AI-based digital project performance evaluation data processing method and system
By using AI technology to achieve unified collection and processing of multi-source data and dynamically adjust evaluation indicators, the problems of data dispersion and static indicators in traditional project performance evaluation are solved. This enables accurate prediction of project risks and improves management efficiency and information transparency.
Patent Information
- Application Number
- CN202511279576.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-09
- Publication Date
- 2025-12-09
AI Technical Summary
Traditional project performance evaluation methods cannot effectively handle multi-source heterogeneous data, lack unified semantic mapping rules, cannot iterate quickly, and static KPI indicators lack dynamism, cannot accurately reflect the project status, and cannot predict risks in advance.
Employing AI-based multi-source data acquisition, entity alignment, dynamic indicator generation, and performance prediction modules, and utilizing unstructured data processing, graph neural networks, reinforcement learning, and temporal graph convolutional networks, this system enables cross-system data entity association, dynamic indicator weight adjustment, and task dependency modeling, while providing a visualization interface for data display.
It enables comprehensive collection and efficient processing of multi-source data, ensuring the accuracy and consistency of data integration, dynamically adjusting the weights of evaluation indicators to improve the accuracy and adaptability of evaluation, predicting project delay risks in advance, providing intuitive data display functions, and improving project management efficiency.
Smart Images

Figure CN121094748A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of artificial intelligence and digital project management technology, specifically to an AI-based digital project performance evaluation data processing method and system. Background Technology
[0002] With the deepening of digital transformation, project management models are undergoing profound changes, exhibiting characteristics such as frequent cross-team collaboration, fragmented toolchains, and diversified data types. These changes pose serious challenges to traditional project performance evaluation methods. 1. The core data of current digital projects are scattered across heterogeneous toolchains. In addition to JIRA (requirements management), GitLab (code repository), Splunk (log analysis), and financial systems (cost accounting), dozens of other tools are also involved, such as Confluence (document collaboration), Jenkins (continuous integration), WeChat Work (communication records), and Figma (design assets). The data formats and semantic standards of the above systems differ greatly, and there is a lack of unified semantic mapping rules, which makes it impossible to directly link the data. 2. Traditional ETL tools rely on predefined data patterns. When a new data source is added, the data adaptation script needs to be redeveloped, which cannot meet the needs of rapid project iteration. Moreover, with the development of the times, the proportion of unstructured data continues to rise, while traditional tools can only process structured data such as tables and databases, resulting in a large amount of key information not being included in the evaluation system. 3. Traditional project performance evaluation uses static KPI indicators, which do not include "dynamic influencing factors". The weights of the indicators are set manually in advance, but the stage characteristics of the entire project life cycle are not considered. For example, in the project initiation phase, the key indicators should be "completeness of requirements documents" and "resource availability rate", but traditional KPIs still focus on "delivery efficiency". The manual setting of weights also has subjective bias, resulting in the evaluation results lacking objectivity and failing to truly reflect the overall status of the project. 4. Traditional tools can only trigger reminders after a task is overdue, but cannot predict risks in advance, and rely heavily on human experience to make judgments when conducting root cause analysis.
[0003] Based on this, we now provide an AI-based digital project performance evaluation data processing method and system that can eliminate the drawbacks of existing technical solutions. Summary of the Invention
[0004] The purpose of this invention is to provide an AI-based digital project performance evaluation data processing method and system to address the serious challenges posed by traditional project performance evaluation methods in the background art.
[0005] To achieve the above objectives, the present invention provides the following technical solution: An AI-based digital project performance evaluation data processing system includes: A multi-source data acquisition module is used to acquire structured, semi-structured, and unstructured data from the project management toolchain. The multi-source data acquisition module includes an unstructured data processing unit and a data lineage tracing unit. The entity alignment module establishes the association relationship between cross-system data entities based on the BERT-based semantic matching algorithm and graph neural network. The entity alignment module includes a data preprocessing unit and an entity association unit. A dynamic indicator generation module is used to dynamically adjust the weights of evaluation indicators using a reinforcement learning framework. The dynamic indicator generation module includes a project phase classifier and a reinforcement learning agent unit. The performance prediction module is used to model task dependencies based on a temporal graph convolutional network and outputs the probability of delay risk. The performance prediction module includes a task feature extraction unit and a temporal prediction unit. A visualization interface is used to intuitively display data in a visual style and to support users in viewing project data, evaluation results and performance prediction reports. The visualization interface includes a data report generation unit and an interactive dashboard unit. The multi-source data acquisition module, entity alignment module, dynamic indicator generation module, performance prediction module, and visualization interface communicate with each other via a network.
[0006] Preferably, the unstructured data processing unit is used to parse the chart data in the project document using OCR technology, convert the chart information in image form into structured data, and to convert the meeting recording into text data using ASR technology, converting unstructured data into structured data. The data lineage tracing unit is used to record the original information of multi-source data, including the original system name, data storage path, conversion path during data processing, and operation records.
[0007] Preferably, the data preprocessing unit is used to clean, deduplicate, and normalize the collected multi-source data. The cleaning operation is used to remove missing values, outliers, and invalid data from the multi-source data. The deduplicate operation is used to identify duplicate data based on the combination of key fields of the multi-source data and retain unique valid data. The normalization operation is used to convert data of different formats and units into a preset standard format and unit. The entity association unit is used to perform entity recognition on the preprocessed data using graph neural networks, identify the core entities in the digital project, and then establish the association between the core entities through entity attribute matching and relationship mining operations to construct a knowledge graph about the project data. The core entities include tasks, resources and personnel.
