Project progress intelligent evaluation method, equipment and medium
By employing multimodal semantic alignment, dynamic weight calculation, and knowledge graph reasoning, the problem of insufficient identification of multimodal evidence association and dependency relationships in project progress assessment is solved, enabling accurate assessment of project health and automated report generation, thereby improving the efficiency and accuracy of the assessment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-04-07
AI Technical Summary
Existing project progress assessment methods lack precise correlation and dynamic weight adjustment of multimodal evidence, making it difficult to adapt to the unique needs of different project types. Furthermore, insufficient identification of dependencies leads to a lack of subjectivity and accuracy in the assessment results.
By employing the domain-adaptive multimodal semantic alignment principle, dynamic weight calculation driven by multi-level analysis and expert models, and full-element knowledge graph reasoning, a project knowledge graph is constructed to perform progress logic verification and resource matching, generating project health scores and evaluation reports.
It achieves accuracy and real-time performance in project progress assessment, improves the efficiency and operability of the assessment, and provides automated comprehensive quantitative assessment and report generation.
Smart Images
Figure CN121809893A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of project management technology, and in particular to a method, device and medium for intelligent assessment of project progress. Background Technology
[0002] In current project management practices, as project scale and complexity increase, traditional schedule assessment methods are struggling to meet the demands of multi-source heterogeneous data fusion and intelligent decision-making. Existing technologies, when processing multimodal evidence for project assessment, typically rely on general cross-modal semantic analysis models. These models lack a deep understanding of specific project management domain terminology and acceptance criteria, resulting in insufficient semantic accuracy in the semantic association between heterogeneous evidence such as text reports, data tables, images, and videos and project schedule milestones. Furthermore, the weighting systems in project assessments often rely on manually preset fixed values or are generated using simple statistical methods, failing to dynamically adjust based on differences in project type, risk level, core objectives, and other inherent characteristics. This makes it difficult for assessment models to adapt to the unique needs of projects of different natures, such as basic research and applied research, thus limiting the relevance and accuracy of the assessment results.
[0003] Existing solutions generally lack the ability to systematically model and automate the complex logical relationships between project tasks, deliverables, and evidence. Project progress assessment still largely relies on the experience and judgment of managers, making it difficult to automatically and accurately verify the completeness of the deliverable evidence chain and identify dependencies between tasks. This leads to the risk of subjective bias and logical flaws in the assessment of progress status. Summary of the Invention
[0004] This application provides a method, device, and medium for intelligent project progress assessment to solve the aforementioned technical problems.
[0005] On the one hand, embodiments of this application provide a method for intelligent project progress assessment, including: Acquire multimodal evidence data related to project progress, and based on the domain-adaptive multimodal semantic alignment principle, perform semantic alignment on the multimodal evidence data to obtain the association data between evidence and progress node standards; The project feature data is acquired, encoded into a multi-dimensional project feature vector, and processed based on the dynamic weight calculation principle of project dynamic weight driven by multi-level analysis and expert model to generate dynamic weight data of the project; the dynamic weight data includes schedule dimension weight, cost dimension weight, outcome dimension weight and risk dimension weight. Based on the reasoning principle of full-element knowledge graph, a project knowledge graph data is constructed, and predefined reasoning rules are applied to the project knowledge graph data to perform progress logic verification and resource matching analysis, generating progress evaluation result data. The project knowledge graph data includes the attributes and relationship networks of task entities, evidence entities, resource entities, and outcome entities, and the progress evaluation result data includes task completion status data, evidence missing data, and risk root cause data. Based on the progress assessment results, the project health score is calculated, and the project assessment report is generated to output the project health score and the project assessment report. The project health score comprehensively reflects the weighted result of progress performance, cost performance, outcome achievement rate, and risk index. The project assessment report includes task completion status, evidence missing prompts, and risk warning information.
[0006] In one implementation of this application, based on the principle of multimodal semantic alignment for domain adaptation, the multimodal evidence data is semantically aligned to obtain the association data between the evidence and the progress node standard, specifically including: Domain semantic dictionary data is constructed using expert annotations and professional terminology from the project database. For text-based evidence data, a pre-trained text branching model is used to extract deep semantic feature data. The domain semantic dictionary data includes triples of entity, attribute, and semantic description. For tabular evidence data, the tabular data is converted into standardized text description data in the form of feature values, and the standardized text description data is processed using the text branching model to extract semantic feature data. Image evidence data and video evidence data are processed through the visual branching model. Based on the domain semantic dictionary data, the semantic similarity data between the multimodal evidence data and the progress node standard is calculated, and a dynamic threshold mechanism is applied to adjust the similarity judgment threshold to obtain evidence association data; the dynamic threshold mechanism automatically fine-tunes the threshold according to the project type and evidence type.
