Software development collaborative interaction method

Abstract

The invention relates to the technical field of software development, and discloses a software development collaborative interaction method, which comprises the following steps of: deploying a user behavior monitoring component to acquire multi-user group operation sequence data and converting the data into a structured demand item; constructing a demand evolution graph to record a historical demand evolution path; establishing a multi-subject collaborative prediction system integrated multi-user group local prediction model; a distributed verification system is implemented to verify and correct the prediction result; and generating collaborative development tasks and priority suggestions based on the verification result. According to the method, accurate capture and prediction of multi-user group demands are realized, the software development efficiency and quality are improved, and prospective planning and resource optimization allocation of a development team are supported.

Description

Technical Field

[0001] The present invention relates to the technical field of software development, and more particularly, to a collaborative interaction method for software development, which is used for capturing, predicting and planning requirements of multiple user groups. Background Art

[0002] In large-scale software development projects, accurately capturing and predicting the needs of multiple user groups is a key factor in ensuring project success. Traditional requirements acquisition methods rely primarily on static documents and user interviews, which are typically conducted at specific points in the development cycle and cannot continuously capture dynamic changes in user behavior. In addition, users' implicit needs are often difficult to capture. Existing technologies typically use a single data source and model for requirements analysis, making it difficult to effectively integrate scattered insights from multiple user groups. Existing requirements management methods lack the ability to predict future trends in demand evolution. Although the demand evolution paths of diverse user groups are related, they each have their own characteristics, making them difficult to handle in a unified manner. These technical challenges lead to frequent demand changes and rework during the software development process, seriously affecting development efficiency and product quality. Summary of the Invention

[0003] The present invention provides a software development collaborative interaction method to solve the technical problems in related technologies of being unable to accurately capture the needs of multiple user groups and lacking the ability to predict the evolution of needs.

[0004] The present invention discloses a software development collaborative interaction method, including: deploying a user behavior monitoring component, collecting operation sequence data of multiple user groups, and converting the user behavior data into structured demand items through semantic inference technology; constructing a demand evolution graph model, recording the historical demand evolution path of each user group, the demand evolution graph model is a directed graph structure, wherein nodes represent structured demand items, and edges represent the evolutionary relationship between demand items; establishing a multi-agent collaborative prediction system, integrating the local prediction models of multiple user groups, and generating a global demand prediction result, wherein the local prediction model of each user group is fused with the global demand evolution graph through dynamic weights; implementing a distributed verification system, verifying and correcting the prediction results through actual operation behavior feedback of each user group, forming a closed-loop optimization of demand prediction and verification; based on the verified prediction results, generating structured collaborative development tasks and priority suggestions, supporting the software development team to carry out forward-looking planning and resource allocation.

[0005] Furthermore, the step of deploying a user behavior monitoring component to collect operation sequence data of multiple user groups includes: embedding low-interference operation data collection probes in different functional modules of the software application, and the probes collect user interaction events and their context information; cleaning, session division and feature extraction of the collected original operation data to form a standardized operation sequence; classifying the standardized operation sequence according to user groups and storing it in a distributed database.

[0006] Furthermore, the step of constructing a demand evolution graph model includes: performing time series analysis on the historical demand data of each user group to identify the evolutionary correlation between demand items; constructing a weighted directed graph structure, in which the node weight represents the frequency of occurrence of the demand item and the edge weight represents the strength of the evolutionary relationship; applying a graph traversal algorithm in the weighted directed graph structure to extract high-frequency evolution paths and key branch points.

[0007] Furthermore, the steps of establishing a multi-agent collaborative forecasting system include: establishing a local forecasting model for each user group, which integrates a graph neural network and a sequence forecasting algorithm; setting a dynamic weight allocation mechanism to automatically adjust the weight according to the historical forecast accuracy of each user group and the relevance of the current task; and generating a global demand forecast by fusing the results of each local forecasting model through a weighted voting algorithm.

[0008] Furthermore, the steps of implementing the distributed verification system include: converting the predicted demand items into a set of observable expected user behavior hypotheses; deploying distributed verification probes to collect users' actual operating behaviors in specific scenarios; calculating the matching degree between the actual operating behaviors and the expected behavior hypotheses, and quantitatively evaluating and correcting the prediction results.

[0009] Furthermore, the step of generating structured collaborative development tasks based on the verified prediction results includes: converting the verified demand prediction results into functional development items; prioritizing the functional development items based on dependency analysis and resource constraints; generating development task cards containing task descriptions, technical dependencies, resource requirements and time estimates, and providing them to the development team through a visual interface.

[0010] Furthermore, the following steps are included: establishing a continuous feedback mechanism to collect and analyze feedback information from the development team during the execution of tasks; dynamically adjusting and optimizing the prediction model and verification system based on feedback information and actual execution status; and generating collaborative interaction history records to provide a reference basis for demand forecasting in similar scenarios in the future.