[0008] Preferably, the project phase classifier uses the Transformer model to analyze project data, extract key project features, and identify the current phase of the project based on a preset project phase division standard. The key project features include project progress, task completion status, and resource investment status. The phases include the requirements analysis phase, development phase, testing phase, and launch phase. The reinforcement learning agent unit uses a reinforcement learning algorithm to learn the weight allocation strategy of evaluation indicator data under different project stages, and is used to generate an indicator weight adjustment scheme based on the key characteristics of the project at the current stage and historical evaluation indicator data.
[0009] Preferably, the task feature extraction unit is used to extract key task features from project data. The key task features include task dependencies, resource requirements, and task duration. The task dependencies include the association between preceding and subsequent tasks and the collaborative relationship between parallel tasks. The resource requirements include the types and quantities of hardware resources, software resources, and human resources required by the task. The task duration includes the planned start time, planned completion time, actual start time, actual completion time, and estimated time. The time-series prediction unit is used to capture the changing patterns of task dependencies over time using a time-series graph convolutional network based on extracted key task features and historical project data, and to construct a project delay risk prediction model, outputting the delay risk probability for different project stages and the project as a whole.
[0010] Preferably, the data report generation unit is used to generate data reports according to user needs and to display project data in the form of a combination of charts and text. The data reports include project progress reports, resource usage reports and cost consumption reports. The interactive dashboard unit provides a customizable display interface, allowing users to configure and view key indicators, trend charts, and warning information as needed.
[0011] An AI-based data processing method for evaluating the effectiveness of digital projects, applied to the aforementioned data processing system for evaluating the effectiveness of digital projects, specifically includes the following steps: Step S1: Collect project management data from independent systems in the project management toolchain in real time through the multi-source data acquisition module. The data types include structured data, semi-structured data, and unstructured data. Step S2: Based on the data processed by the entity alignment module, identify the task, resource, and personnel entities in the project, establish the relationships between entities, and construct a task-resource-personnel entity relationship network; Step S3: Based on reinforcement learning, and combined with the current project stage identified by the project stage classifier in the dynamic indicator generation module, the key project features in the project documents are extracted by the reinforcement learning agent unit to generate an initial indicator set. Then, the indicator weight allocation is automatically adjusted according to the project stage transition and historical evaluation indicator data to generate a stage-adapted performance indicator system. Step S4: Call the performance prediction module. Based on the performance indicator system, the task feature extraction unit extracts key task features from the project data and inputs them into the time series graph convolutional network of the time series prediction unit to construct a task dependency graph and output a performance deviation warning. Step S5: Use the SHAP interpretability model to analyze the performance prediction results, identify key factors affecting project performance, analyze the causes of risks, and clarify the specific links and problems that lead to performance deviations. Step S6: Based on the risk analysis results, call the visualization interface to generate data reports synchronously by the data report generation unit. Generate a visual root cause analysis map through the interactive dashboard unit, mark key blocking paths in the visual root cause analysis map, and display the risk propagation path and impact scope to assist in the formulation of targeted risk response measures.
[0012] Preferably, in step S1, different collection strategies are adopted for different types of data: for structured data, data interface connections are established with each system through API adapters, and data is acquired in real time according to a preset collection frequency; for semi-structured data, data is crawled from the system through log crawlers according to specified crawling rules and frequencies; for unstructured data, it is processed by OCR technology and ASR technology respectively, and then converted into structured and semi-structured data before collection.
[0013] Preferably, step S2 specifically includes: the data preprocessing unit preprocesses the collected data, removes redundant and erroneous data, and unifies the data format and standards; then, using the graph neural network algorithm in the entity association unit, it performs entity recognition on the preprocessed data, determines the boundaries and attributes of task, resource, and personnel entities, establishes the association between entities based on the similarity of entity attributes and business logic relationships, and stores the entities and the associations between entities as a graph structure according to the knowledge graph construction specifications, forming a task-resource-personnel entity relationship network.
[0014] Preferably, step S4 specifically includes: calling the task feature extraction unit of the performance prediction module to sort out the sequence and dependency relationship of all tasks in the project data, determine the predecessor and successor tasks of each task, construct the task dependency graph of the project, the node attributes of the task dependency graph include task complexity and the skill value of the person in charge, assign attribute information to each node, and the task feature extraction unit performs quantitative evaluation based on the technical difficulty of the task and the historical performance of the personnel. The time series prediction unit calculates the delay probability of downstream tasks in the task dependency graph based on the time series graph convolutional network and the risk propagation algorithm. Starting from the upstream tasks with delay risk, the unit calculates the degree of impact on each downstream task step by step according to the correlation strength between tasks and the delay time of the upstream tasks, thereby obtaining the delay probability of the downstream tasks, and generating an performance deviation warning based on the delay probability threshold.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention establishes an AI-based digital project performance evaluation data processing system and method, achieving comprehensive collection, efficient processing, and evaluation of project data. The multi-source data acquisition module automatically extracts data from multiple heterogeneous systems, solving the problem of data dispersion. Simultaneously, the unstructured data processing unit utilizes OCR and ASR technologies to effectively transform unstructured data into usable information, enhancing data integrity and usability. This addresses existing challenges in data integration, evaluation accuracy, and decision support, and also improves the management efficiency of digital projects. The entity alignment module uses advanced semantic matching algorithms and graph neural networks to ensure precise associations between cross-system data entities, constructing a unified data view. The dynamic indicator generation module adjusts the weights of evaluation indicators according to project stages and actual needs, avoiding the limitations of traditional fixed KPI systems and improving the accuracy and adaptability of evaluation. The performance prediction module models task dependencies through time-series graph convolutional networks, accurately predicting project delay risks and providing decision support for project managers. The visualization interface provides intuitive and interactive data display functions, promoting information transparency and communication efficiency. Attached Figure Description
[0016] Figure 1 This is a schematic diagram of the processing system of the present invention.