[0007] In one implementation of this application, the project feature data is processed based on the project dynamic weight calculation principle driven by both multi-level analysis and expert models to generate dynamic weight data for the project, specifically including: The project feature data is encoded, and the encoded project feature vector is processed by a pre-built hybrid expert model to dynamically calculate the weight allocation data of each expert sub-model through a gating network; the encoding process uses the One-Hot method to process the dimensions of project type, risk level, core objective, project cycle and discipline type; Based on the weighted data, pairwise comparison scale data for the dimensions of schedule, cost, outcome and risk are generated, and quantitative evaluation is carried out using the 1-9 scale method to construct judgment matrix data; The maximum eigenvalue and consistency index data of the judgment matrix data are calculated using the analytic hierarchy process (AHP) to verify the consistency of the judgment matrix data and to calculate the initial weight data. Based on historical project data, and using gradient descent to minimize historical project evaluation errors, the initial weight data is calibrated to generate dynamic weight data.
[0008] In one implementation of this application, project knowledge graph data is constructed based on the reasoning principle of full-element knowledge graphs, specifically including: Extract task entity data, evidence entity data, resource entity data, and outcome entity data from the project management system to construct project knowledge graph data; The task entity data includes a task ID and acceptance criteria attribute; the evidence entity data includes an evidence ID and credibility score attribute; the resource entity data includes skill tags and available time attribute; and the result entity data includes a result ID and completion status attribute. Define the types of relationships between entities and extract the associated data of tasks, evidence and results from the project management system to construct knowledge graph data; The relationship types include the demand relationship between tasks and results, the verification relationship between evidence and results, the support relationship between resources for tasks, and the dependency relationship between software tools and computing resources.
[0009] In one implementation of this application, predefined inference rules are applied to the project knowledge graph data to perform progress logic verification and resource matching analysis, generating progress evaluation result data, specifically including: The results evidence chain integrity rules are applied to process the results entity data and evidence entity data to generate evidence chain integrity status data; the results evidence chain integrity rules require at least two types of related evidence and the credibility score of each piece of evidence is not lower than a preset credibility threshold. Apply cross-achievement dependency rules to handle logical dependencies between achievement entities, trace dependency chains through SPARQL queries, and generate achievement dependency status data; By integrating the evidence chain integrity status data, the outcome dependency status data, and the dynamic weight data, resource matching analysis is performed to generate progress assessment result data.
[0010] In one implementation of this application, the project health score data is calculated based on the progress assessment result data, specifically including: Acquire schedule performance data, cost performance data, outcome achievement rate data, and risk index data, and use dynamic weight data to weight and integrate the schedule performance data, cost performance data, outcome achievement rate data, and risk index data; The progress performance data is calculated based on task completion status data, the achievement rate data is calculated based on evidence correlation data, and the risk index data is calculated based on risk root cause data. The weighted fusion process uses a linear weighted summation algorithm. The weighted comprehensive score is calculated as the project health score data. The health score data is normalized and mapped to a predefined health level standard to generate project health level data.
[0011] In one implementation of this application, generating project evaluation report data specifically includes: By integrating the project health score data, the task completion status data, and the risk root cause data, a project health status description data is generated using natural language generation technology. Based on the data of incomplete tasks and missing evidence, generate a list of task completion status data, and based on the risk root cause data, generate risk warning measure suggestion data; The project health status description data, the task completion status list data, and the risk warning measure suggestion data are formatted to form project evaluation report data.
[0012] In one implementation of this application, obtaining project feature data and encoding the project feature data into a multi-dimensional project feature vector specifically includes: The system receives raw information about project characteristics from the project management system. This raw information includes at least the project type, project risk level, core project objectives, project duration, and the subject area to which the project belongs. The original information of the project type, project risk level, project core objective, project cycle length and the discipline to which the project belongs is standardized and classified to map the unstructured text description to a predefined set of classification labels; One-hot encoding is used to encode each category label after mapping. The feature vectors of the project type, project risk level, project core objective, project cycle length and the discipline type to which the project belongs are concatenated to form a high-dimensional sparse vector. The concatenated high-dimensional sparse vector is dimensionality reduced to transform it into a low-dimensional dense vector, generating a project feature vector containing multiple preset dimensional features.
[0013] On the other hand, embodiments of this application also provide a project progress intelligent assessment device, the device comprising: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are then executed by the at least one processor to enable the at least one processor to perform a project progress intelligent assessment method as described above.
[0014] On the other hand, this application also provides a non-volatile computer storage medium storing computer-executable instructions, which, when executed, implement the above-described intelligent project progress assessment method.