[0011] Furthermore, in the step of collecting operation sequence data of multiple user groups: the user groups include end users, business analysts, developers and testers; according to the characteristics of different user groups, specific data collection strategies and feature extraction methods are designed respectively.

[0012] The present invention solves the technical problems of the inability to accurately capture the needs of multiple user groups and the lack of demand evolution prediction capabilities in software development by deploying user behavior monitoring components, constructing a demand evolution graph model, establishing a multi-agent collaborative prediction system and implementing a distributed verification system, and has achieved significant technical effects: First, it breaks through the traditional demand acquisition method's reliance on static documents and user interviews, and realizes dynamic demand capture based on actual user behavior, especially the effective identification of implicit needs; second, through the innovative demand evolution graph model and sequence prediction algorithm, it realizes accurate prediction of future demand evolution trends, providing forward-looking planning support for software development teams; third, the innovative multi-agent collaborative prediction system effectively integrates the demand insights of different user groups while maintaining sensitivity to the characteristics of each group, thereby improving the accuracy and comprehensiveness of the overall prediction; finally, the closed-loop distributed verification mechanism ensures the reliability of the prediction results and supports the continuous optimization of the prediction model, significantly improving software development efficiency and product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is an overall flow chart of the software development collaborative interaction method of the present invention; Figure 2 It is a flow chart of the deployment of the user behavior monitoring component of the present invention; Figure 3 It is a flow chart of the demand evolution graph model construction of the present invention; Figure 4 This is a flow chart of the multi-agent collaborative prediction system established in the present invention; Figure 5 is a flow chart of an implementation of the distributed verification system of the present invention; Figure 6 It is a flow chart of structured collaborative development task generation according to the present invention. DETAILED DESCRIPTION

[0014] Before describing the present application in detail, in order to facilitate understanding of the technical solution of the present application, the following is an explanation of the terms used in the present application: User behavior monitoring component: refers to a program module installed in a software application to collect user operation behavior data. It is low-intrusive and does not affect the user's normal usage experience.

[0015] Behavior-intent mapping model: refers to an algorithmic model that converts a user's operation behavior sequence into corresponding user intentions and needs, and infers their potential needs by analyzing the user's operation patterns.

[0016] Demand evolution graph: refers to a directed graph structure that describes the evolution relationship between demand items, where nodes represent specific demand items and edges represent the evolution relationship between demands.

[0017] Sequence prediction algorithm: refers to a computational method that predicts possible future demand items based on historical demand sequences and contextual information.

[0018] Multi-agent collaborative prediction: refers to a method that integrates the local prediction models of multiple user groups and generates comprehensive prediction results through dynamic weight adjustment.

[0019] Distributed verification mechanism: refers to a method of verifying and correcting prediction results by utilizing the actual operational behaviors of multiple user groups.

[0020] Structured requirement item: refers to a standardized requirement description, which contains a data structure with attributes such as requirement type, priority, and association relationship.

[0021] In response to the above technical challenges, this application provides a collaborative interaction method for software development, which integrates innovative technologies such as user behavior monitoring, demand evolution graph modeling, multi-agent collaborative prediction and distributed verification to form a complete software demand capture, prediction and planning solution.

[0022] Specifically, this application deploys a low-intrusion user behavior monitoring component to collect actual operational sequence data from multiple user groups during software usage. It then uses semantic inference technology to build a behavior-intent mapping model, converting user behavior data into structured requirements in real time. This demand capture method based on actual behavior data breaks through the traditional demand acquisition method's reliance on static documents and user interviews, and can more accurately reflect users' real needs, especially implicit needs.

[0023] This application further constructs a demand evolution graph model to record the evolution path of historical demand for each user group and applies an innovative sequence prediction algorithm to predict future demand. This algorithm comprehensively considers the demand evolution graph structure, historical demand sequences, and product context information to generate demand forecast results with time series characteristics.

[0024] Furthermore, this application innovatively establishes a multi-agent collaborative forecasting system that integrates local forecasting models from different user groups and assigns them dynamic weights to generate global forecast results. This system simultaneously maintains sensitivity to the needs of each group while also understanding common needs, effectively addressing the technical challenges of addressing the needs of diverse user groups.

[0025] This application also implements a distributed verification system to verify and correct the prediction results through the actual operational behavior feedback of each user group, forming a closed-loop optimization of demand prediction and verification to ensure the accuracy and reliability of the prediction results.

[0026] Ultimately, this application generates structured collaborative development tasks and priority recommendations based on prediction results and verification feedback, providing data-driven decision support for software development teams and enabling forward-looking planning and optimal resource allocation.

[0027] The software development collaborative interaction method proposed in this application is applicable to various large-scale software development projects, especially those involving multiple user groups and complex and changing requirements. The following describes a typical application scenario of this application.

[0028] In enterprise-level software development environments, such as those used in large-scale software projects like enterprise resource planning (ERP) systems and customer relationship management (CRM) systems, multiple user groups exist, including management-level users, business department users, and technical support personnel. These user groups have different functional requirements and usage preferences for the software system. The proposed method can be deployed in such environments to monitor the actual operational behavior of each user group, predict the evolution of their needs, and provide forward-looking development guidance to software development teams.

