analysis modules according to text domain characteristics, user style characteristics, and task characteristics. Furthermore, by determining the parameters of each language analysis module within the combination based on these factors, the embodiment allows for fine-grained control over the operation of each language analysis module, ensuring that each module adapts to the current text and analysis task.

[0031] Step S3, determining the execution order of each language analysis module in the language analysis module combination, refers to arranging the processing order of the selected language analysis modules. Step S3 can determine the execution order of each language analysis module in the language analysis module combination by querying a pre-built mapping table based on text domain features and task features. Alternatively, Step S3 can determine the execution order of each language analysis module in the language analysis module combination by inputting text domain features and task features into a pre-trained execution order allocation model. For example, for a task requiring sentiment analysis and entity recognition, word segmentation and part-of-speech tagging are usually performed first, followed by named entity recognition, and finally sentiment analysis. Preferably, if the text domain features indicate that the text contains a large number of technical terms, a terminology recognition module can be added before word segmentation. The requirements analysis process refers to the processing flow for end-to-end analysis of natural language requirement text constructed based on the determined language analysis module combination and execution order. Step S3 can achieve the construction of the requirements analysis process based on the execution order and language analysis module combination by chaining all selected language analysis modules in the execution order to form a complete processing chain.

[0032] Step S4, which involves performing requirements analysis on the natural language requirements text using the requirements analysis process, refers to inputting the original natural language requirements text into the constructed requirements analysis process. After being processed sequentially by each language analysis module, the original natural language requirements text will be transformed into structured requirements information. In this instance, the requirements information refers to the structured data (such as requirements items, requirements types, priorities, related entities, constraints, etc.) extracted from the natural language requirements text that describes the user's requirements.

[0033] The core innovation of this application lies in constructing a requirements analysis process that adapts to specific domains, personalized user expressions, and analysis tasks. This is achieved by first utilizing AI-based feature extraction technology to acquire text domain features, user style features, and task features, and then dynamically determining the combination of language analysis modules, the parameters of each language analysis module within the combination, and the execution order of each language analysis module based on these features. In essence, this application enables the analysis process to adapt to personalized user expressions and accurately analyze different analysis tasks. For example, when a user's style leans towards colloquialism, the system can automatically select a language analysis module that focuses more on colloquial processing and adjust its parameters to improve recognition accuracy. When the task emphasizes technical details, the system will prioritize and configure modules that excel in entity extraction and relation recognition. Therefore, this application effectively solves the problem that fixed and universal analysis processes lack adaptability when facing differences in user style and the diversity of analysis tasks, resulting in requirements analysis results that fail to accurately reflect users' actual needs. This effectively improves the accuracy and reliability of requirements analysis.

[0034] Specifically, this method first acquires the natural language requirement text to be analyzed. Next, it utilizes AI-based feature extraction technology to conduct in-depth analysis of the text, extracting text domain features reflecting the professionalism of the content, user style features reflecting user expression habits, and task features indicating the analysis objective. These features serve as the foundation for dynamic configuration in subsequent steps. Then, based on the extracted text domain features, user style features, and task features, the system dynamically selects the most suitable set of language analysis modules for the current text and task, and configures optimal operating parameters for these selected modules. This step ensures the applicability and effectiveness of the analysis tools. Further, the system dynamically determines the execution order of the selected language analysis modules when processing the text based on the text domain features and task features, and constructs a customized requirement analysis process based on the determined module combination, parameters, and execution order to optimize the output structure of the requirement analysis process. Finally, the requirement analysis process is used to analyze the natural language requirement text, ultimately obtaining structured requirement information. The entire process achieves adaptive adjustment of the analysis process by perceiving and responding to the features of the input text and the analysis task, overcoming the limitations of fixed and generic analysis processes.

[0035] As a preferred implementation, the solution of this application is implemented as follows: First, the natural language requirement text input by the user is acquired, and a natural language processing model based on deep learning (AI-based feature extraction technology) is used to analyze the natural language requirement text to extract the domain to which the natural language requirement text belongs (text domain features), the user's writing style (user style features), and the task type of this analysis (task features). Then, the system queries a preset configuration strategy library based on the extracted text domain features, user style features, and task features to dynamically select a set of language analysis modules, and loads model parameters or dictionaries optimized for the current domain and style for each language analysis module. At the same time, the system queries a preset order strategy library based on the text domain features and task features to determine the execution order of these language analysis modules (e.g., first perform named entity recognition, then perform dependency parsing to determine the subject-verb-object structure). Subsequently, the selected language analysis modules are organized into a processing pipeline according to the execution order to form a customized requirement analysis process. Finally, the natural language requirement text is input into this customized requirement analysis process for processing to gradually extract key requirement information from the text, such as identifying "the user needs to purchase a game that can be played by multiple people".

[0036] This application provides an AI-based natural language requirements analysis method. It first utilizes AI-based feature extraction technology to obtain text domain features, user style features, and task features. Then, based on these features, it dynamically determines the combination of language analysis modules, the parameters corresponding to each language analysis module in the combination, and the execution order of each language analysis module. This enables the construction of a requirements analysis process adapted to specific domains, user-personalized expressions, and analysis tasks. In other words, this application enables the analysis process to adapt to user-personalized expressions and accurately analyze different analysis tasks. Therefore, this application effectively solves the problem that fixed and universal analysis processes lack adaptability when facing differences in user style and the diversity of analysis tasks, resulting in requirements analysis results that cannot accurately reflect users' actual needs. This effectively improves the accuracy and reliability of requirements analysis. Furthermore, different business domains have their own unique terminology, conceptual systems, and expression habits. For example, in the financial field, "account" usually refers to a bank account or trading account, while in the software development field, "account" may refer to a user's login credentials. Since the text domain features of this embodiment can determine the professional domain to which the natural language requirement text belongs, and this embodiment determines the combination of language analysis modules, the parameters corresponding to each language analysis module in the combination of language analysis modules, and the execution order of each language analysis module in the combination of language analysis modules based on the text domain features, this embodiment can effectively avoid the situation where the final requirement information is not accurate due to the failure to consider the professional domain to which the natural language requirement text belongs during requirement analysis.