[0015] This application provides a method, device, and medium for intelligent project progress assessment, which has at least the following beneficial effects: By introducing a domain-adaptive multimodal semantic alignment principle, the problem of insufficient accuracy in associating multimodal evidence with progress milestone standards in existing technologies is solved. A dynamic weight calculation principle driven by both multi-level analysis and expert models is adopted to automatically generate an adapted weight system based on project characteristics, overcoming the subjectivity and rigidity of traditional fixed weight methods. Through the principle of full-element knowledge graph reasoning, a systematic network of relationships between tasks, evidence, resources, and results is constructed, and intelligent reasoning rules are applied to automate the verification of progress logic and resource matching, effectively solving the problems of inconsistent acceptance standards and missing dependencies, and enhancing the real-time nature of project status tracking and the accuracy of risk identification. By integrating multimodal evidence association, dynamic weight calculation, and knowledge graph reasoning, a comprehensive quantitative assessment of project health and automated report generation are achieved, comprehensively improving the efficiency, accuracy, and operability of project progress assessment. Attached Figure Description
[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating an intelligent project progress assessment method provided in this application embodiment; Figure 2 This is a schematic diagram of the internal structure of an intelligent project progress assessment device provided in an embodiment of this application. Detailed Implementation
[0017] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0018] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.
[0019] Figure 1 This is a flowchart illustrating an intelligent project progress assessment method provided in an embodiment of this application.
[0020] The analysis method involved in the embodiments of this application can be implemented by a terminal device or a server, and this application does not impose any special limitations on it. For ease of understanding and description, the following embodiments are all described in detail using a server as an example.
[0021] It should be noted that the server can be a single device or a system composed of multiple devices, i.e., a distributed server. This application does not make any specific limitations on this.
[0022] like Figure 1 As shown in the embodiment of this application, a project progress intelligent assessment method includes: Step 101: Obtain multimodal evidence data related to project progress, and based on the domain-adaptive multimodal semantic alignment principle, perform semantic alignment on the multimodal evidence data to obtain the association data between evidence and progress node standards.
[0023] In this embodiment, the acquisition of multimodal evidence data is achieved through a data interface integrated with the project management system. For example, this evidence data includes, but is not limited to, various forms of project progress documentation such as project progress report texts, experimental data tables, equipment operation status images, and operation record videos. It should be noted that this evidence data may be stored in different system modules, and the system automatically collects and standardizes preprocessing through a unified API interface.
[0024] In the domain semantic dictionary construction phase, the system builds a professional terminology database through a combination of expert annotation and project database mining. It's important to note that the expert annotation process involves domain experts manually defining core entities and their attribute descriptions based on the characteristics of different project types. For example, in scientific research project management, experts would define the entity "academic paper," with publication level attributes corresponding to specific semantic descriptions such as SCI inclusion and EI inclusion. Simultaneously, the system automatically extracts high-frequency professional terms from the historical project database and uses natural language processing technology to identify the relationships between terms, forming a complete entity-attribute-semantic description triple knowledge system. Understandably, this construction method, combining expert wisdom and data mining, ensures both the professional accuracy of the dictionary and the comprehensiveness of terminology coverage.
[0025] During the evidence feature extraction phase, the system employs differentiated processing paths for different types of evidence data. Specifically, for textual evidence data, the system uses a pre-trained language model fine-tuned for domain-specific text to extract deep semantic features. For example, the text branching model, based on the BERT architecture, is incrementally trained on professional corpora of project management, enabling it to accurately understand professional expressions in project reports and technical documents. For tabular evidence data, the system first converts it into standardized text descriptions in the form of feature values. For instance, it transforms multiple rows of data in a performance test table into continuous text descriptions such as "Accuracy: Specific Percentage, Recall: Specific Percentage," before inputting it into the text branching model for feature extraction. It should be noted that this standardization process preserves the structured information of the tabular data while making it suitable for processing by the text model.
[0026] When processing image and video evidence data, the system employs a visual branching model for feature extraction. Specifically, for image evidence, the model directly extracts the visual feature vectors of the image; for video evidence, the system first extracts keyframes, selecting the most representative frames as the processing objects. For example, in a software development project, the system extracts keyframes containing important operation steps or results from screen-recorded videos, and then encodes the features of these keyframes. Understandably, this approach ensures the effectiveness of feature extraction while avoiding the computational burden of processing all video frames.
[0027] During the semantic similarity calculation phase, the system calculates the semantic correlation between multimodal evidence features and progress node standards based on a constructed domain semantic dictionary. It's important to note that the system employs a dynamic threshold mechanism to adjust the similarity judgment criteria. This mechanism automatically fine-tunes the threshold based on project type and evidence type. For example, for basic research projects, due to the diverse forms of deliverables, the system uses a relatively lenient similarity threshold; while for application development projects, due to the clearer acceptance criteria, a stricter judgment threshold is used. Specifically, the system establishes a threshold adjustment rule base by analyzing the matching of various types of evidence and standards in historical projects. When processing new projects, the system automatically selects the most suitable threshold parameters based on the project characteristics.