[0029] In the field of internet product development, such as social media platforms, e-commerce systems, and online education applications, user groups are becoming more diverse and their needs are changing more rapidly. This application method can effectively capture the operational behavior data of different user groups (such as users of different age groups, different regions, and different occupations), analyze their implicit needs, predict the evolution of future functional requirements, and support product teams in rapid iteration and innovation.

[0030] In open source software communities, contributors and users are distributed globally, making requirements collection and integration even more challenging. This application method can monitor user behavior worldwide, identify common and specific needs among users from different regions and cultural backgrounds, provide requirement prioritization recommendations to the open source community, and promote global collaborative development.

[0031] The deployment environment of this application method mainly includes two parts: Client environment: Deploy user behavior monitoring components on the client side of the software application to collect user action sequence data. These components must be low-intrusive, not impacting the user's experience, while ensuring data privacy and security.

[0032] Server environment: Deploy data processing, model training, and predictive analysis systems on the server side. The server environment needs to have sufficient computing and storage capabilities to support the real-time processing and analysis of large-scale user behavior data, as well as the training and operation of complex algorithm models.

[0033] By deploying this application method in the above application scenarios and environments, software development teams can obtain more accurate insights and predictions of user needs, reduce development rework, and improve development efficiency and product quality.

[0034] The technical solution of the present application is described in detail below in conjunction with the accompanying drawings and embodiments. It should be noted that the following embodiments are only used to illustrate the present application and are not used to limit the scope of the present application.

[0035] According to one embodiment of the present application, the specific steps of the software development collaborative interaction method are as follows: Step S01: Deploy user behavior monitoring components, collect operation sequence data of multiple user groups, and build a behavior-intent mapping model to capture structured demand items in real time.

[0036] It should be understood that in this step, a low-intrusion user behavior monitoring component must first be deployed in the target software application. This monitoring component has the following characteristics: Step S01-1: embed a data collection probe in each functional module of the software application to collect user operation sequence data.

[0037] The operation sequence data here includes but is not limited to: user click behavior, input content, page dwell time, function usage frequency, operation path, number of error attempts, etc. The data collection probe adopts a lightweight design, with an impact of less than 3% on software performance, ensuring that it does not interfere with the normal user experience.

[0038] In some implementations, data collection probes can be implemented in different ways depending on the application environment. For example, in web applications, user interaction events can be captured using JavaScript event listeners; in mobile applications, this can be achieved through the behavior tracking interface provided by the native SDK; and in desktop applications, this can be achieved through UI component event hooks. Furthermore, the probe can optionally automatically adjust the data collection granularity based on user privacy settings, collecting sufficient behavioral data while protecting user privacy.

[0039] In the enterprise resource planning (ERP) system scenario, data collection probes can focus on monitoring users' operation paths and pause points in key business processes. For example, the system may discover that users in the finance department frequently switch between multiple menus when generating monthly reports, suggesting that the current report generation process is not convenient enough and there is room for optimization.

[0040] Step S01-2: Preprocess the collected operation sequence data to form a standardized user behavior data set.

[0041] Preprocessing includes steps such as data cleaning, session segmentation, and feature extraction. Data cleaning removes abnormal operations and noisy data; session segmentation divides continuous operation sequences into meaningful operation sessions; and feature extraction extracts key features from the original operation sequence, such as operation type, target object, and timestamp. Through these processes, the unstructured raw operation data is converted into a structured behavioral feature matrix.

[0042] In some implementations, preprocessing may also include data enhancement and noise reduction. For example, for sparse user behavior data, data enhancement methods based on similar user groups can be used to increase data density and representativeness. Optionally, time window sliding techniques can be applied to capture temporal characteristics and periodic patterns of user behavior.

[0043] In a customer relationship management (CRM) system, the pre-processing phase can identify typical salesperson behavioral patterns during customer follow-up. For example, the system might discover that high-performing salespeople typically review more complete customer history information before a customer visit and update customer status information more quickly after the visit. These patterns can be retained and reinforced as valuable behavioral characteristics.

[0044] Step S01-3: Use semantic inference technology to build a behavior-intent mapping model to convert user behavior data into structured demand items.

[0045] According to the embodiment of the present application, the model is based on a sequence-to-sequence architecture, with the input being the user operation sequence features and the output being the corresponding demand intention representation. The core of the model consists of an encoder and a decoder: The encoder uses a bidirectional long short-term memory network (Bi-LSTM) to process the user operation sequence and encode it into a context vector. The specific algorithm is as follows: in, Represents the time step The operational eigenvector of Indicates hidden state.

[0046] The decoder generates a structured demand representation based on the attention mechanism: in, is the context vector, is the attention weight, is the hidden state of the decoder at the previous time step, The encoder The hidden state of time steps, Function is used to calculate similarity.