[0037] In some preferred embodiments, step S2 includes:

[0038] S21. Obtain confidence information on text domain features, user style features, and task features;

[0039] S22. Based on the confidence information, the text domain features, user style features, and task features are weighted to obtain weighted domain features, weighted style features, and weighted type features;

[0040] S23. Select language analysis modules from the preset language analysis module library based on weighted domain features, weighted style features, and weighted type features to obtain a combination of language analysis modules;

[0041] S24. Determine the parameters corresponding to each language analysis module in the language analysis module combination based on the weighted domain features, weighted style features, and weighted type features.

[0042] Confidence information refers to a quantitative assessment of the reliability, certainty, or importance of text domain features, user style features, and task features. This confidence information can be a probability value, a confidence interval, a score, or a level. It reflects the effectiveness and guidance of each feature (text domain features, user style features, and task features) in the current natural language requirement text. This confidence information can be output by AI-based feature extraction technology when identifying features, or it can be obtained by evaluating the degree of matching between features and a preset knowledge base or historical data. Weighted processing refers to adjusting the original text domain features, user style features, and task features based on confidence information to generate weighted features. Specifically, features with higher confidence are assigned greater weights to increase their influence in subsequent decisions; conversely, features with lower confidence are assigned smaller weights. In this embodiment, weighted processing can be achieved by multiplying the feature vector with the corresponding confidence value. Compared to the original text domain features, user style features, and task features, the weighted domain features, weighted style features, and weighted type features in this embodiment can more accurately reflect the actual situation and analysis focus of the current natural language demand text. The pre-defined language analysis module library is a collection of various language analysis modules with different functions. Each module has a clear functional description, scope of application, performance indicators, and compatibility information with other modules. For example, the module library may include named entity recognition module, sentiment analysis module, syntactic analysis module, and topic modeling module. This embodiment can intelligently select the most suitable language analysis module from the module library based on weighted domain features, weighted style features, and weighted type features to form a customized language analysis module combination. The selection process in this embodiment can be based on a matching degree algorithm and prioritize modules that highly match the weighted features. In step S24, once the language analysis module combination is determined, the system will determine the corresponding operation parameters for each language analysis module in the combination based on weighted domain features, weighted style features, and weighted type features. These parameters may include, but are not limited to, model version, threshold setting, dictionary selection, output format, or specific algorithm configuration. This embodiment can maximize the analysis effect by dynamically determining the module parameters so that each selected module runs in the optimal state in the current requirements analysis process. For example, if the weighted domain features indicate that the text belongs to the legal domain, the parameters of the legal analysis module may be adjusted to focus more on identifying legal entities and regulatory citations.Step S24 can use predefined rules or lookup tables to determine the parameters of each language analysis module in the language analysis module combination based on weighted domain features, weighted style features, and weighted type features. For example, a series of configuration files can be pre-created, each corresponding to a text feature combination (a combination of domain features, style features, and type features) and a complete set of language analysis module parameters (e.g., the dictionary path of the word segmenter, the model file of the part-of-speech tagger, the entity type list of the named entity recognizer, and the grammar rule set of the syntactic analyzer). After obtaining the weighted domain features, weighted style features, and weighted type features, the system first combines the weighted domain features, weighted style features, and weighted type features into a feature vector or a multi-dimensional index. Then, based on this feature vector or multi-dimensional index, the system searches for the most matching configuration file in the lookup table. The matching process is based on similarity calculation (e.g., if the feature vector has the highest similarity to the feature description of a certain file, then that file is selected). Finally, the parameters of each language analysis module in the language analysis module combination are set according to the language analysis module parameters corresponding to the matched configuration file.

[0043] After acquiring the various features of the natural language requirement text, this embodiment further evaluates the confidence level of these features. For example, if the confidence level of a certain text's domain feature recognition is high, it indicates that the domain feature has stronger guiding significance for understanding the text content, and will be given higher weight during weighted processing. Therefore, this embodiment can, by introducing confidence information and weighting the text's domain features, user style features, and task features, make the weighted domain features, style features, and type features more accurately reflect the actual situation and analysis focus of the current requirement text. This allows for priority selection of modules that highly match these weighted features when choosing modules from the preset language analysis module library, and avoids module selection bias caused by uncertainty in feature information. Simultaneously, the parameters of each module can be finely adjusted based on these weighted features, ensuring that each module operates optimally in the requirement analysis process. This effectively solves the problems of inaccurate module selection and parameter configuration or insufficient adaptability that may exist in the basic solution.

[0044] In some preferred embodiments, suppose there is a natural language requirement text whose content relates to "tax compliance analysis of cryptocurrency". In step S1, through AI-based feature extraction technology, the domain features of the text may be identified as "finance" and "law", the user style feature as "professional", and the task features as "compliance analysis" and "information extraction". Without confidence information and weighted processing, the system may treat these features equally, thus selecting a general financial analysis module and a general legal analysis module, and using default parameters. However, according to the scheme of this embodiment, in step S21, the system obtains the confidence information of these features. For example, through AI model evaluation, it may be found that the confidence of the "finance" domain feature is 0.9, while the confidence of the "law" domain feature is 0.7; the confidence of the "compliance analysis" task feature is 0.8, while the confidence of the "information extraction" task feature is 0.6. In step S22, the system will perform weighted processing on the original features based on these confidence information. For example, the weight of the "finance" domain features will be increased, the weight of the "law" domain features will be appropriately increased, and the weight of the "compliance analysis" task features will be increased, thereby obtaining weighted domain features, weighted style features, and weighted type features.

[0045] In step S23, based on these weighted features, the system selects modules from a pre-defined language analysis module library. Since "finance" and "compliance analysis" have higher weights, the system may prioritize selecting a module specifically for "financial compliance text analysis" and a module optimized for "legal clause information extraction," rather than a general module. In step S24, the system determines the parameters of these selected modules based on the weighted features. For example, the parameters for the "financial compliance text analysis" module may be adjusted to focus more on identifying financial terms, regulatory requirements, and risk indicators; the parameters for the "legal clause information extraction" module may be adjusted to focus more on identifying legal entities, regulatory citations, and case law information. In this way, this embodiment can dynamically adjust the analysis strategy based on the confidence level of the features to ensure that the selected modules and parameter configurations more accurately match the specific need for "cryptocurrency tax compliance analysis," thereby providing more accurate and in-depth needs analysis results.