[0028] Finally, the system integrates the similarity calculation results of all evidence to generate complete evidence association data. This data not only includes the degree of matching between each piece of evidence and the standard, but also records key feature information in the matching process, providing data support for subsequent weight calculation and health assessment. Understandably, this domain-semantic-based precise alignment method effectively overcomes the problem of insufficient adaptability of general multimodal models in specialized domains.
[0029] Step 102: Obtain project feature data, encode the project feature data into multi-dimensional project feature vectors, and process the project feature data based on the project dynamic weight calculation principle driven by multi-level analysis and expert models to generate project dynamic weight data.
[0030] It should be noted that the dynamic weight data in this application embodiment includes schedule dimension weight, cost dimension weight, outcome dimension weight and risk dimension weight.
[0031] In this embodiment, firstly, during the project feature data acquisition and feature vector generation stage, the system receives raw information of project feature data from the project management system. It should be noted that this raw information includes at least five key dimensions: project type, project risk level, core project objectives, project duration, and the discipline to which the project belongs. For example, project type may include specific categories such as basic research, applied research, and industry-academia-research collaboration; project risk level is typically divided into low-risk, medium-risk, and high-risk levels; and core project objectives include different types such as prioritizing results, cost control, and risk avoidance. It is understood that this raw information is initially stored in unstructured text form in different modules of the project management system, and the system automatically collects and integrates this scattered information through data interfaces.
[0032] Next, the system performs standardized classification processing on the collected raw information. Specifically, the system uses a predefined set of classification labels to map unstructured text descriptions to a unified classification system. For example, for the feature of project cycle length, the system maps the specific number of months or years to three standard categories: short-term, medium-term, and long-term; for the feature of subject type, the system maps the specific subject name to standard classifications such as single-discipline, interdisciplinary, and multidisciplinary. This standardization process ensures the comparability of features of different types of projects.
[0033] After standardizing the classification, the system uses one-hot encoding to encode each category label. One-hot encoding is a common method for converting categorical variables into binary vectors, where each category value corresponds to a position in the vector with a value of 1, and all other positions have a value of 0. Specifically, for the feature "item type," if there are three possible types, it is represented using a three-bit binary code; the risk level feature is also represented using a three-bit binary code. The system concatenates the binary vectors obtained from the one-hot encoding of each dimension's features to form a high-dimensional sparse vector. It should be noted that while this high-dimensional sparse vector completely preserves all the original feature information, its high dimensionality and the fact that most elements are zero can lead to computational inefficiency if used directly.
[0034] To address the computational efficiency issue of high-dimensional sparse vectors, the system performs dimensionality reduction on the concatenated vectors. For example, the system employs principal component analysis to transform the high-dimensional sparse vectors into low-dimensional dense vectors. Specifically, the system projects the original features onto a new feature space through a linear transformation, significantly reducing the dimensionality of the feature vectors while preserving most of the original information. The final generated project feature vectors contain multiple pre-defined dimensional features, which retain the key information of the original data while offering improved computational performance.
[0035] After obtaining the project feature vectors, the system enters the dynamic weight calculation phase. First, the system processes the encoded project feature vectors using a pre-built hybrid expert model. It's worth noting that the hybrid expert model consists of five expert sub-models and a two-layer fully connected gating network. Each expert sub-model specializes in handling a specific type of typical project scenario; for example, some experts excel at basic research projects, while others focus on application development projects. The gating network dynamically calculates the weight allocation data for each expert sub-model based on the input project feature vectors. This design allows the model to automatically select the most suitable expert combination for decision-making based on specific project characteristics, enhancing the model's adaptability.
[0036] Based on the weighted data output by the gating network, the system generates pairwise comparative scaling data for four dimensions: schedule, cost, outcomes, and risk. Specifically, a 1-9 scaling method is used for quantitative evaluation, where 1 indicates that the two dimensions are equally important, and 9 indicates that one dimension is extremely important compared to the other. For example, in outcome-oriented projects, the outcome dimension may be assigned a higher scaling value relative to the cost dimension; while in risk-sensitive projects, the risk dimension may receive a higher relative importance evaluation. Based on these pairwise comparative scaling data, the system constructs a judgment matrix that fully describes the relative importance relationships between each evaluation dimension.
[0037] Next, the system processes the judgment matrix data using the Analytic Hierarchy Process (AHP). It's important to note that the AHP is a commonly used multi-criteria decision analysis method. It verifies the logical consistency of the judgment matrix by calculating its largest eigenvalue and a consistency index. Specifically, the system first calculates the largest eigenvalue of the judgment matrix, and then calculates the consistency index and random consistency ratio based on the matrix order. Understandably, consistency verification is a crucial step in ensuring the logical rationality of expert judgments. If the consistency requirements are not met, the system will automatically prompt for a readjustment of the scaling values.