[0047] In some implementations, different types of attention mechanisms can be employed. For example, multi-head attention can be used to improve the model's ability to perceive different patterns. Alternatively, a self-attention mechanism can be introduced to capture long-range dependencies within user operation sequences. Furthermore, to enhance the model's generalization capabilities, transfer learning techniques can be combined to extract more general operational semantic representations using large-scale pre-trained models.

[0048] In online education platforms, the behavior-intention mapping model can analyze students' course browsing and learning behaviors to infer their potential learning needs. For example, if the system observes a student repeatedly reviewing different explanations of the same knowledge point and spending a long time on related exercises, the model can infer that the student has difficulty understanding this knowledge point and thus generate a structured requirement item, such as "needing more detailed explanations of this knowledge point and targeted exercises."

[0049] Through this model, the system can map user action sequences into structured requirement items. Each requirement item contains attributes such as requirement type, description, priority, and associated objects. For example, if a user repeatedly attempts to find a feature within a functional module but fails, the model can infer the user's need for that feature and generate a corresponding structured requirement item.

[0050] Step S02: Construct a demand evolution graph model, apply a sequence prediction algorithm, and record the historical demand evolution path of each user group.

[0051] According to an embodiment of the present application, this step creates a new demand evolution graph model for recording and analyzing the evolution path of historical demands of each user group and predicting future demands based on this.

[0052] Step S02-1: Based on the structured requirement items obtained from step S01, a requirement evolution graph model is constructed.

[0053] The demand evolution graph is a directed graph structure. Indicates that It is a set of nodes, each node represents a structured requirement item; is a set of edges, representing the evolutionary relationship between demand items; is a set of edge weights, representing the strength of the evolutionary relationship. The following algorithm is used in the construction process: For each newly identified requirement , create the corresponding node Join middle.

[0054] For temporally adjacent demand items and , if there is an evolutionary relationship, then Add directed edges to ,from point to .

[0055] Calculate edge weights based on the frequency and co-occurrence patterns of evolutionary relationships Join middle.

[0056] In some implementations, the requirements evolution graph structure can be expanded. For example, attributes can be added to nodes and edges in the graph. Node attributes include requirement type, source user group, and urgency; edge attributes include evolution type (e.g., refinement, generalization, substitution), time interval, etc. Optionally, a multi-level graph structure can be constructed, with different levels representing requirements at different levels of abstraction, to capture the hierarchical relationships between requirements.

[0057] In the context of e-commerce platforms, demand evolution maps can illustrate the evolution of user needs related to the shopping experience. For example, the system might discover that users' initial demand for "simplified payment processes" gradually evolves over time into more detailed and advanced requirements such as "multiple payment method integration," "one-click payment," and "biometric payment." These evolutionary paths provide clear direction for platform feature iteration.

[0058] Step S02-2: Apply time series data mining technology to extract evolution patterns from the demand evolution graph.

[0059] Use the frequent subgraph mining algorithm to identify common evolution patterns from the demand evolution graph. This algorithm can effectively discover recurring demand evolution paths that represent the typical development patterns of user needs. The core steps of the algorithm include: Set the minimum support threshold .

[0060] Starting from a single-node subgraph, by gradually expanding the subgraph scale, we can mine all nodes with support not less than Frequent subgraph patterns of .

[0061] The frequent subgraph patterns mined are screened and merged to form the final set of evolutionary patterns.

[0062] In some implementations, the frequent subgraph mining algorithm can be further optimized. For example, a sampling-based approximation algorithm can be applied to improve computational efficiency while ensuring quality results. Optionally, constraints, such as time constraints and structural constraints, can be introduced to extract evolutionary patterns that better meet specific business needs. Furthermore, the system can employ an incremental mining strategy, dynamically updating identified evolutionary patterns as new demand data is continuously added.

[0063] In social media application scenarios, time series data mining can extract patterns in the evolution of user preferences for social features from demand evolution graphs. For example, the system might identify that the demand for "photo sharing" typically evolves into a series of related needs, such as "photo editing," "photo classification," and "smart albums." The satisfaction of these needs, in turn, triggers demands for new forms of social interaction, such as "photo-based topic discussions." Identifying these typical evolutionary patterns helps product teams predict the appropriate path for feature iteration.

[0064] Step S02-3: Apply sequence prediction algorithm to predict possible future demand based on demand evolution graph and historical demand sequence.

[0065] This step uses an innovative sequence prediction algorithm that comprehensively considers the demand evolution graph structure, historical demand sequences, and product context information to predict future demand. The algorithm expression is as follows: in, Represents a given historical demand sequence Condition, the next requirement The probability of occurrence; It is the demand evolution map; is the representation vector of the historical demand sequence; It is product context information, including product life cycle stage, market environment characteristics, etc.

[0066] function This is achieved by integrating the graph convolutional network (GCN) and the long short-term memory network (LSTM). The specific calculation process is as follows: Use GCN to extract graph features: Use LSTM to encode historical sequences: Combine context features and the results of the first two steps to predict the probability of the next demand item: in, and are model parameters, Represents a vector concatenation operation.