[0046] In some preferred embodiments, the confidence information includes the initial confidence levels corresponding to the weighted domain features, weighted style features, and weighted type features, as well as the correlation confidence levels between the features. Step S22 includes:

[0047] S221. Based on the initial confidence level, the association confidence level, and the preset weighting rules, perform preliminary weighting processing on the text domain features, user style features, and task features to obtain preliminary domain features, preliminary style features, and preliminary type features;

[0048] S222. Based on the preliminary domain features, preliminary style features, and preliminary type features, identify whether there is cross-domain content, style mixing, or multi-task requirements in the natural language requirement text. If so, adjust the weighting rules according to the identification results and execute step S223. If not, use the preliminary domain features as weighted domain features, the preliminary style features as weighted style features, and the preliminary type features as weighted type features, and execute step S23.

[0049] S223. Perform secondary weighting on the preliminary domain features, preliminary style features, and preliminary type features according to the adjusted weighting rules to obtain weighted domain features, weighted style features, and weighted type features, and then execute step S23.

[0050] Initial confidence can be understood as an assessment of the reliability or importance of a single feature in a specific context. In this embodiment, the initial confidence can be generated using predicted probabilities based on machine learning models (such as support vector machines or neural networks) or expert system rules. Association confidence refers to a quantitative representation of the degree of mutual influence and dependence between different features. In this embodiment, mutual information or Pearson correlation coefficients can be used to calculate association confidence. Preliminary weighting processing refers to the process of initially assigning weights to text domain features, user style features, and task features based on the initial confidence and association confidence. Specifically, this can be achieved using linear weighting, non-linear weighting, or attention-based weighting methods. For example, the original feature vector can be scaled by multiplying or adding the initial confidence and association confidence, combined with a preset weighting rule. The preset weighting rule refers to a set of weight allocation strategies pre-defined before the preliminary weighting processing. Specifically, it can be defined using a rule set based on expert experience, a weight distribution based on historical data statistics, or a weight function trained through a machine learning model (such as reinforcement learning). Preliminary domain features, preliminary style features, and preliminary type features refer to the feature representations obtained after preliminary weighting. They reflect the feature importance after considering the initial confidence and association confidence. In step S222, identifying whether there is cross-domain content, style mixing, or multi-task requirements in the natural language requirement text refers to the process of determining whether the text has complex features by analyzing the preliminary weighted features. Specifically, this can be achieved using existing clustering analysis algorithms. For example, the presence of cross-domain content can be determined by analyzing the uniformity of the distribution of preliminary domain features across multiple domains. Adjusting the weighting rules refers to modifying the preset weighting rules based on the recognition results to better adapt to the complexity of the text. Specifically, this can be achieved using dynamic adjustment based on a preset rule engine or parameter adjustment methods based on Bayesian optimization. For example, if cross-domain content is identified, the weight of domain features can be increased to ensure the recognition of different domains.

[0051] This solution addresses the shortcomings of existing weighting methods in complex natural language text analysis by introducing more comprehensive confidence information and a mechanism for dynamically adjusting weighting rules. This improves the accuracy and flexibility of feature weighting, providing a more reliable foundation for the selection of subsequent language analysis modules and parameter determination. The confidence information includes not only the initial confidence levels corresponding to weighted domain features, weighted style features, and weighted type features, but also the correlation confidence levels between features. This expanded confidence information more comprehensively reflects the inherent attributes of features and their interrelationships, providing a richer data foundation for subsequent weighting processing. Step S221, through the introduction of correlation confidence, allows the initial weighting processing to consider the mutual influence between different features. For example, in a specific domain, a certain user style may be more indicative, or a certain task type may be strongly correlated with specific domain features. Correlation confidence allows for more accurate initial weighting, enabling the initial features to better reflect the actual situation of the text. In step S222, the complexity of the natural language text is actively detected through further analysis of the preliminary weighted results. If the recognition result indicates the presence of cross-domain content, mixed styles, or multi-task requirements, it suggests that the preset weighting rules may be insufficient to handle these complexities. The weighting rules are adjusted based on the recognition results. This dynamic recognition mechanism allows the system to adaptively adjust according to the actual complexity of the text. By adjusting the weighting rules, the system can selectively strengthen or weaken the weights of certain features to better adapt to the complexity of the text. For example, for cross-domain content, the weights of domain features can be increased to ensure recognition of different domains; for mixed styles, the weights of style features can be adjusted to balance the influence of different styles. If the recognition result indicates the absence of these complexities, the preliminary domain features, preliminary style features, and preliminary type features are directly used as the weighted domain features, weighted style features, and weighted type features, and step S23 is executed. This avoids unnecessary rule adjustments and improves processing efficiency. In step S223, the preliminary domain features, preliminary style features, and preliminary type features are subjected to secondary weighting processing according to the adjusted weighting rules. Since the secondary weighting processing is carried out under the adjusted weighting rules, this embodiment can ensure that the final weighted domain features, weighted style features, and weighted type features can more accurately reflect the complexity and diversity of the text, thereby providing more accurate input for the subsequent selection of language analysis module combinations and parameter determination.

[0052] In some preferred embodiments, the step of identifying whether there is cross-domain content, style mixing, or multi-task requirements in the natural language requirement text based on preliminary domain features, preliminary style features, and preliminary type features includes:

[0053] A1. Based on the preliminary domain characteristics and the hierarchical relationship in the pre-set domain knowledge graph, analyze the co-occurrence frequency and semantic association strength of terms in each domain in the natural language requirement text, so as to identify whether there is cross-domain content in the natural language requirement text and determine the boundary information between each domain.

[0054] A2. Based on preliminary style characteristics, analyze the distribution patterns of syntactic structure, vocabulary selection, sentiment tendency, and rhetorical devices in natural language requirement texts to identify whether style mixing exists in natural language requirement texts.

[0055] A3. Based on the preliminary type characteristics analysis, analyze the frequency of verbs, noun phrases and preset keywords in the natural language requirement text and the context to identify whether there are multi-task requirements in the natural language requirement text and determine the priority of each task.

[0056] The steps for adjusting the weighting rules based on the identification results include:

[0057] Based on the cross-domain content recognition results, style mixing recognition results, and multi-task requirement recognition results, and combined with the preset adjustment strategy, the weighting rules are adjusted. The adjustment strategy includes increasing, decreasing, or balancing the weights of preliminary domain features, preliminary style features, and / or preliminary type features.