[0038] After passing consistency verification, the system calculates the initial weight data for each evaluation dimension. However, these initial weights are derived only from the characteristics of the current project and need to be optimized incorporating historical experience. Therefore, the system calibrates the initial weight data by using gradient descent to minimize historical project evaluation errors based on historical project data. For example, the system analyzes the deviation between the actual performance and evaluation results of similar historical projects and iteratively adjusts the weight parameters to make the evaluation results more consistent with reality. This mechanism of calibration based on historical data ensures that the generated dynamic weight data is both theoretically sound and has practical guiding value.
[0039] Ultimately, the dynamic weighted data output by the system comprises four components: schedule weight, cost weight, outcome weight, and risk weight. These weight values will be directly used in subsequent comprehensive project health assessments. Understandably, this weight calculation method, which combines project characteristic analysis with historical data calibration, effectively overcomes the subjectivity and rigidity of traditional fixed-weight systems.
[0040] Step 103: Based on the reasoning principle of full-element knowledge graph, construct project knowledge graph data, apply predefined reasoning rules to the project knowledge graph data, perform progress logic verification and resource matching analysis, and generate progress evaluation result data.
[0041] It should be noted that the project knowledge graph data in this application embodiment includes the attributes and relationship network of task entities, evidence entities, resource entities and outcome entities, and the progress evaluation result data includes task completion status data, evidence missing data and risk root cause data.
[0042] In this embodiment, firstly, during the project knowledge graph construction phase, the system automatically extracts four types of core entity data from the project management system. It should be noted that the task entity data includes key attributes such as task ID and acceptance criteria. The task ID uniquely identifies each task node, while the acceptance criteria define the completion conditions of the task in detail. The evidence entity data includes attributes such as evidence ID and credibility score. The evidence ID ensures the unique identifiability of each piece of evidence, while the credibility score reflects the quality level of the evidence. The resource entity data includes attributes such as skill tags and availability time. Skill tags describe the capability characteristics of the resource, and availability time indicates the resource's availability. The outcome entity data includes attributes such as outcome ID and completion status. The outcome ID uniquely identifies the project output, while the completion status records the current progress of the outcome. It can be understood that these entity data collectively constitute the basic nodes of the knowledge graph, forming a digital mapping of project elements.
[0043] After extracting entity data, the system needs to define and establish the types of relationships between entities. Specifically, tasks and deliverables are linked through requirement relationships, which clarify the specific deliverable goals that each task needs to achieve. Evidence and deliverables are linked through verification relationships, reflecting the supporting role of evidence in achieving the deliverables. Resources and tasks are linked through support relationships, reflecting the resource's ability to support task execution. Software tools and computing resources are interconnected through dependency relationships, revealing the collaborative needs between infrastructures. For example, in a software development project, the user management module development task is linked to the user authentication function deliverable through requirement relationships, which in turn is linked to the unit test report evidence through verification relationships. Meanwhile, backend development engineer resources support the execution of this task through support relationships. It should be noted that this relationship data is also automatically extracted from the project management system. By parsing management documents such as task decomposition structures and resource allocation matrices, the system can automatically identify and establish the relationship network between entities.
[0044] Next, in the knowledge reasoning application phase, the system first applies the evidence chain integrity rule to process the results. This rule requires that each result be supported by a sufficient quantity and quality of credible evidence. Specifically, the system checks the number of evidence entities associated with each result entity to ensure that at least two different types of evidence are included, and verifies whether the credibility score of each piece of evidence reaches a preset credibility threshold. For example, for the result that the algorithm model's accuracy reaches an industry-leading level, the system requires that both a performance indicator table from the test dataset and a third-party evaluation report be provided, and the credibility score of each piece of evidence must meet the passing standard. If any type of evidence is found to be missing or insufficiently credible, the system automatically generates corresponding evidence chain integrity status data, clearly indicating the missing evidence type or the substandard evidence item.
[0045] Meanwhile, the system applies cross-achievement dependency rules to handle logical dependencies between outcome entities. It's important to note that in complex projects, outcomes often have strict sequential dependencies; the achievement of a prerequisite outcome is a precondition for the realization of a subsequent outcome. Specifically, the system uses the SPARQL query language to precisely track the task-outcome-evidence dependency chain in the knowledge graph, automatically identifying dependencies between outcomes and verifying the fulfillment of dependency conditions. For example, in a research project, the completion of experimental data collection must precede the data analysis report outcome. If the system, through SPARQL querying, finds that a prerequisite outcome has not yet provided valid evidence, it will automatically classify the subsequent outcome as incomplete and generate corresponding outcome dependency status data. Understandably, this knowledge graph-based dependency tracking effectively avoids the problem of misjudgment of progress caused by unclear dependencies in traditional management.