[0067] In some implementations, the sequence prediction algorithm can have multiple variations. For example, a Graph Attention Network (GAT) can be used instead of a GCN to better capture important information in the graph structure. Alternatively, a Gated Recurrent Unit (GRU) or Transformer can be used instead of an LSTM to encode the historical sequence. Furthermore, to improve the interpretability of the prediction, attention visualization techniques can be introduced to intuitively display the key historical requirements and graph features that influence the prediction results.

[0068] In smart home applications, sequence prediction algorithms can predict future smart home control needs based on a user's historical needs. For example, after observing that a user has requested "smart lighting control," "timed temperature adjustment," and "remote device monitoring," the system can predict that the user is likely to require advanced features like "scene linkage" or "intelligent voice control" next, enabling the development team to plan the development of these features in advance.

[0069] Through the above algorithm, the system can establish an exclusive demand evolution map model for each user group, record its historical demand evolution path, and predict possible demand items in the future.

[0070] Step S03: Establish a multi-agent collaborative forecasting system, integrate the local forecasting models of different user groups, and generate global demand forecast results.

[0071] According to the embodiment of the present application, this step creates an innovative multi-agent collaborative prediction system, which forms a global prediction result by integrating local prediction models of different user groups and assigning dynamic weights.

[0072] Step S03-1: Parameterize the local prediction model of each user group.

[0073] For each user group (common groups), and use the sequence prediction algorithm in step S02 to build a local prediction model ,in is the historical demand data of this group, is a contextual characteristic of the group.

[0074] In some implementations, local prediction models can employ different structures and training strategies. For example, for user groups with abundant data, more complex deep learning models can be used; for groups with sparse data, lighter models or data augmentation techniques can be selected. Alternatively, specific loss functions can be designed for local models based on the characteristics of different user groups to optimize specific prediction tasks.

[0075] In multilingual software localization scenarios, local prediction models for each language user group can capture the unique needs of that language user group. For example, a model for an Asian language user group might focus more on text input methods and display adaptation, while a model for a European language user group might focus more on date formats and cultural adaptability. By parameterizing these characteristics, the system can comprehensively account for these differences in subsequent steps.

[0076] Step S03-2: Design a dynamic weight allocation algorithm to assign appropriate weights to the local prediction models of each user group based on the characteristics of the prediction task and historical prediction accuracy.

[0077] The dynamic weight allocation algorithm calculates weights based on the following three factors: Historical prediction accuracy : Measures the historical forecast accuracy of the local model for this group.

[0078] Correlation coefficient : Measures the relevance of the group to the current prediction task.

[0079] Diversity coefficient : Ensure the diversity of prediction results and prevent a single group from dominating the overall prediction.

[0080] The weight calculation formula is as follows: in, is the final weight after normalization.

[0081] In some implementations, the dynamic weighting algorithm can incorporate a time-series adjustment factor. For example, recent trends in forecast accuracy can be considered to assign higher weights to user groups with consistently improving forecasting capabilities. Alternatively, a rare demand discovery factor can be incorporated into the weighting calculation, appropriately increasing the weights of groups that excel at discovering emerging and rare demands to enhance the overall forecast's foresight.

[0082] In medical software development scenarios, dynamic weighting algorithms can effectively balance the predictive influence of doctors, nurses, and management teams. For example, when forecasting demand for diagnosis-related functions, the system might assign higher weight to the local model of the doctors; while for ward management-related functions, it might assign higher weight to the models of the nurses and management teams. This dynamic adjustment ensures that the prediction results are targeted to each functional area.

[0083] Step S03-3: Implement a multi-agent collaborative forecasting algorithm, integrate the local forecasting results of each user group, and generate a global demand forecasting result.

[0084] The multi-agent collaborative prediction algorithm performs weighted fusion of the local prediction results of each user group, while taking into account the constraints of the global demand evolution map to generate the final global prediction result. The core formula of the algorithm is as follows: in, Represents the global prediction result, that is, the next demand item The probability distribution of occurrence; It is Dynamic weight of each user group; It is Local prediction results of a group; It is the current global demand evolution map; It is a fusion function implemented using the attention mechanism, where attention is allocated based on the consistency between the prediction results and the global graph.

[0085] In some implementations, the collaborative prediction algorithm can employ a multi-stage fusion strategy. For example, it can first perform local fusion on user groups with similar functions, followed by a higher-level global fusion across functional groups. Optionally, an uncertainty estimation module can be introduced to assign a confidence score to each prediction result for subsequent decision-making. Furthermore, the system can employ ensemble learning techniques, such as boosting or stacking, to further enhance fusion effectiveness.

[0086] In enterprise collaborative software development scenarios, multi-agent collaborative prediction algorithms can integrate demand forecasts from users across different departments. For example, when predicting the evolution of document collaboration features, the system comprehensively considers the marketing department's needs for external sharing capabilities, the R&D department's needs for version control, and the legal department's needs for permissions management. This creates a comprehensive prediction that balances the interests of all parties and guides the priorities and direction of feature development.