[0058] In step A1, preliminary domain features refer to the text's tendency representation in different domains obtained through preliminary weighting processing. These preliminary domain features can be represented using word vectors, topic models, or knowledge graph embedding. The hierarchical relationship in the preset domain knowledge graph refers to the hierarchical and parallel relationships between different domains, subdomains, and related concepts in the domain knowledge graph. The co-occurrence frequency of domain terms refers to the number of times different domain terms appear simultaneously in the natural language demand text. This embodiment can use methods such as sliding windows, dependency parsing, or co-occurrence matrices to obtain the collinearity frequency of domain terms. Semantic association strength refers to the degree of semantic relevance between different domain terms. This embodiment can use word embedding similarity, topic similarity, or... This embodiment uses methods such as knowledge graph path length to obtain semantic association strength. It can identify whether a natural language requirement contains content from multiple domains by analyzing the co-occurrence frequency and semantic association strength of terms in various domains in the natural language requirement text, and clarify the boundaries between these domains. For example, when analyzing a requirement text about "smart home security system", if the text contains terms such as "Internet of Things devices", "data encryption" (information security domain) and "user interface design" (human-computer interaction domain), and these terms have a high co-occurrence frequency and strong semantic association strength, it can be identified that the text contains cross-domain content, and it can be determined that "Internet of Things", "information security" and "human-computer interaction" are the domains involved in the text. In step A2, preliminary style features refer to the text's tendency to exhibit different writing styles after preliminary weighting. In this embodiment, preliminary style features can be represented by syntactic complexity indicators, lexical diversity indicators, or sentiment polarity scores. Syntactic structure refers to the organization of sentences in the text. In this embodiment, syntactic structure can be analyzed using methods such as syntactic trees, dependency parsing graphs, or syntactic pattern recognition. Lexical selection refers to the types and frequencies of words used in the text. In this embodiment, lexical selection can be analyzed using methods such as word frequency statistics, part-of-speech tagging, or professional dictionary matching. Sentiment tendency refers to the emotional tone expressed in the text. In this embodiment, a preset sentiment dictionary or sentiment classifier can be used. This embodiment analyzes sentiment tendencies using methods such as metaphor, parallelism, and hyperbole. Rhetorical devices refer to the expressive methods used in a text, such as metaphor, parallelism, and hyperbole. This embodiment can analyze rhetorical devices using preset rhetorical rules or pattern matching algorithms. By analyzing the distribution patterns of syntactic structure, vocabulary selection, sentiment tendencies, and rhetorical devices in natural language demand text, this embodiment can identify whether there is a mixture of multiple writing styles in the text. For example, when analyzing a product review, if the text contains both statements that objectively describe the product's functions (such as "This phone is equipped with a 12-megapixel camera") and statements that express strong emotions (such as "This phone is simply amazing!"), then the text can be identified as having a mixture of styles.In step A3, preliminary type features refer to the tendency of the text to be processed in different task types through preliminary weighting. In this embodiment, preliminary type features can be represented by task-related word frequency, task verb recognition, or task pattern matching. Verbs refer to words in the text that express actions or states. Noun phrases refer to combinations of words in the text that express entities or concepts. Preset keywords refer to words that are highly related to a specific task type. These preset keywords can be obtained by expert definition or statistical analysis. Contextual context refers to the surrounding environment of words or phrases in natural language requirement text. In this embodiment, the contextual environment can be analyzed by windowing mechanisms or dependency relationships. This embodiment can identify whether the text contains multiple task requirements and determine the priority of each task by analyzing the frequency of occurrence of verbs, noun phrases, and preset keywords in natural language requirement text and the contextual context. For example, when analyzing a software requirement document, if the text contains both "implement user login function" (functional requirement) and "system response time less than 1 second" (performance requirement), it can be identified that the text contains multiple task requirements, and the priority of functional requirements and performance requirements can be determined based on their frequency of occurrence and contextual context. The adjustment strategy can be understood as a series of predefined rules or algorithms used to dynamically adjust the weights of preliminary domain features, preliminary style features, and / or preliminary type features based on the identified cross-domain content recognition results, style mixing recognition results, or multi-task requirement recognition results. For example, when significant cross-domain content is identified in the text, the weights of relevant domain features can be increased, or the weights of different domain features can be balanced to ensure that all relevant domain information is fully considered; when style mixing is identified, the weights of user style features can be adjusted to better adapt to mixed style analysis; when multi-task requirements are identified, the weights of task features can be adjusted according to task priority to prioritize high-priority tasks.

[0059] This embodiment achieves a more comprehensive and in-depth understanding of text complexity by meticulously identifying cross-domain content, style mixing, and multi-task requirements in natural language processing texts. This refined identification allows for more accurate and adaptive adjustments to subsequent weighting rules. Specifically, when cross-domain content is identified, analyzing the co-occurrence frequency and semantic association strength of domain terms ensures a reasonable allocation of feature weights across different domains, preventing single-domain features from overdoing or ignoring important auxiliary domain information. When style mixing is identified, analyzing syntactic structure and vocabulary selection allows for adjustments to style feature weights that better align with the text's actual expression habits, thereby improving the accuracy of understanding user intent. When multi-task requirements are identified, analyzing verbs, noun phrases, and keywords allows for adjustments to task feature weights based on task priority, ensuring that critical tasks are prioritized. Thus, the adjustment of weighting rules is no longer static but dynamically optimized based on the text's actual complexity, providing more accurate and adaptive input for the selection of subsequent language analysis module combinations and parameter determination.

[0060] In some preferred embodiments, step S23 includes:

[0061] S231. Obtain the functional description, applicable scope and compatibility information of each module in the preset language analysis module library;

[0062] S232. Based on the weighted domain characteristics, weighted style characteristics, and weighted type characteristics, and combined with the functional descriptions and applicable scope of each module, a set of language analysis modules related to the current analysis needs is initially selected.

[0063] S233. Based on the compatibility information between modules in the initially selected language analysis module set, identify whether there are modules with overlapping or mutually exclusive functions in the module set, and perform conflict resolution and redundancy removal on the module set according to the identification results to obtain the language analysis module combination.