[0046] Finally, the system integrates evidence chain integrity status data, outcome dependency status data, and dynamic weight data to conduct a comprehensive resource matching analysis. Specifically, the system checks the matching degree between resource entities and task entities, identifying problems such as skill mismatch, resource conflicts, or uneven load. For example, when the system finds that a task requires resources tagged with the "machine learning" skill, but the currently allocated resources only have the "data processing" skill, it automatically generates a resource matching warning. Simultaneously, the system prioritizes identified problems based on the relative importance of each dimension in the dynamic weight data, ensuring that critical issues and risks are addressed first. The final progress assessment results include a complete task completion status assessment, details of missing evidence, dependency violations, and resource matching suggestions, providing project managers with a comprehensive and accurate view of the progress status. It should be noted that this knowledge graph-based intelligent reasoning method not only automates the assessment of progress status but, more importantly, provides precise identification of the root causes of problems and improvement suggestions, significantly enhancing the intelligence level of project management.
[0047] Step 104: Based on the progress assessment results data, calculate the project health score data and generate the project assessment report data to output the project health score data and the project assessment report data.
[0048] It should be noted that the project health score data in this application embodiment comprehensively reflects the weighted result of schedule performance, cost performance, outcome achievement rate and risk index, and the project evaluation report data includes task completion status, evidence missing prompts and risk warning information.
[0049] In this embodiment, during the project health score calculation phase, the system needs to acquire basic assessment data across four key dimensions. It should be noted that the schedule performance data is calculated based on task completion status data. The system analyzes the ratio of completed tasks to the total task load, while also considering the completion status of critical path tasks, to comprehensively evaluate the project's execution efficiency in the time dimension. Cost performance data is calculated by comparing actual expenditures with the budget plan, combined with key cost control indicators. The calculation of outcome achievement rate data is more complex. The system not only needs to statistically analyze the proportion of achieved outcomes but also performs a weighted evaluation of outcome quality based on evidence-related data, ensuring that outcomes supported by sufficient high-quality evidence receive higher recognition. Risk index data is calculated based on risk root cause data. The system comprehensively considers the number, severity, and potential impact of identified risks to form a quantitative risk assessment result.
[0050] After obtaining the basic data across these four dimensions, the system uses dynamic weighted data for weighted fusion processing. Understandably, this dynamic weighted data is specifically generated based on project characteristics during previous processing stages, ensuring a high degree of alignment between weight allocation and project features. Specifically, the weighted fusion process employs a linear weighted summation algorithm, multiplying the score of each dimension by its corresponding weight and then summing the results to obtain a preliminary comprehensive score. For example, in a results-oriented R&D project, the weight of the results achievement rate dimension will be relatively high, and its score will have a more significant impact on overall health; while in a construction project with a tight schedule, the weight of the schedule performance dimension will be correspondingly increased. It should be noted that this mechanism of dynamically adjusting weights based on project characteristics makes the evaluation results more reflective of the project's true state.
[0051] Next, the system normalizes the weighted composite score. Specifically, the system uses a min-max normalization method to map the score to a uniform scoring range, eliminating dimensional differences caused by different project types and evaluation standards. After normalization, the system maps the numerical health score data to the corresponding health level according to predefined health level standards. For example, the system may divide the health level into multiple levels such as Excellent, Good, Attention, and Warning, each corresponding to a specific score range and status description. Understandably, this level division makes the evaluation results more intuitive and easier to understand, allowing project managers to quickly grasp the overall project status.
[0052] During the project evaluation report generation phase, the system first integrates various evaluation data using natural language generation technology. It's important to note that the natural language generation engine, based on predefined templates and rules, transforms structured evaluation data into fluent natural language descriptions. Specifically, the system combines project health score data, task completion status data, and root cause risk data to generate a description of the project's health status. For example, for a project with a health score at the "concern" level, the system might generate a description stating that the overall project progress is basically normal, but that improvements are needed in deliverable quality and risk control, further detailing specific problem areas and improvement suggestions.
[0053] Simultaneously, the system generates a list of task completion status data based on incomplete tasks and missing evidence data. This list includes not only simple task completion statistics but also a detailed analysis of the reasons for each incomplete task, especially those tasks hindered by missing or insufficient evidence. For each identified problem, the system provides specific improvement suggestions and follow-up requirements. Based on risk root cause data, the system generates risk warning measure suggestions, which not only identify existing risks but also provide targeted response strategies and preventative measures. For example, regarding the identified risk of losing key technical personnel, the system might suggest establishing a knowledge transfer mechanism, improving the document management system, and initiating a talent development program for future candidates.
[0054] Finally, the system formats and integrates all generated assessment content to form a complete project assessment report. Specifically, the system uses standardized report templates to ensure that each generated report has a consistent structure and style. The report typically includes key sections such as an executive summary, overall health assessment, detailed task analysis, risk warnings, and improvement recommendations. It's worth noting that this automated report generation process not only significantly saves time compared to manual report writing but, more importantly, ensures the consistency of assessment standards and the objectivity of the assessment results.
[0055] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide an intelligent project progress assessment device, the structure of which is as follows: Figure 2 As shown.