[0017] Figure 2 This is a schematic diagram of each module of the present invention.
[0018] Figure 3 This is a flowchart illustrating the processing method of the present invention.
[0019] Figure label annotations: Multi-source data acquisition module 100, unstructured data processing unit 110, data lineage tracing unit 120, entity alignment module 200, data preprocessing unit 210, entity association unit 220, dynamic indicator generation module 300, project stage classifier 310, reinforcement learning agent unit 320, performance prediction module 400, task feature extraction unit 410, time series prediction unit 420, visualization interface 500, data report generation unit 510, interactive dashboard unit 520. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.
[0021] In this embodiment, as Figure 1 and Figure 2 As shown, an AI-based digital project performance evaluation data processing system includes: The multi-source data acquisition module 100 is used to acquire structured, semi-structured and unstructured data in the project management toolchain. The multi-source data acquisition module 100 includes an unstructured data processing unit 110 and a data lineage tracing unit 120. The unstructured data processing unit 110 is used to parse the chart data in the project document using OCR technology, convert the chart information in the form of images into structured data, and convert the meeting recording into text data using ASR technology, converting unstructured data into structured data. The data lineage tracing unit 120 is used to record the original information of multi-source data, including the original system name, data storage path, conversion path during data processing, and operation records. Specifically, unstructured data refers to data with irregular or incomplete data structures, without predefined data models, and data that is inconvenient to represent using two-dimensional logical tables in a database. This includes all formats of office documents, text, images, XML, HTML, various reports, images, audio, and video information. OCR technology is used for chart data (such as task progress line charts) in project documents (such as design drawings). It uses OCR image recognition algorithms to parse the numerical and label information in the charts and convert them into structured data tables. An example is a "Task ID - Planned Duration - Actual Duration" lookup table, which avoids errors and efficiency losses from manual input. ASR (Automatic Speech Recognition) technology is used for recordings of project meetings. It converts the speech content into text data and automatically extracts key information from the text, ensuring that important decisions and risk information in the meeting are not missed. Both OCR and ASR technologies are existing technologies. The unstructured data processed above is formatted and stored in an intermediate database in a unified manner with structured and semi-structured data. Specifically, existing project data is mostly scattered across independent systems such as JIRA, GitLab, and Splunk. The multi-source data acquisition module 100 can cover structured, semi-structured, and unstructured data through multiple technical means, achieving comprehensive and real-time acquisition of project data. The acquisition tools include API adapters and log crawlers, filling the gap in traditional ETL tools that "cannot handle unstructured data" and solving the data credibility problem caused by "unclear data sources and untraceable data processing" in existing technologies. The entity alignment module 200 establishes the association relationship between cross-system data entities based on the BERT-based semantic matching algorithm and graph neural network. The entity alignment module 200 includes a data preprocessing unit 210 and an entity association unit 220. The data preprocessing unit 210 is used to clean, deduplicate and normalize the collected multi-source data. The cleaning operation is used to remove missing values, outliers and invalid data in the multi-source data. The deduplication operation is used to identify duplicate data based on the combination of key fields of multi-source data and retain unique valid data. The normalization operation is used to convert data of different formats and units into preset standard formats and units. The entity association unit 220 is used to perform entity recognition on the preprocessed data using graph neural networks, identify the core entities in the digital project, and then establish the association between the core entities through entity attribute matching and relationship mining operations to construct a knowledge graph about the project data. The core entities include tasks, resources and personnel. Specifically, entity alignment refers to the process of using natural language processing and graph computing techniques to identify entities representing the same actual object from different data sources and establish relationships. BERT is a pre-trained language model that is fine-tuned for project entities. It can understand text semantics through bidirectional context and is used for semantic matching of entity names and feature extraction of unstructured text. The input is multi-source text and structured data, and the output is a knowledge graph after entity alignment. BERT is used to determine whether they are the same entity. Graph Neural Network (GNN) is a deep learning model that is specifically designed to process graph structure data. It is used to build entity relationship networks and calculate the relationship strength between entities. A knowledge graph is a semantic network with entities as nodes and relationships as edges. It is used to store the relationships between tasks, resources, and people. Specifically, the entity alignment module 200 uses semantic matching and association to construct a unified project data standard format for the "heterogeneous data scattered across different systems" acquired by the multi-source data acquisition module 100. This ensures that subsequent performance predictions are semantically consistent and clearly correlated. For example, it maps "task priority: High / Medium / Low" in JIRA and "task urgency: 1 / 2 / 3" in GitLab to "priority: high / medium / low", and unifies the "date format (YYYY-MM-DD / MM / DD / YYYY)" of different systems to "YYYY-MM-DD". The dynamic indicator generation module 300 is used to dynamically adjust the weights of evaluation indicators using a reinforcement learning framework. The dynamic indicator generation module 300 includes a project phase classifier 310 and a reinforcement learning agent unit 320. The project phase classifier 310 uses the Transformer model to analyze project data, extract key project features, and identify the current phase of the project based on preset project phase division criteria. Key project features include project progress, task completion status, and resource input status. The phases include requirements analysis phase, development phase, testing phase, and launch phase. The reinforcement learning agent unit 320 uses a reinforcement learning algorithm to learn the weight allocation strategy of evaluation indicator data under different project stages, which is used to generate