[0064] The functional description can be understood as the specific language processing tasks that the module can perform, such as lexical analysis, syntactic analysis, named entity recognition, sentiment analysis, etc. The scope of application refers to the domains, styles, or task types for which the module performs best. The compatibility information indicates whether there are dependencies, functional overlaps, or mutual exclusions between the module and other modules. This information is usually stored in the form of structured data, such as metadata tags, configuration files, or knowledge graphs. This information can provide comprehensive basic data for subsequent module selection and combination optimization. In step S232, weighted domain features, weighted style features, and weighted type features are used to guide the initial screening process. Specifically, these weighted features reflect the domain orientation, user expression style, and core task type of the current natural language processing text. The system matches these weighted features with the functional descriptions and applicable scopes of each module in the preset language analysis module library. For example, if the weighted domain feature indicates that the text belongs to the "finance" domain, then the named entity recognition module or event extraction module, which performs well in the finance domain, will be prioritized. If the weighted style feature indicates that the text is in a "formal" style, then the syntactic analysis module, which has strong capabilities in processing formal language, will be preferred. In this way, a set of language analysis modules highly relevant to the current analysis needs can be initially screened, with the aim of narrowing the selection range and improving the efficiency of subsequent processing. In step S233, the initially selected set of language analysis modules is further optimized. Specifically, the system uses the compatibility information obtained in step S231 to identify whether there are modules with overlapping or mutually exclusive functions in the module set. Functional overlap refers to two or more modules performing similar or identical tasks. Overlapping modules can lead to redundant calculations and wasted resources. Mutually exclusive modules refer to two modules that cannot logically or technically coexist in the same analysis process, otherwise, it will lead to erroneous or inconsistent results. For example, two different sentiment analysis modules may both aim to identify text sentiment, but their internal algorithms or output formats may differ. Using them simultaneously may cause redundancy or conflict. Once these problems are identified, the system will process them according to preset conflict resolution and redundancy removal strategies. Specifically, this embodiment can resolve conflicts by retaining only the best-performing module among the conflicting modules. This embodiment can remove redundant modules by removing modules with overlapping functions and retaining only the best-performing or most suitable module for the current needs. This embodiment can obtain a streamlined, efficient, and collaborative combination of language analysis modules by resolving conflicts and removing redundancy in the module set, ensuring the accuracy and efficiency of the final requirements analysis process.

[0065] This embodiment effectively solves the problems of module function overlap and mutual exclusion that may exist in traditional methods by introducing the acquisition and utilization of module function descriptions, applicable scope and compatibility information in the process of determining the combination of language analysis modules, and further performing preliminary screening, conflict resolution and redundancy removal. Specifically, step S231 provides comprehensive metadata support for subsequent intelligent selection; step S232 performs preliminary screening based on weighted features and module metadata to ensure the relevance of the selected modules to the current requirements; and step S233 actively identifies and resolves potential functional overlap and mutual exclusion problems through in-depth analysis of the compatibility between modules. This not only optimizes the utilization of computing resources and reduces unnecessary processing overhead, but also ensures the synergy and efficiency of the selected module combination. Therefore, this embodiment can avoid problems such as low analysis efficiency, inaccurate results or waste of resources caused by improper module selection, thereby effectively improving the accuracy and robustness of the entire requirements analysis process. In other words, this embodiment enables the finally constructed requirements analysis process to analyze natural language requirements text more efficiently and accurately, thereby effectively improving the practical value and performance of AI-based natural language requirements analysis methods.

[0066] In some preferred embodiments, step S3 includes:

[0067] S31. Determine the initial execution order of each language analysis module in the language analysis module combination based on the text domain characteristics and task characteristics;

[0068] S32. Construct the initial requirements analysis process based on the initial execution order and the combination of language analysis modules;

[0069] S33. During the execution of the initial requirements analysis process, obtain the intermediate processing results of any language analysis module;

[0070] S34. Determine whether the intermediate processing results contain specific language phenomena or missing information. Specific language phenomena or missing information indicate that the subsequent language analysis module needs to be adjusted.

[0071] S35. When the intermediate processing result contains specific language phenomena or missing information, select the corresponding adjustment rule from the preset adjustment rule set according to the intermediate processing result.

[0072] S36. Adjust the subsequent language analysis modules according to the adjustment rules, including redetermining the execution order of the subsequent analysis modules;

[0073] S37. Based on the adjusted execution order, update the requirements analysis process and continue executing the requirements analysis process.

[0074] In step S31, the initial execution order can be determined based on a preset rule base or expert knowledge. These rules or knowledge comprehensively consider the impact of text domain features and task features on the processing order of different language analysis modules. For example, for a task focusing on sentiment analysis, the sentiment analysis module may be arranged after lexical analysis and syntactic analysis to better utilize its output. In step S33, the intermediate processing result refers to the output data generated by any language analysis module after completing its processing task. This output data can be preliminarily annotated text, extracted entity lists, syntactic tree structures, semantic relation graphs, or any other form of intermediate analysis product. The purpose of obtaining the intermediate processing result is to monitor the progress and quality of the analysis process in real time. Step S34 determines whether the intermediate processing result contains specific linguistic phenomena or missing information. Specific linguistic phenomena may include, but are not limited to, ambiguous sentences, unclear referents, sentiment reversals, multiple negations, and domain terminology confusion. Missing information means that key entities have not been extracted, core intentions have not been identified, or necessary semantic connections have not been established at the current processing stage. The existence of specific linguistic phenomena or missing information indicates that the current analysis path is not optimal, and therefore, the current analysis path needs to be adjusted. In step S35, when the intermediate processing result contains specific linguistic phenomena or missing information, a corresponding adjustment rule is selected from a preset adjustment rule set based on the intermediate processing result. The adjustment rule set can be a knowledge base containing multiple condition-action pairs, where the condition part describes the specific linguistic phenomenon or missing information type, and the action part specifies the corresponding adjustment strategy, such as inserting a new module, skipping certain modules, changing the execution order of modules, or adjusting the parameters of modules. In step S36, subsequent language analysis modules are adjusted according to the adjustment rules. This means that the entire requirements analysis process is not reconstructed, but rather dynamically adjusted from the part of the process after the current module. For example, if serious ambiguity is found in the text, the adjustment rules may instruct the insertion of a ambiguity resolution module after the current module, or advance the semantic disambiguation module originally scheduled for a later stage. In step S37, based on the adjusted execution order, the requirements analysis process is updated and continues to be executed. This update is real-time and dynamic, ensuring that the analysis process can adaptively adjust according to the actual complexity of the text, thereby improving the accuracy and efficiency of the analysis.