[0087] Through this multi-agent collaborative forecasting system, the system can effectively integrate the demand insights of different user groups while maintaining sensitivity to the characteristics of each group, thereby improving the accuracy and comprehensiveness of the overall forecast.

[0088] Step S04: Implement a distributed verification system to verify and correct the prediction results through feedback from the actual operation behavior of each user group, forming a closed-loop optimization.

[0089] According to the embodiment of the present application, this step establishes an innovative distributed verification system to verify and correct the prediction results through actual user behavior to ensure the accuracy and reliability of the prediction results.

[0090] Step S04-1: Convert the predicted demand items into observable behavioral hypotheses.

[0091] The system converts the demand forecast results from step S03 into a series of observable user behavior hypotheses. For example, if the forecast results indicate that user group A needs a new feature, the corresponding behavior hypothesis might be "user group A will try to find and use this feature." These behavior hypotheses are stored in a structured format and contain the following attributes: Forecast demand item ID; A description of the corresponding expected user behavior; Observation window time setting; Verify threshold conditions; In some implementations, the behavioral hypothesis generation process can utilize template-based methods and natural language processing techniques. For example, the system can establish a template library that maps requirement types to behavioral patterns, automatically generating corresponding behavioral hypotheses for different types of requirements. Optionally, contextual factors can be introduced to generate differentiated behavioral hypotheses for the same requirement in different usage scenarios.

[0092] In game development scenarios, if predictions indicate players may need more social interaction features, the system will generate multiple observable behavioral hypotheses, such as "players will frequently attempt to send messages to other players" or "players will seek out team or guild features in the game." These hypotheses have clear observation conditions and verification criteria, making them easy to verify later using actual player behavior data.

[0093] Step S04-2: Collect actual user operation behavior data through the user behavior monitoring component to verify the hypothesis.

[0094] The system continues to collect actual user behavior data within the set observation window through the user behavior monitoring component deployed in step S01. This data is used to verify the behavior hypothesis generated in step S04-1. The verification process uses the hypothesis testing method to calculate the degree of conformity between the actual observed data and the behavior hypothesis: in, Indicates the A behavioral hypothesis, represents the observation dataset, Representation and Assumptions The amount of observation data matched, Represents the total amount of observation data.

[0095] In some implementations, the hypothesis verification process can employ a multi-level matching strategy. For example, strict and loose matching criteria can be defined, with compliance calculated at different levels of precision. Optionally, a time-weighting factor can be introduced to give a higher weight to observations from the most recent time period, reflecting the timeliness of user needs. Furthermore, the system can employ A / B testing, providing a subset of users with prototypes of hypothetical functions and observing their feedback to further validate the accuracy of demand forecasts.

[0096] In financial software development scenarios, the system might predict that users have a new need for transaction data visualization. To verify this hypothesis, the system observes whether users frequently attempt to export data to external analysis tools, whether they spend extended time on data pages, and whether they search for related features. By statistically analyzing this observational data, the system can objectively assess the accuracy of this prediction, providing a basis for subsequent development decisions.

[0097] Step S04-3: Based on the verification results, the prediction model parameters are adjusted in real time to form a closed-loop optimization.

[0098] The system adjusts the prediction model parameters in real time based on the verification results to achieve closed-loop optimization of the model. The adjustment process uses an online learning algorithm to dynamically correct the model parameters based on the verification deviation: in, represents the current model parameters, is the learning rate, is the gradient of the loss function with respect to the parameters, is the verification compliance vector of each hypothesis, is the expected target vector.

[0099] In some implementations, the parameter adjustment strategy can employ an adaptive learning rate. For example, the learning rate can be dynamically adjusted based on the stability of the verification results, using a smaller learning rate when the verification results fluctuate significantly and a larger learning rate when they do not. Optionally, a regularization term can be introduced to prevent the model from overfitting to the behavioral patterns of specific user groups. Furthermore, the system can employ a memory-enhanced learning algorithm to specifically focus on and memorize cases where predictions were incorrect, improving the model's adaptability to unpredictable situations.

[0100] In a cloud service platform development scenario, the system, through distributed verification, found that while demand forecasts for the "Automatic Resource Scaling" feature were highly accurate, forecasts for the "Multi-Region Data Synchronization" feature exhibited significant deviations. Based on this verification, the system automatically adjusts forecast model parameters to better learn user behavior patterns within specific technical areas of the cloud service and potentially downweights irrelevant features, thereby improving accuracy in the next round of forecasts.

[0101] Through this distributed verification system, the system can continuously verify the accuracy of the prediction results and dynamically adjust the prediction model based on actual user behavior feedback, forming a closed-loop optimization and improving the prediction accuracy.

[0102] Step S05: Based on the prediction results and verification feedback, generate structured collaborative development tasks and priority recommendations to support the software development team in forward-looking planning and resource allocation.