indicator weight adjustment schemes based on the key features of the project at the current stage and historical evaluation indicator data. Specifically, the reinforcement learning framework is an AI framework that learns the optimal decision-making strategy through trial and error in the interaction between the agent and the environment. The input is the current stage and historical indicator data, and the output is the indicator weight allocation scheme, which is used to dynamically adjust the performance evaluation indicator weights for different project stages. The temporal graph convolutional network is a deep learning model that combines graph convolutional networks and temporal models, which can simultaneously capture task dependencies and their changes over time. The dynamic indicator generation module 300 dynamically generates an "adaptive performance indicator system for the current stage" based on the characteristics of different project stages, ensuring the timeliness and accuracy of the evaluation indicators. The Transformer model trains a classifier for project stage classification tasks, with the input being project data (such as progress, resources, task completion status, etc.) and the output being the current stage of the project. The performance prediction module 400 is used to model task dependencies based on the temporal graph convolutional network and outputs the delay risk probability. The performance prediction module 400 includes a task feature extraction unit 410 and a temporal prediction unit 420. The task feature extraction unit 410 is used to extract key task features from project data. Key task features include task dependencies, resource requirements, and task duration. Task dependencies include the relationship between preceding and subsequent tasks and the collaborative relationship between parallel tasks. Resource requirements include the types and quantities of hardware resources, software resources, and human resources required by the task. Task duration includes the planned start time, planned completion time, actual start time, actual completion time, and estimated time. The time series prediction unit 420 is used to capture the changing patterns of task dependencies over time based on extracted key task features and historical project data, and to build a project delay risk prediction model, outputting the delay risk probability of different project stages and the project as a whole. Specifically, temporal graph convolutional networks are a hybrid deep learning model that combines "graph convolutional networks (GCNs)" with "temporal models (such as RNNs and LSTMs)". The input is a task dependency graph and temporal data, and the output is the probability of delay risk. The steps for modeling task dependencies using temporal graph convolutional networks are as follows: Each independent task in the project is treated as a "node" in the graph, and each node is assigned attributes such as task complexity and the skill value of the person in charge. It can also be extended to include core project data such as "planned task duration" and "percentage of resources consumed". Based on the actual dependencies between tasks, "edges" connecting nodes are defined. If "task Y can only be started after task X is completed", then a "directed edge" (from X to Y) is established between "task X node" and "task Y node", indicating "X's prerequisite dependency on Y". If "task M and task N share the same development resources", then an "undirected edge" is established between them, indicating "resource competition dependency". The task dependency graph is input into a temporal graph convolutional network, and the task dependency graph is convolved through graph convolutional layers to calculate the local association features of each task node. For example: By analyzing the "complexity" and "responsible person's skill value" of task A, and combining this with the progress status of its dependent task B, the influence weight of task A's current progress on subsequent dependent task C is derived. Then, a time-series layer is introduced to introduce a sliding window with a time dimension to perform time-series modeling of the task dependency graph features for each time slice. For example: Tracking the "completion rate of task A over time" and combining it with the "delay status of task B," the "influence weight of task A on task C" is dynamically updated. The final output of the time-series graph convolutional network is a global representation vector that "integrates task dependencies and dynamic time features." This vector can fully reflect the current status of all tasks, the dependency influence between tasks, and the trend of change over time. Based on the global representation vector, through structures such as fully connected layers, the overall project delay risk probability and the delay risk probability of key tasks are calculated and output. For example: If the model detects that "task B (preceding task) has been delayed by X days, and its responsible person's skill value is lower than the task requirement threshold (e.g., 70 points)," and "task C (dependent on task B) has only Y days remaining in its planned duration," then the influence coefficient of task B on downstream tasks is calculated. ,in Quantify the percentage of time spent in delay. The amplified risk of delays due to insufficient quantification skills, combined with the remaining planned duration of dependent task C, is used to calculate the buffer coefficient for task C: When the buffer coefficient is less than 1, it is determined that task C does not have enough time to absorb the impact of the delay of the preceding task, and the delay probability of task C is output: When the buffer coefficient is greater than 1, task C is determined to have no risk of delay. The overall project delay probability is calculated as follows: Based on the delay probability of all critical path tasks (including task C), a weighted summation logic is used. The weight of critical path tasks is set according to their proportion in the total project duration. Starting from the upstream task with delay risk, the impact on each downstream task is calculated step by step according to the correlation strength between tasks and the delay time of upstream tasks. The delay probability of downstream tasks is then obtained, and an efficiency deviation warning is generated based on the delay probability threshold. At the same time, root cause analysis is performed in combination with task attribute information (such as insufficient skill value of the person in charge) and historical data to determine the main reasons for the delay risk. For example: the person in charge of task B has mismatched skills, which leads to delay. The time buffer of task C is insufficient. It is necessary to adjust personnel or extend the time of task C. The visualization interface 500 is used to intuitively display data in a visual style and supports users to view project data, evaluation results and performance prediction reports. The visualization interface 500 includes a data report generation unit 510 and an interactive dashboard unit 520. The data report generation unit 510 is used to generate data reports according to user needs and to