[0056] Figure 2 This is a schematic diagram of the internal structure of a project progress intelligent assessment device provided in an embodiment of this application. Figure 2 As shown, the device includes: At least one processor; And, a memory that is communicatively connected to at least one processor; The memory stores instructions that can be executed by at least one processor, and the instructions, when executed by at least one processor, enable at least one processor to: Acquire multimodal evidence data related to project progress, and based on the domain-adaptive multimodal semantic alignment principle, perform semantic alignment on the multimodal evidence data to obtain the association data between evidence and progress node standards; The project feature data is acquired, encoded into multi-dimensional project feature vectors, and processed based on the dynamic weight calculation principle driven by multi-level analysis and expert models to generate dynamic weight data for the project. The dynamic weight data includes schedule dimension weights, cost dimension weights, outcome dimension weights, and risk dimension weights. Based on the reasoning principle of full-element knowledge graph, a project knowledge graph data is constructed, and predefined reasoning rules are applied to the project knowledge graph data to perform progress logic verification and resource matching analysis, generating progress assessment result data. The project knowledge graph data includes the attributes and relationship networks of task entities, evidence entities, resource entities, and outcome entities, and the progress assessment result data includes task completion status data, evidence missing data, and risk root cause data. Based on the progress assessment results, the project health score is calculated, and the project assessment report is generated to output the project health score and the project assessment report. The project health score comprehensively reflects the weighted results of progress performance, cost performance, outcome achievement rate, and risk index. The project assessment report includes task completion status, evidence missing prompts, and risk warning information.
[0057] This application also provides a non-volatile computer storage medium storing computer-executable instructions, which, when executed, can: Acquire multimodal evidence data related to project progress, and based on the domain-adaptive multimodal semantic alignment principle, perform semantic alignment on the multimodal evidence data to obtain the association data between evidence and progress node standards; The project feature data is acquired, encoded into multi-dimensional project feature vectors, and processed based on the dynamic weight calculation principle driven by multi-level analysis and expert models to generate dynamic weight data for the project. The dynamic weight data includes schedule dimension weights, cost dimension weights, outcome dimension weights, and risk dimension weights. Based on the reasoning principle of full-element knowledge graph, a project knowledge graph data is constructed, and predefined reasoning rules are applied to the project knowledge graph data to perform progress logic verification and resource matching analysis, generating progress assessment result data. The project knowledge graph data includes the attributes and relationship networks of task entities, evidence entities, resource entities, and outcome entities, and the progress assessment result data includes task completion status data, evidence missing data, and risk root cause data. Based on the progress assessment results, the project health score is calculated, and the project assessment report is generated to output the project health score and the project assessment report. The project health score comprehensively reflects the weighted results of progress performance, cost performance, outcome achievement rate, and risk index. The project assessment report includes task completion status, evidence missing prompts, and risk warning information.
[0058] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.
[0059] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.
Claims
1. A method for intelligently assessing project progress, characterized in that, The method includes: Acquire multimodal evidence data related to project progress, and based on the domain-adaptive multimodal semantic alignment principle, perform semantic alignment on the multimodal evidence data to obtain the association data between evidence and progress node standards; The project feature data is acquired, encoded into a multi-dimensional project feature vector, and processed based on the dynamic weight calculation principle of project dynamic weight driven by multi-level analysis and expert model to generate dynamic weight data of the project; the dynamic weight data includes schedule dimension weight, cost dimension weight, outcome dimension weight and risk dimension weight. Based on the reasoning principle of full-element knowledge graph, a project knowledge graph data is constructed, and predefined reasoning rules are applied to the project knowledge graph data to perform progress logic verification and resource matching analysis, generating progress evaluation result data. The project knowledge graph data includes the attributes and relationship networks of task entities, evidence entities, resource entities, and outcome entities, and the progress evaluation result data includes task completion status data, evidence missing data, and risk root cause data. Based on the progress assessment results, the project health score is calculated, and the project assessment report is generated to output the project health score and the project assessment report. The project health score comprehensively reflects the weighted result of progress performance, cost performance, outcome achievement rate, and risk index. The project assessment report includes task completion status, evidence missing prompts, and risk warning information.
2. The intelligent project progress assessment method according to claim 1, characterized in that, Based on the principle of domain-adaptive multimodal semantic alignment, the multimodal evidence data is semantically aligned to obtain the association data between the evidence and the progress node standard, specifically including: Domain semantic dictionary data is constructed using expert annotations and professional terminology from the project database. For text-based evidence data, a pre-trained text branching model is used to extract deep semantic feature data. The domain semantic dictionary data includes triples of entity, attribute, and semantic description. For tabular evidence data, the tabular data is converted into standardized text description data in the form of feature values, and the standardized text description data is processed using the text branching model to extract semantic feature data. Image evidence data and video evidence data are processed through the visual branching model. Based on the domain semantic dictionary data, the semantic similarity data between the multimodal evidence data and the progress node standard is calculated, and a dynamic threshold mechanism is applied to adjust the similarity judgment threshold to obtain evidence association data; the dynamic threshold mechanism automatically fine-tunes the threshold according to the project type and evidence type.