[0103] According to an embodiment of the present application, this step converts the output of the previous steps into actionable collaborative development tasks and priority recommendations, providing decision support for the software development team.

[0104] Step S05-1: Convert the verified demand forecast results into structured development tasks.

[0105] The system converts the verified demand forecast results into structured software development tasks. The conversion process includes the following steps: Requirements classification and aggregation: Aggregate similar or related requirements into functional groups.

[0106] Task decomposition: Decompose the functional group into specific and implementable development tasks.

[0107] Dependency analysis: Identify dependencies between tasks and build a task dependency graph.

[0108] Each development task contains the following attributes: task ID, task description, associated requirements, required skill tags, estimated workload, pre-dependencies, etc.

[0109] In some implementations, the task conversion process can be assisted by template libraries and knowledge graphs. For example, a library mapping requirement types to development task templates can be established to quickly generate standardized task descriptions. Optionally, historical project data analysis can be incorporated to provide more accurate workload estimates and dependency identification based on historical development records of similar requirements. Furthermore, the system can employ natural language generation technology to convert machine-readable requirement representations into human-understandable task descriptions, improving communication efficiency.

[0110] In collaborative office software development scenarios, if the system predicts and verifies user demand for real-time collaborative editing, it translates this demand into multiple structured development tasks, such as implementing a conflict detection algorithm for document changes, developing a real-time collaborative status synchronization mechanism, and building a user editing permission control module. Each task includes a clear description, skill requirements, workload estimates, and dependencies, enabling the development team to clearly understand and effectively plan implementation steps.

[0111] Step S05-2: Generate task priority recommendations using a multi-objective optimization algorithm.

[0112] The system uses a multi-objective optimization algorithm to generate task priority recommendations, taking into account the following objectives: Demand urgency: The urgency of demand determined based on forecast results and verification feedback.

[0113] Implementation complexity: The technical complexity and effort estimate of the development task.

[0114] Resource constraints: Available development resources and skill distribution.

[0115] Dependency: Predependencies between tasks.

[0116] Value assessment: The value that is expected to be brought to users after the task is completed.

[0117] The formal expression of the multi-objective optimization problem is as follows: in, is a vector composed of multiple objective functions, and is a constraint, is the feasible solution space. By solving it with the Pareto optimization method, a set of non-dominated solutions are obtained as priority recommended options.

[0118] In some implementations, multi-objective optimization algorithms can employ different weighting strategies. For example, different objective weighting schemes can be set for different business stages, such as emphasizing user value goals in the early stages of a product, and focusing more on resource efficiency goals during the stabilization phase. Alternatively, interactive optimization methods can be introduced to allow the development team to fine-tune the initial priority recommendations provided by the algorithm, combining human experience and algorithmic intelligence. Furthermore, the system can employ reinforcement learning methods to continuously optimize objective weights and constraint settings by recording the actual execution results of historical task plans.

[0119] In mobile payment app development scenarios, multi-objective optimization algorithms balance security, user experience, and performance requirements to generate recommended task priorities. For example, while the "biometric payment" feature may have a high user value assessment, the system will consider its implementation complexity and security risks and may prioritize the "payment security framework upgrade" task to lay the foundation for advanced features. Meanwhile, tasks like "payment result animation optimization," which have few dependencies, are simple to implement, but can significantly improve the user experience, may be prioritized for rapid implementation in short iterations.

[0120] Step S05-3: Generate resource allocation recommendations and forward-looking planning reports.

[0121] Based on the priority recommendations in step S05-2, the system further generates detailed resource allocation recommendations and forward-looking planning reports, including: Human resource allocation plan: Propose the optimal personnel allocation plan based on the skills required for the task and the current team skill distribution.

[0122] Iteration plan suggestions: Organize tasks into multiple iteration cycles and set reasonable delivery nodes.

[0123] Risk warning: Identify possible development risk points and propose preventive measures.

[0124] Forward-looking technology reserve recommendations: Based on long-term demand forecasts, propose technology research directions that need to be planned in advance.

[0125] In some implementations, planning report generation can incorporate a variety of visualization and interactive methods. For example, Gantt charts, dependency network diagrams, and other visualizations can be used to intuitively display planning results. Optionally, scenario simulations can be provided, allowing teams to explore the impact of different resource allocations or priority adjustments on the overall development progress. Furthermore, the system can implement real-time monitoring and dynamic adjustment mechanisms, automatically updating planning recommendations based on the actual development progress to ensure planning timeliness.

[0126] In IoT platform development scenarios, forward-looking planning reports will identify the "device auto-discovery and configuration" feature area, which is likely to become a hot topic in the future, based on demand forecasts. They will recommend that teams conduct relevant technical research and talent development in advance. Furthermore, for the near-term development of "multi-protocol device access," the system will analyze the current team's skill set and may recommend bringing in external experts with specialized knowledge of specific communication protocols or arranging technical training for some team members to reduce development risks and improve implementation efficiency.