display project data in the form of a combination of charts and text. The data reports include project progress reports, resource usage reports and cost consumption reports. The interactive dashboard unit 520 provides a customizable display interface, allowing users to configure and view key indicators, trend charts, and alerts as needed. The multi-source data acquisition module 100, entity alignment module 200, dynamic indicator generation module 300, efficiency prediction module 400 and visualization interface 500 communicate with each other via a network. Specifically, the system includes a multi-source data acquisition module 100, an entity alignment module 200, a dynamic indicator generation module 300, an efficiency prediction module 400, and a visualization interface 500. It can achieve comprehensive collection and fusion processing of structured, semi-structured, and unstructured data in the project management toolchain. Through AI technologies such as BERT semantic matching, graph neural networks, reinforcement learning, and temporal graph convolutional networks, it can achieve cross-system entity alignment, dynamic indicator weight adjustment, task delay risk prediction, and root cause analysis. With the help of the visualization interface, it provides intuitive data display and interactive functions, and has good application prospects. Among them, such as Figure 3 As shown, an AI-based data processing method for digital project effectiveness evaluation, applied to a digital project effectiveness evaluation data processing system, specifically includes the following steps: Step S1: Collect project management data from independent systems in the project management toolchain in real time through the multi-source data acquisition module 100. The data types include structured data, semi-structured data, and unstructured data. Specifically, in step S1, different collection strategies are adopted for different types of data: for structured data, data interface connections are established with each system through API adapters, and data is acquired in real time according to the preset collection frequency; for semi-structured data, data is crawled from the system through log crawlers according to the specified crawling rules and frequency; for unstructured data, it is processed by OCR technology and ASR technology respectively, and then converted into structured and semi-structured data before collection. Step S2: Based on the data processed by the entity alignment module 200, identify the task, resource and personnel entities in the project, establish the relationship between entities, and construct a task-resource-personnel entity relationship network. The aligned entity network provides structured input for stage identification and indicator weight adjustment. Specifically, the data preprocessing unit 210 preprocesses the collected data, removes redundant and erroneous data, and standardizes the data format and standards. Then, using the graph neural network algorithm in the entity association unit 220, entity recognition is performed on the preprocessed data to determine the boundaries and attributes of task, resource, and personnel entities. Based on the similarity of entity attributes and business logic relationships, the associations between entities are established. In accordance with the knowledge graph construction specifications, the entities and the relationships between entities are stored as a graph structure to form a task-resource-personnel entity relationship network. Step S3: Based on reinforcement learning, and combined with the current project stage identified by the project stage classifier 310 in the dynamic indicator generation module 300, the reinforcement learning agent unit 320 extracts key project features from the project documents to generate an initial indicator set. Then, based on the project stage transition and historical evaluation indicator data, the indicator weight allocation is automatically adjusted to generate a stage-adapted performance indicator system. The dynamically generated indicator system serves as the input features for the prediction model. The initial indicator set consists of preset indicators such as project type, industry standards, and historical best practices, for example: task completion rate, code quality, resource utilization, and frequency of requirement changes. The performance indicator system includes quantitative indicators in multiple dimensions, such as progress, quality, resources, and cost. The weight allocation can be adjusted according to actual needs to ensure that the evaluation results are more timely and targeted. Step S4: Invoke the performance prediction module 400. Based on the performance indicator system, the task feature extraction unit 410 extracts key task features from the project data and inputs them into the time series graph convolutional network of the time series prediction unit 420 to construct a task dependency graph and output a performance deviation warning. The prediction results are input into the SHAP interpretability model for root cause analysis. The task dependency graph is a directed graph (DAG), where nodes are tasks and edges represent dependencies. Node attributes include task complexity, responsible person's skill value, planned duration, etc. When the risk probability exceeds a preset threshold, a performance deviation warning is triggered. Root cause analysis is then performed in conjunction with historical data and task attributes to identify key factors causing delays (such as resource conflicts, skill mismatch, and frequent requirement changes). Specifically, this includes: calling the task feature extraction unit 410 of the performance prediction module 400 to sort out the sequence and dependency relationships of all tasks in the project data, determine the predecessor and successor tasks of each task, and construct the task dependency graph of the project. The node attributes of the task dependency graph include task complexity and the skill value of the person in charge. Attribute information is assigned to each node, and the task feature extraction unit 410 performs quantitative evaluation based on the technical difficulty of the task and the historical performance of the personnel. The task complexity is quantitatively evaluated based on the technical difficulty and workload required by the task, such as using a 1-5 level rating, with level 5 being the highest difficulty. The skill value of the person in charge is comprehensively scored based on the professional skill level of the personnel, project experience, and the quality of historical task completion. If a percentage system is used, the threshold is set to 70 points. If the score is lower than this value, it is judged as insufficient skill matching. The time series prediction unit 420 is based on the time series graph convolutional network and calculates the delay probability of downstream tasks in the task dependency graph through a risk propagation algorithm. Starting from the upstream tasks with delay risk, the influence of each downstream task is calculated step by step according to the correlation strength between tasks and the delay time of the upstream tasks, thereby obtaining the delay probability of