[0075] This embodiment effectively addresses the limitations of traditional static processes when handling complex, ambiguous, or polysemous natural language requirement texts by introducing a real-time monitoring and dynamic adjustment mechanism for intermediate processing results during the requirements analysis process. Specifically, after the intermediate processing result of any language analysis module is acquired, the system determines whether specific linguistic phenomena or missing information exist. These phenomena or omissions signal potential problems in the current analysis path. Once such problems are identified, the system selects the most appropriate adjustment rule based on a pre-set set of adjustment rules. This rule guides the system to adjust subsequent language analysis modules, such as inserting new analysis modules to solve specific problems (e.g., pronoun resolution) or adjusting the priority of existing modules to better handle complex semantics. In this way, the requirements analysis process is no longer rigid but adaptively evolves according to the actual situation of the text, ensuring the depth and accuracy of the analysis. This significantly improves the robustness and adaptability of AI-based natural language requirements analysis methods and avoids analysis biases caused by initial judgment errors or unexpected text complexity, thereby effectively improving the quality and efficiency of requirements analysis.

[0076] In some preferred embodiments, a specific example is given below. Suppose a natural language requirement text is "I need a tool that can automatically identify cats and dogs in pictures and tell me what breeds they are." 1. Initially, based on text domain features (image recognition, animal recognition) and task features (recognition, classification), the system may construct an initial requirement analysis process, such as: lexical analysis -> syntactic analysis -> entity recognition (recognizing "cat", "dog", "breed") -> intent recognition (recognizing "recognize", "tell me") -> image recognition module call -> animal breed classification module call. 2. When the initial requirement analysis process reaches the entity recognition module, its intermediate processing results show that although "cat" and "dog" are recognized, for the word "breed," due to the complexity of the context, it cannot be clearly identified whether it is a modifier of "cat breed" or "dog breed," or it cannot be identified that "breed" itself is an entity that needs further detailed analysis. This is judged as a specific linguistic phenomenon of "information missing" or "unclear reference." 3. At this point, based on the intermediate processing result, the system selects a rule from a preset set of adjustment rules. This rule may indicate that when "breed" is identified and its modification relationship is unclear, a "semantic disambiguation module" or "reference resolution module" should be inserted into the subsequent process and placed before the "image recognition module call" to ensure that the user's specific requirements for "breed" can be accurately understood before image recognition and breed classification. 4. Based on this adjustment rule, the requirements analysis process is updated to: Lexical analysis -> Syntactic analysis -> Entity recognition -> Intent recognition -> Semantic disambiguation module -> Image recognition module call -> Animal breed classification module call. The system then continues execution according to this updated process, thereby more accurately understanding the user's specific requirements for "breed" and guiding subsequent image recognition and classification tasks.

[0077] In some preferred embodiments, step S36 includes:

[0078] S361. Identify multiple adjustment rules in a preset adjustment rule set that match a specific language phenomenon or information gap;

[0079] S362. For each matching adjustment rule, evaluate its degree of matching with the intermediate processing results and its potential impact on subsequent analysis processes.

[0080] S363. Sort the adjustment rules of multiple matches according to the degree of matching, potential impact and preset rule priority to obtain the sorting result;

[0081] S364. Select the optimal adjustment rule as the final adjustment rule based on the sorting results.

[0082] Step S361 refers to the fact that when the intermediate processing result is detected to contain a specific language phenomenon or missing information, the system does not simply select a preset adjustment rule, but first performs a comprehensive search in the preset adjustment rule set to identify all adjustment rules that match the current specific language phenomenon or missing information. These adjustment rules may differ in function, scope of application or processing strategy, but they all have the potential to solve or deal with the specific problem. Step S362 refers to the system's in-depth evaluation of each matching adjustment rule after identifying multiple matching adjustment rules. This evaluation includes two main aspects: first, assessing the degree of matching between the adjustment rule and the current intermediate processing result, i.e., the extent to which the rule can accurately resolve or handle the currently detected specific language phenomena or information gaps; second, assessing the potential impact of the adjustment rule on the subsequent analysis process, such as whether the introduction of the rule will lead to new ambiguities, increase computational complexity, affect the normal operation of other modules, or significantly improve the accuracy and efficiency of subsequent analysis. This embodiment can adopt the following approach: first, run a complete subsequent analysis process using preset data and record all key analysis results, model performance indicators, statistics, and visualizations to obtain the baseline results for the impact evaluation; then, run the subsequent analysis process after applying the adjustment rule using the same preset data and record all key analysis results, model performance indicators, statistics, and visualizations to obtain the actual running results of the adjustment rule; finally, compare the actual running results with the baseline results to evaluate the potential impact of the adjustment rule on the subsequent analysis process. Step S363 refers to the system ranking the rules after evaluating all matching adjustment rules, taking into account matching degree, potential impact, and preset rule priority. Adjustment rules with higher matching degree, smaller potential negative impact, and higher priority are ranked higher. In this embodiment, rule priority can be set based on expert experience, historical data, or preset strategies. For example, some rules may be given higher priority because they exhibit greater robustness or better performance in handling specific types of language phenomena. Step S364 refers to the system selecting the highest-ranked adjustment rule as the final adjustment rule after obtaining the ranking results. This optimal adjustment rule is considered the strategy that most effectively addresses specific language phenomena or information gaps in the current situation and has the best impact on subsequent requirements analysis processes.

[0083] This embodiment introduces a mechanism for identifying, evaluating, ranking, and selecting the optimal match adjustment rules when determining the execution order of subsequent language analysis modules. This multi-dimensional and refined rule selection process enables the system to avoid blind or suboptimal selection, ensuring that the selected adjustment rules best adapt to the current requirements analysis context, thereby improving the accuracy and effectiveness of dynamic adjustments. By comprehensively considering the degree of matching, potential impact, and rule priority, this application can more intelligently address complex linguistic phenomena and information gaps, avoiding the introduction of new problems due to inappropriate adjustments, and guaranteeing the stability and high-quality output of the requirements analysis process.

[0084] In some preferred embodiments, step S1 includes:

[0085] S11. Obtain natural language demand text and user personal information, including occupation type;

[0086] S12. Confirm the parameters of AI-based feature extraction technology based on the user's personal information;

[0087] S13. Use AI-based feature extraction technology to extract features from natural language requirement texts to obtain text domain features, user style features, and task features.