3. The intelligent project progress assessment method according to claim 1, characterized in that, Based on the principle of dynamic weight calculation for projects driven by both multi-level analysis and expert models, the project feature data is processed to generate dynamic weight data for the projects, specifically including: The project feature data is encoded, and the encoded project feature vector is processed by a pre-built hybrid expert model to dynamically calculate the weight allocation data of each expert sub-model through a gating network; the encoding process uses the One-Hot method to process the dimensions of project type, risk level, core objective, project cycle and discipline type; Based on the weighted data, pairwise comparison scale data for the dimensions of schedule, cost, outcome and risk are generated, and quantitative evaluation is carried out using the 1-9 scale method to construct judgment matrix data; The maximum eigenvalue and consistency index data of the judgment matrix data are calculated using the analytic hierarchy process (AHP) to verify the consistency of the judgment matrix data and to calculate the initial weight data. Based on historical project data, and using gradient descent to minimize historical project evaluation errors, the initial weight data is calibrated to generate dynamic weight data.
4. The intelligent project progress assessment method according to claim 1, characterized in that, Based on the reasoning principle of full-element knowledge graphs, a project knowledge graph data is constructed, specifically including: Extract task entity data, evidence entity data, resource entity data, and outcome entity data from the project management system to construct project knowledge graph data; The task entity data includes a task ID and acceptance criteria attribute; the evidence entity data includes an evidence ID and credibility score attribute; the resource entity data includes skill tags and available time attribute; and the result entity data includes a result ID and completion status attribute. Define the types of relationships between entities and extract the associated data of tasks, evidence and results from the project management system to construct knowledge graph data; The relationship types include the demand relationship between tasks and results, the verification relationship between evidence and results, the support relationship between resources for tasks, and the dependency relationship between software tools and computing resources.
5. The intelligent project progress assessment method according to claim 1, characterized in that, Applying predefined inference rules to the project knowledge graph data, performing schedule logic verification and resource matching analysis, and generating schedule evaluation result data, specifically including: The results evidence chain integrity rules are applied to process the results entity data and evidence entity data to generate evidence chain integrity status data; the results evidence chain integrity rules require at least two types of related evidence and the credibility score of each piece of evidence is not lower than a preset credibility threshold. Apply cross-achievement dependency rules to handle logical dependencies between achievement entities, trace dependency chains through SPARQL queries, and generate achievement dependency status data; By integrating the evidence chain integrity status data, the outcome dependency status data, and the dynamic weight data, resource matching analysis is performed to generate progress assessment result data.
6. The intelligent project progress assessment method according to claim 1, characterized in that, Based on the progress assessment results, the project health score is calculated, specifically including: Acquire schedule performance data, cost performance data, outcome achievement rate data, and risk index data, and use dynamic weight data to weight and integrate the schedule performance data, cost performance data, outcome achievement rate data, and risk index data; The progress performance data is calculated based on task completion status data, the achievement rate data is calculated based on evidence correlation data, and the risk index data is calculated based on risk root cause data. The weighted fusion process uses a linear weighted summation algorithm. The weighted comprehensive score is calculated as the project health score data. The health score data is normalized and mapped to a predefined health level standard to generate project health level data.
7. The intelligent project progress assessment method according to claim 6, characterized in that, The data used to generate the project evaluation report includes: By integrating the project health score data, the task completion status data, and the risk root cause data, a project health status description data is generated using natural language generation technology. Based on the data of incomplete tasks and missing evidence, generate a list of task completion status data, and based on the risk root cause data, generate risk warning measure suggestion data; The project health status description data, the task completion status list data, and the risk warning measure suggestion data are formatted to form project evaluation report data.
8. The intelligent project progress assessment method according to claim 1, characterized in that, Obtain project feature data and encode the project feature data into a multi-dimensional project feature vector, specifically including: The system receives raw information about project characteristics from the project management system. This raw information includes at least the project type, project risk level, core project objectives, project duration, and the subject area to which the project belongs. The original information of the project type, project risk level, project core objective, project cycle length and the discipline to which the project belongs is standardized and classified to map the unstructured text description to a predefined set of classification labels; One-hot encoding is used to encode each category label after mapping. The feature vectors of the project type, project risk level, project core objective, project cycle length and the discipline type to which the project belongs are concatenated to form a high-dimensional sparse vector. The concatenated high-dimensional sparse vector is dimensionality reduced to transform it into a low-dimensional dense vector, generating a project feature vector containing multiple preset dimensional features.
9. A project progress intelligent assessment device, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform a project progress intelligent assessment method as described in any one of claims 1-8.
10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, When the computer-executable instructions are executed, they implement a project progress intelligent assessment method as described in any one of claims 1-8.