the downstream tasks, and generating an performance deviation warning based on the delay probability threshold. Specifically, the risk propagation algorithm is a calculation method that simulates the process of risk transmission in the task dependency chain. By quantifying the impact of upstream task delays on downstream tasks, it predicts the overall project risk. At the same time, it combines task attribute information and historical data to conduct root cause analysis and determine the main causes of delay risks. Step S5: Use the SHAP interpretability model to analyze the performance prediction results, identify key factors affecting project performance, analyze the causes of risks, clarify the specific links and problems that lead to performance deviations, and visualize the analysis results to assist decision-making. Specifically, this step introduces the SHAP interpretability model, which is a model interpretation method based on Shapley value theory. The input is the prediction results and feature data, and the output is the contribution of each feature to the prediction results. By quantifying the contribution of each factor (such as task delay time, resource gap, and personnel skills) to the prediction results, the key factors affecting project effectiveness are identified, the specific links and problems that lead to performance deviation are clarified, and the credibility and acceptability of the evaluation results are improved. Step S6: Based on the risk analysis results, call the visualization interface 500, and the data report generation unit 510 synchronously generates a data report. The interactive dashboard unit 520 generates a visual root cause analysis map, marks key blocking paths in the visual root cause analysis map, displays the risk propagation path and impact range, and assists in formulating targeted risk response measures. Specifically, the entire process of the AI-based digital project performance evaluation data processing method is achieved through the collaborative operation of the processing system modules. Step S1 is executed by the multi-source data acquisition module 100, step S2 is executed by the entity alignment module 200, step S3 is executed by the dynamic indicator generation module 300, step S4 is executed by the performance prediction module 400, and steps S5-S6 are executed collaboratively by the performance prediction module 400 and the visualization interface 500. Each module achieves real-time data flow and functional linkage through network communication. Example 1
[0022] Let's take the "e-commerce platform V3.0 version upgrade project" as the application scenario: The project period is from January 1, 2025 to June 30, 2025 (180 days in total). It involves multiple project management toolchains, including JIRA (requirements / task management), GitLab (code management), Splunk (log analysis), Confluence (document collaboration), WeChat Work (communication records), and a financial system (cost accounting). The core tasks are divided into five phases: requirements analysis (1-30 days), architecture design (31-60 days), feature development (61-120 days), testing and acceptance (121-165 days), and deployment and maintenance (166-180 days). In the data acquisition phase, differentiated acquisition technologies and strategies are adopted for different types of data: For structured data, task progress and cost data are collected from JIRA and the financial system through API adapters; for semi-structured data, system operation logs are crawled from Splunk through log crawlers; for project documents in unstructured data, design diagrams, schedules, etc. in Confluence are parsed using OCR technology to convert image-based chart information into structured data; for meeting recordings in unstructured data, several meeting recordings in WeChat Work are processed using ASR technology to extract key text information such as task allocation, progress commitments, and risk feedback from the recordings. At the same time, the data lineage tracking unit 120 records the flow information of all data sources in real time and generates data lineage records, covering the original system name, data storage path, and processing and transformation path to ensure data traceability. An example is: WeChat Work audio → ASR → text → keyword extraction → risk event table. The entity alignment module 200 processing flow is as follows: The data preprocessing unit 210 cleans, deduplicates, and normalizes the collected multi-source data. For example, it cleans tasks with a "cancelled" status in JIRA, unifies the time format DD / MM / YYYY in GitLab to YYYY-MM-DD, aligns the "high / medium / low" priority parsed from Confluence with the "high / medium / low" mapping in JIRA, removes invalid and duplicate data, and converts date data of different formats and cost data of different units into a preset standard format. The entity association unit 220 uses a graph neural network to perform entity recognition on the preprocessed project data, identifying three core categories: tasks, resources, and personnel. Entities, for example: tasks such as "ID:ECOM-205, Name: Develop Recommendation Algorithm Interface, Type: Development Task", personnel such as "ID:P1001, Name: Zhang San, Role: Backend Development Engineer", resources such as "ID:SERVER-08, Type: Test Server, Configuration: 8C-16G", a knowledge graph is constructed through entity attribute matching and relationship mining. The graph contains entity nodes and associated edges. Examples of association relationships are: "(Zhang San) → [Responsible for] → (ECOM-205)", "(ECOM-205) → [Dependency] → (ECOM-200), (ECOM-200) is a database design task", "(ECOM-205) → [Consumes] → (SERVER-08)"; During the requirements analysis phase, the project phase classifier 310 uses the Transformer model to analyze project data and extracts key features such as requirement document completion rate and resource availability rate based on the current data. This identifies that the project is in the "requirements analysis phase" on January 20, 2025. The reinforcement learning agent unit 320 combines data from several similar historical project phases to set the weights of core evaluation indicators. For example, the weights for the requirements analysis phase are "requirements document completeness 35%, requirements review pass rate 30%, resource availability rate 35%". When entering the feature development phase, the module automatically adjusts the weights of the core indicators to "code submission frequency 25%, unit test pass rate 40%, development task completion rate 35%". During the testing and acceptance phase, the weight of "defect density" is increased to 45% to ensure that the evaluation indicators match the phase characteristics. All of the above data are hypothetical values. Performance prediction and root cause analysis were performed: The current date is April 10, 2025 (mid-development phase). The problem is that the task "User Service Reconstruction" (Task B) is behind schedule. The planned duration is 20 days (March 21 to April 10), but it is actually expected to be delayed by 3 days (completed on April 13). The person in charge has a skill score of 65 (below the 70-point threshold). The performance prediction module 400 