[0127] Through the above steps, the system converts complex requirement prediction results into intuitive and operable development tasks and decision suggestions, supporting software development teams to conduct data-driven forward-looking planning and resource optimization allocation.

[0128] It should be noted that through the above five main steps, the software development collaborative interaction method provided by the present application forms a complete closed-loop system, from user behavior monitoring, requirement evolution analysis, multi-agent collaborative prediction, distributed verification to task planning and resource allocation, realizing intelligent support for the whole process of software development. The method can effectively solve the technical challenges in multi-user group requirement capture and prediction, and improve the efficiency and quality of software development.

Claims

1. A software development collaborative interaction method, characterized in that: The following steps are involved: Deploy user behavior monitoring components to collect operation sequence data from multiple user groups and convert user behavior data into structured demand items through semantic inference technology; Construct a demand evolution graph model to record the historical demand evolution path of each user group. The demand evolution graph model is a directed graph structure, in which nodes represent structured demand items and edges represent the evolutionary relationship between demand items; Establish a multi-agent collaborative forecasting system to integrate the local forecasting models of multiple user groups to generate global demand forecast results, where the local forecasting model of each user group is fused with the global demand evolution map through dynamic weights; Implement a distributed verification system to verify and correct forecast results through feedback from actual operational behaviors of various user groups, thus forming a closed-loop optimization of demand forecasting and verification; Based on the verified prediction results, structured collaborative development tasks and priority recommendations are generated to support software development teams in forward-looking planning and resource allocation.

2. The software development collaborative interaction method according to claim 1, characterized in that: The step of deploying a user behavior monitoring component to collect operation sequence data of multiple user groups includes: Embed low-intrusion operation data collection probes in different functional modules of the software application, and the probes collect user interaction events and their context information; Clean, divide, and extract features from the collected raw operation data to form a standardized operation sequence; The standardized operation sequences are classified according to user groups and stored in a distributed database.

3. The software development collaborative interaction method according to claim 1, characterized in that: The steps of constructing the demand evolution graph model include: Conduct time series analysis on historical demand data of each user group to identify the evolutionary correlation between demand items; Construct a weighted directed graph structure, where the node weight represents the frequency of occurrence of demand items and the edge weight represents the strength of evolutionary relationships; A graph traversal algorithm is applied to the weighted directed graph structure to extract high-frequency evolution paths and key branch points.

4. The software development collaborative interaction method according to claim 1, characterized in that: The steps of establishing a multi-agent collaborative prediction system include: Build a local prediction model for each user group, which combines graph neural networks and sequence prediction algorithms; Set up a dynamic weight allocation mechanism to automatically adjust the weight based on the historical prediction accuracy of each user group and the relevance of the current task; The results of each local forecast model are fused through a weighted voting algorithm to generate a global demand forecast.

5. The software development collaborative interaction method according to claim 1, characterized in that: The steps of implementing the distributed verification system include: Convert the predicted demand items into a set of observable expected user behavior hypotheses; Deploy distributed verification probes to collect users' actual operation behaviors in specific scenarios; Calculate the degree of match between actual operational behavior and expected behavior assumptions, and conduct quantitative evaluation and correction of prediction results.

6. The software development collaborative interaction method according to claim 1, characterized in that: The step of generating a structured collaborative development task based on the verified prediction result includes: Convert verified demand forecast results into functional development items; Prioritize functional development items based on dependency analysis and resource constraints; Generate development task cards containing task descriptions, technical dependencies, resource requirements, and time estimates, and provide them to the development team through a visual interface.

7. The software development collaborative interaction method according to claim 1, characterized in that: The following steps are also included: Establish a continuous feedback mechanism to collect and analyze feedback from the development team during the execution of tasks; Dynamically adjust and optimize the prediction model and verification system based on feedback information and actual execution; Generate a historical record of collaborative interactions to provide a reference for demand forecasting in similar scenarios in the future.

8. The software development collaborative interaction method according to claim 1, characterized in that: In the step of collecting operation sequence data of multiple user groups: Said user groups include end users, business analysts, developers, and testers; According to the characteristics of different user groups, specific data collection strategies and feature extraction methods are designed respectively.

9. A software development collaborative interaction device, characterized in that: include: User behavior monitoring module, which collects operation sequence data of multiple user groups and converts user behavior data into structured demand items through semantic inference technology; The demand evolution graph construction module is used to record the historical demand evolution path of each user group and construct it into a directed graph structure, where nodes represent structured demand items and edges represent the evolutionary relationships between demand items; The multi-agent collaborative forecasting module is used to integrate the local forecasting models of multiple user groups to generate global demand forecast results. The local forecasting model of each user group is fused with the global demand evolution map through dynamic weights. A distributed verification module is used to verify and correct the forecast results through feedback from the actual operation behavior of each user group, forming a closed-loop optimization of demand forecasting and verification; The task generation module is used to generate structured collaborative development tasks and priority recommendations based on verified prediction results, supporting software development teams in forward-looking planning and resource allocation.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images