[0088] User personal information refers to non-textual content information related to the author of the request text, which may include the user's occupation type and historical behavioral data. Parameters of AI-based feature extraction technology refer to configuration items or weights that affect the operation and results of the AI-based feature extraction technology. These parameters can be model architecture selection, number of layers, number of nodes, learning rate, regularization coefficient, or weight factors for specific features. This embodiment can confirm the parameters of the AI-based feature extraction technology based on user personal information by querying a pre-built mapping table of personal information and feature extraction technology parameters. This embodiment can obtain more accurate and representative text domain features, user style features, and task features by dynamically adjusting the parameters of the AI-based feature extraction technology based on user personal information before feature extraction. This allows the extracted features to better reflect the expression habits of different users and different analysis task requirements. Therefore, this embodiment can effectively improve the accuracy and reliability of text domain features, user style features, and task features, thereby further improving the accuracy and reliability of the language analysis module configuration, and ultimately further improving the accuracy and reliability of the requirements analysis.

[0089] In some preferred embodiments, step S11 includes:

[0090] S111. Obtain natural language request text and user personal information;

[0091] S112. Preprocess the natural language requirement text.

[0092] Preprocessing refers to the process of cleaning, standardizing, and structuring raw natural language text. This embodiment can effectively remove noise and redundant information and standardize text format by preprocessing the natural language requirement text, thereby improving the data quality of the natural language requirement text input to AI-based feature extraction technology, and further improving the accuracy and reliability of the extracted text domain features, user style features, and task features. Preferably, the preprocessing in this embodiment includes stop word removal, word segmentation, synonym and dissident word processing, and part-of-speech tagging. Stop word removal refers to identifying and removing common functional words in the text, which can be achieved using a preset stop word list or a method based on statistical frequency. Word segmentation refers to dividing a continuous text sequence into word units with independent semantics, which can be achieved using methods based on dictionary matching, statistical models, or deep learning models. Synonym and dissident word processing refers to normalizing different expressions with the same or similar meanings and handling words with unclear or ambiguous referents, which can be achieved by constructing a thesaurus, using word vector models, or rule-based methods. Part-of-speech tagging refers to identifying the grammatical category of each word in a text, which can be achieved using rule-based, statistical model-based, or neural network model-based methods.

[0093] As can be seen from the above, the AI-based natural language requirements analysis method provided in this application can first obtain text domain features, user style features, and task features using AI-based feature extraction technology, and then dynamically determine the combination of language analysis modules, the parameters corresponding to each language analysis module in the combination, and the execution order of each language analysis module in the combination based on these features. This enables the construction of a requirements analysis process that is adapted to specific domains, user-personalized expressions, and analysis tasks. In other words, this application enables the analysis process to adapt to user-personalized expressions and accurately analyze different analysis tasks. Therefore, this application can effectively solve the problem that the requirements analysis results cannot accurately reflect the actual needs of users due to the insufficient adaptability of fixed and general analysis processes when facing differences in user style and diversity of analysis tasks, thereby effectively improving the accuracy and reliability of requirements analysis.

[0094] Secondly, such as Figure 2 As shown, this application also provides an AI-based natural language requirements analysis system, which includes:

[0095] Feature extraction module 1 is used to acquire natural language requirement text, and then use AI-based feature extraction technology to extract features from the natural language requirement text to obtain text domain features, user style features and task features;

[0096] Parameter confirmation module 2 is used to determine the combination of language analysis modules and the parameters corresponding to each language analysis module in the combination of language analysis modules based on text domain features, user style features and task features.

[0097] Analysis process confirmation module 3 is used to determine the execution order of each language analysis module in the language analysis module combination based on the text domain characteristics and task characteristics, and then construct the requirements analysis process based on the execution order and the language analysis module combination;

[0098] Module 4, the requirements analysis module, is used to perform requirements analysis on natural language requirements text using the requirements analysis process in order to obtain requirements information.

[0099] This application provides an AI-based natural language requirements analysis system, which includes a feature extraction module 1, a parameter confirmation module 2, an analysis process confirmation module 3, and a requirements analysis module 4. The AI-based natural language requirements analysis system provided in this embodiment is used to perform the steps in the AI-based natural language requirements analysis method provided in the first aspect above. The principle of the AI-based natural language requirements analysis system provided in this embodiment is the same as the principle of the AI-based natural language requirements analysis method provided in the first aspect above, and will not be discussed in detail here.

[0100] As can be seen from the above, the AI-based natural language requirements analysis method and system provided in this application can first obtain text domain features, user style features, and task features using AI-based feature extraction technology, and then dynamically determine the combination of language analysis modules, the parameters corresponding to each language analysis module in the combination, and the execution order of each language analysis module in the combination based on these features. This enables the construction of a requirements analysis process that is adapted to specific domains, user-personalized expressions, and analysis tasks. In other words, this application enables the analysis process to adapt to user-personalized expressions and accurately analyze different analysis tasks. Therefore, this application can effectively solve the problem that the requirements analysis results cannot accurately reflect the actual needs of users due to the insufficient adaptability of fixed and general analysis processes when facing differences in user style and diversity of analysis tasks, thereby effectively improving the accuracy and reliability of requirements analysis.

[0101] In the embodiments provided in this application, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of the above units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple units or components may be combined or integrated into another robot, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0102] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0103] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0104] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An AI-based natural language requirements analysis method, characterized in that, The AI-based natural language requirements analysis method includes the following steps: S1. Obtain the natural language requirement text, and then use AI-based feature extraction technology to extract features from the natural language requirement text to obtain text domain features, user style features, and task features; the AI-based feature extraction technology is a deep learning model or a pre-trained language model; the text domain features are the professional field or topic category to which the natural language requirement text belongs; the user style features are the language habits and preferences shown by the user when expressing requirements; the task features are the specific goals or focus of this requirement analysis; S2. Determine the language analysis module combination and the parameters corresponding to each language analysis module in the language analysis module combination based on the text domain features, the user style features and the task features; S3. Determine the execution order of each language analysis module in the language analysis module combination based on the text domain features and the task features, and then construct the requirements analysis process based on the execution order and the language analysis module combination. S4. Analyze the natural language requirement text using the requirement analysis process to obtain requirement information; Step S2 includes: S21. Obtain the confidence information of the text domain features, the user style features, and the task features; S22. Perform preliminary weighting processing on the text domain features, user style features and task features based on the confidence information, and perform secondary weighting processing on the features obtained from the preliminary weighting processing when there is cross-domain content, style mixing or multi-task requirements in the natural language requirement text to obtain weighted domain features, weighted style features and weighted type features. S23. Select a language analysis module from a preset language analysis module library based on the weighted domain features, the weighted style features, and the weighted type features to obtain a combination of language analysis modules; S24. Determine the parameters corresponding to each language analysis module in the language analysis module combination based on the weighted domain features, the weighted style features, and the weighted type features.