uses a temporal graph convolutional network to quantify the risk of delay and identifies that the task "Recommendation Algorithm Interface Development" (Task C) is strongly dependent on Task B. A subgraph containing nodes of Task B and Task C is constructed, and the influence coefficient of Task B on Task C is calculated. Calculate the buffer coefficient for task C: Since the buffer coefficient is greater than 1, task C is determined to have no risk of delay. If task B is delayed by 10 days and the person in charge has a skill score of 50, the impact coefficient will increase. If the buffer coefficient is less than 1, the system will trigger a delay warning for task C and analyze the root cause to find that the main reason is the mismatch of the person in charge of task B's skills and the reduction of buffer time due to the delay of the preceding task. The system will also generate a cause contribution score, so that the situation can be predicted when the task just shows signs of risk, which will facilitate subsequent measures. The data report generation unit 510 automatically generates project progress reports, resource usage reports, and cost consumption reports according to user needs. The reports use a combination of line charts (progress deviation trend), bar charts (resource load rate), and text. The interactive dashboard unit 520 provides a customizable interface. Users can configure the display layout of key indicators (such as task completion rate, defect density, and delay risk value) through drag-and-drop operations, and view data details through zoom operations, ensuring that risk information is intuitively visible on the interactive dashboard. For example, task B is marked in red, showing the number of delay days and the skill value of the person in charge. The dependency edge pointing to task C is displayed in bold and marked "high-risk dependency". The graph shows the risk propagation path of "delay of task B → may affect task C → and thus affect subsequent test tasks". According to the analysis results of the SHAP analysis report, if it is pointed out that "the skill value of the person in charge" is the main contributing factor, then a senior engineer will be assigned to assist task B, and the time of task C will be reassessed with the team, with buffer time reserved. In summary, this system can integrate various types of data, dynamically adapt indicator weights, make the evaluation results more in line with the actual situation of the project, and identify risks in advance, allowing sufficient time for response. With the comprehensive support of this system, the overall project delay situation can be effectively improved, while significantly reducing the cost of manual input in data processing and evaluation, and improving the intelligence and efficiency of project management.
[0023] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A data processing system for evaluating the effectiveness of digital projects based on AI, characterized in that, include: A multi-source data acquisition module (100) is used to acquire structured, semi-structured and unstructured data in the project management toolchain. The multi-source data acquisition module (100) includes an unstructured data processing unit (110) and a data lineage tracing unit (120). The entity alignment module (200) establishes the association relationship between cross-system data entities based on the BERT-based semantic matching algorithm and graph neural network. The entity alignment module (200) includes a data preprocessing unit (210) and an entity association unit (220). A dynamic indicator generation module (300) is used to dynamically adjust the weights of evaluation indicators using a reinforcement learning framework. The dynamic indicator generation module (300) includes a project phase classifier (310) and a reinforcement learning agent unit (320). The performance prediction module (400) is used to model task dependencies based on the temporal graph convolutional network and output the delay risk probability. The performance prediction module (400) includes a task feature extraction unit (410) and a temporal prediction unit (420). A visualization interface (500) is used to intuitively display data in a visual style and to support users in viewing project data, evaluation results and performance prediction reports. The visualization interface (500) includes a data report generation unit (510) and an interactive dashboard unit (520). The multi-source data acquisition module (100), entity alignment module (200), dynamic indicator generation module (300), performance prediction module (400), and visualization interface (500) communicate with each other via a network.
2. The digital project performance evaluation data processing system according to claim 1, characterized in that, The unstructured data processing unit (110) is used to parse the chart data in the project document through OCR technology, convert the chart information in the form of images into structured data, and convert the meeting recording into text data through ASR technology, converting unstructured data into structured data. The data lineage tracking unit (120) is used to record the original information of multi-source data, including the original system name, data storage path, conversion path during data processing, and operation records.
3. The digital project performance evaluation data processing system according to claim 1, characterized in that, The data preprocessing unit (210) is used to clean, deduplicate and normalize the collected multi-source data. The cleaning operation is used to remove missing values, outliers and invalid data in the multi-source data. The deduplication operation is used to determine duplicate data based on the combination of key fields of the multi-source data and retain unique valid data. The normalization operation is used to convert data of different formats and units into preset standard formats and units. The entity association unit (220) is used to perform entity recognition on the preprocessed data using a graph neural network, identify the core entities in the digital project, and then establish the association between the core entities through entity attribute matching and relationship mining operations to construct a knowledge graph about the project data. The core entities include tasks, resources and personnel.
4. The digital project performance evaluation data processing system according to claim 1, characterized in that, The project phase classifier (310) uses the Transformer model to analyze project data, extract key project features, and identify the current phase of the project based on the preset project phase division criteria. The key project features include project progress, task completion status, and resource input status. The phases include the requirements analysis phase, development phase, testing phase, and launch phase. The reinforcement learning agent unit (320) uses a reinforcement learning algorithm to learn the weight allocation strategy of evaluation indicator data under different project stages, and is used to generate an indicator weight adjustment scheme based on the key features of the project at the current stage and historical evaluation indicator data.
5. The digital project performance evaluation data processing system according to claim 1, characterized in that, The task feature extraction unit (410) is used to extract key task features from project data. The key task features include task dependencies, resource requirements, and task duration. The task dependencies include the association between preceding and subsequent tasks and the collaborative relationship between parallel tasks. The resource requirements include the types and quantities of hardware resources, software resources, and human resources required by the task. The task duration includes the planned start time, planned completion time, actual start time, actual completion time, and estimated time. The time-series prediction unit (420) is used to capture the changing patterns of task dependencies over time using a time-series graph convolutional network based on extracted key task features and historical project data, and to construct a project delay risk prediction model, outputting the delay risk probability of different project stages and the project as a whole.
6. The digital project performance evaluation data processing system according to claim 1, characterized in that, The data report generation unit (510) is used to generate data reports according to user needs and to display project data in the form of a combination of charts and text. The data reports include project progress reports, resource usage reports and cost consumption reports. The interactive dashboard unit (520) is used to provide a customizable display interface, allowing users to configure and view key indicators, trend charts and warning information as needed.
7. A data processing method for evaluating the effectiveness of digital projects based on AI, characterized in that, The data processing system for digital project performance evaluation according to any one of claims 1-6 specifically includes the following steps: Step S1: Collect project management data from independent systems in the project management toolchain in real time through the multi-source data acquisition module (100). The data types cover structured data, semi-structured data and unstructured data. Step S2: Based on the data processed by the entity alignment module (200), identify the task, resource and personnel entities in the project, establish the relationship between entities, and construct the task-resource-personnel entity relationship network; Step S3: Based on reinforcement learning, the current stage of the project is identified by the project stage classifier (310) in the dynamic indicator generation module (300). The key features of the project in the project document are extracted by the reinforcement learning agent unit (320) to generate an initial indicator set. Then, the indicator weight allocation is automatically adjusted according to the project stage transition and historical evaluation indicator data to generate a stage-adapted performance indicator system. Step S4: Call the performance prediction module (400), based on the performance index system, the task feature extraction unit (410) extracts key task features from the project data, inputs them into the time series prediction unit (420) convolutional network, constructs the task dependency graph and outputs performance deviation warning; Step S5: Use the SHAP interpretability model to analyze the performance prediction results, identify key factors affecting project performance, analyze the causes of risks, and clarify the specific links and problems that lead to performance deviations. Step S6: Based on the risk analysis results, call the visualization interface (500), and the data report generation unit (510) generates a data report synchronously. The interactive dashboard unit (520) generates a visualization root cause analysis map, marks the key blocking paths in the visualization root cause analysis map, and displays the risk propagation path and impact range to assist in the formulation of targeted risk response measures.
8. The data processing method for digital project performance evaluation according to claim 7, characterized in that, In step S1, different collection strategies are adopted for different types of data: For structured data, data interface connections are established with each system through API adapters, and data is acquired in real time according to the preset collection frequency; for semi-structured data, data is crawled from the system through log crawlers according to the specified crawling rules and frequency; for unstructured data, it is processed by OCR technology and ASR technology respectively, and then converted into structured and semi-structured data before collection.
9. The data processing method for digital project performance evaluation according to claim 7, characterized in that, Step S2 specifically includes: the data preprocessing unit (210) preprocesses the collected data, removes redundant and erroneous data, and unifies the data format and standards. Then, using the graph neural network algorithm in the entity association unit (220), entity recognition is performed on the preprocessed data to determine the boundaries and attributes of task, resource, and personnel entities. Based on the similarity of entity attributes and business logic relationships, the association between entities is established. According to the knowledge graph construction specifications, the entities and the association between entities are stored as a graph structure to form a task-resource-personnel entity relationship network.
10. The data processing method for digital project performance evaluation according to claim 7, characterized in that, Step S4 specifically includes: calling the task feature extraction unit (410) of the performance prediction module (400), sorting out the sequence and dependency relationship of all tasks in the project data, determining the predecessor and successor tasks of each task, constructing the task dependency graph of the project, the node attributes of the task dependency graph include task complexity and the skill value of the person in charge, assigning attribute information to each node, and the task feature extraction unit (410) quantitatively evaluating the task technical difficulty and the historical performance of the personnel. The time series prediction unit (420) calculates the delay probability of the downstream tasks in the task dependency graph based on the time series graph convolutional network and the risk propagation algorithm. Starting from the upstream tasks with delay risk, the unit calculates the degree of impact on each downstream task step by step according to the correlation strength between tasks and the delay time of the upstream tasks, thereby obtaining the delay probability of the downstream tasks, and generating an performance deviation warning based on the delay probability threshold.
Citation Information
Patent Citations
Nursing risk intervention decision-making system and method based on knowledge graph
CN119560121A
Electric power capital construction process early warning method and system based on high-precision three-dimensional model
CN119919024A
Intelligent customer relationship knowledge graph construction method
CN120146177A
Project progress tracking and risk prediction system and method based on improved knowledge graph and multi-view graph neural network
CN120181781A
Intelligent programming auxiliary method and system based on multi-mode AI language model
CN120315685A
Cited By
Evaluation method and system for Chinese herbal medicine-feed compatibility interaction effect
CN121329249A
Intelligent construction management system and method
CN121544056A
Machine learning modeling method and system for dynamic prediction of housing construction project progress
CN121882834A