2. The AI-based natural language requirements analysis method according to claim 1, characterized in that, The confidence information includes the initial confidence information corresponding to the confidence information of the text domain features, the user style features, and the task features, as well as the correlation confidence between each feature. Step S22 includes: S221. Perform preliminary weighting processing on the text domain features, user style features, and task features according to the initial confidence level, the association confidence level, and the preset weighting rules to obtain preliminary domain features, preliminary style features, and preliminary type features; S222. Based on the preliminary domain features, preliminary style features, and preliminary type features, identify whether there is cross-domain content, style mixing, or multi-task requirements in the natural language requirement text. If yes, adjust the weighting rules according to the identification results and execute step S223. If no, use the preliminary domain features as weighted domain features, the preliminary style features as weighted style features, and the preliminary type features as weighted type features and execute step S23. S223. Perform secondary weighting on the preliminary domain features, preliminary style features, and preliminary type features according to the adjusted weighting rules to obtain weighted domain features, weighted style features, and weighted type features, and then execute step S23.

3. The AI-based natural language requirements analysis method according to claim 2, characterized in that, The step of identifying whether cross-domain content, mixed styles, or multi-task requirements exist in the natural language requirement text based on the preliminary domain features, the preliminary style features, and the preliminary type features includes: A1. Based on the preliminary domain characteristics and the hierarchical relationship in the preset domain knowledge graph, analyze the co-occurrence frequency and semantic association strength of terms in each domain in the natural language requirement text, so as to identify whether there is cross-domain content in the natural language requirement text and determine the boundary information between each domain. A2. Based on the preliminary style characteristics, analyze the distribution patterns of syntactic structure, vocabulary selection, sentiment tendency, and rhetorical devices in the natural language requirement text to identify whether style mixing exists in the natural language requirement text and determine the degree of mixing of each style. A3. Analyze the frequency of verbs, noun phrases and preset keywords in the natural language requirement text based on the preliminary type characteristics to identify whether there are multi-task requirements in the natural language requirement text and determine the priority of each task. The step of adjusting the weighting rule based on the recognition result includes: Based on the cross-domain content recognition results, style mixing recognition results, and multi-task requirement recognition results, and combined with a preset adjustment strategy, the weighting rules are adjusted. The adjustment strategy includes increasing, decreasing, or balancing the weights of the preliminary domain features, the preliminary style features, and / or the preliminary type features.

4. The AI-based natural language requirements analysis method according to claim 1, characterized in that, Step S23 includes: S231. Obtain the functional description, applicable scope and compatibility information of each module in the preset language analysis module library; S232. Based on the weighted domain features, the weighted style features, and the weighted type features, and in conjunction with the functional descriptions and applicable scopes of each module, a set of language analysis modules related to the current analysis needs is initially selected. S233. Based on the compatibility information between the modules in the initially selected language analysis module set, identify whether there are modules with overlapping or mutually exclusive functions in the module set, and perform conflict resolution and redundancy removal processing on the module set according to the identification results to obtain a language analysis module combination.

5. The AI-based natural language requirements analysis method according to claim 1, characterized in that, Step S3 includes: S31. Determine the initial execution order of each language analysis module in the language analysis module combination based on the text domain features and the task features; S32. Construct an initial requirements analysis process based on the initial execution order and the combination of the language analysis modules; S33. During the execution of the initial requirements analysis process, obtain the intermediate processing results of any language analysis module; S34. Determine whether the intermediate processing result contains specific language phenomena or missing information. The specific language phenomena or missing information indicate that the subsequent language analysis module needs to be adjusted. S35. When the intermediate processing result contains the specific language phenomenon or missing information, select the corresponding adjustment rule from the preset adjustment rule set according to the intermediate processing result. S36. Adjust the subsequent language analysis modules according to the adjustment rules, the adjustment including redetermining the execution order of the subsequent analysis modules; S37. Based on the adjusted execution order, update the requirements analysis process and continue executing the requirements analysis process; Step S31 specifically includes: The initial execution order of each language analysis module in the language analysis module combination is determined by querying a pre-built mapping table based on text domain features and task features. Step S32 specifically includes: The initial requirements analysis process is constructed by chaining all selected language analysis modules together in the initial execution order to form a complete processing chain.

6. The AI-based natural language requirements analysis method according to claim 5, characterized in that, Step S36 includes: S361. Identify multiple adjustment rules in the preset adjustment rule set that match the specific language phenomenon or information loss; S362. For each matching adjustment rule, evaluate its degree of matching with the intermediate processing results and its potential impact on subsequent analysis processes. S363. Sort the adjustment rules of the multiple matches according to the matching degree, the potential impact and the preset rule priority to obtain the sorting result; S364. Select the optimal adjustment rule as the final adjustment rule based on the sorting results.

7. The AI-based natural language requirements analysis method according to claim 1, characterized in that, Step S1 includes: S11. Obtain natural language demand text and user personal information, wherein the user personal information includes occupation type; S12. Confirm the parameters of the AI-based feature extraction technology based on the user's personal information; S13. Use the AI-based feature extraction technology to extract features from the natural language requirement text to obtain text domain features, user style features, and task features.

8. The AI-based natural language requirements analysis method according to claim 7, characterized in that, Step S11 includes: S111. Obtain natural language request text and user personal information; S112. Preprocess the natural language requirement text.

9. An AI-based natural language requirements analysis system, characterized in that, The AI-based natural language requirements analysis system is used to perform the steps of the AI-based natural language requirements analysis method as described in any one of claims 1-8, wherein the AI-based natural language requirements analysis system includes: The feature extraction module is used to acquire natural language requirement text, and then use AI-based feature extraction technology to extract features from the natural language requirement text to obtain text domain features, user style features and task features; The parameter confirmation module is used to determine the combination of language analysis modules and the parameters corresponding to each language analysis module in the combination of language analysis modules based on the text domain features, the user style features and the task features. The analysis process confirmation module is used to determine the execution order of each language analysis module in the language analysis module combination based on the text domain features and the task features, and then construct the requirements analysis process based on the execution order and the language analysis module combination. The requirements analysis module is used to perform requirements analysis on the natural language requirements text using the requirements analysis process to obtain requirements information.

Images