Project group performance intelligent assessment method and device

Abstract

The invention relates to the technical field of artificial intelligence and engineering project management, and particularly discloses a project group performance intelligent assessment method and device. The method is especially suitable for a project group management scene, and examination work is carried out from multiple dimensions of task completion, organization coordination, cross-project collaboration, research, development, innovation and the like. The method comprises the following steps: acquiring three types of basic data including characters, images and structured data; semantic decoding is carried out through a large language model, image recognition is carried out through a convolutional neural network, structured data is processed through a standardized algorithm, and multi-source data preprocessing is achieved; extracting static features based on a multi-layer sensing network, and extracting dynamic time sequence features based on a long short-term memory network; and a big language model is utilized to generate an interpretive performance comment, and finally a comprehensive assessment report is output. The system is fused with a multi-mode AI model, has high scoring accuracy and high interpretability, and is particularly suitable for intelligent performance evaluation under a multi-project collaborative management situation.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and engineering project management technology, and in particular to a method and apparatus for intelligent performance evaluation of project portfolios. Background Technology

[0002] With the continuous advancement of national large-scale infrastructure construction, project portfolio management has been widely applied in numerous complex engineering projects. A project portfolio, as a collection of interconnected engineering projects with coordinated resources and objectives, plays a crucial role in large-scale engineering scenarios such as cascade hydropower development and the construction of clean energy bases integrating hydropower, wind power, solar power, and energy storage. This management model effectively integrates resources, coordinates schedules, achieves collaborative and efficient management among projects, and improves overall project performance.

[0003] Project portfolio management involves complex project types, numerous stakeholders, dispersed data sources, and multiple assessment dimensions, placing extremely high demands on performance evaluation. Traditional manual or decentralized scoring mechanisms are insufficient to meet the needs of refined and transparent management. Existing assessment methods generally suffer from problems such as strong data subjectivity, slow response to dynamic changes, low adaptability of assessment indicators, and difficulty in horizontal comparison between projects. Therefore, constructing a performance intelligent assessment system based on artificial intelligence technology that is adaptive, interpretable, and adaptable to the organizational structure of project portfolios has significant practical implications.

[0004] The rapid development of technologies such as artificial intelligence, big data, and multimodal learning has provided intelligent technical support for performance evaluation in complex scenarios. For example, natural language processing can assist in parsing project logs and meeting minutes, computer vision can identify anomalies in construction site images, and deep learning models can integrate multi-source data and achieve intelligent scoring. These technologies lay a solid foundation for building intelligent performance evaluation systems for project portfolios.

[0005] Currently, technologies related to performance evaluation indicator systems for engineering projects mainly focus on indicator design, online evaluation process design, and automation of quantitative indicator calculation rules. In terms of performance evaluation technology, existing achievements often involve evaluation system design, evaluation rule design, and intelligent scoring methods. However, these technologies have not yet been able to customize performance evaluation to suit the specific characteristics of engineering projects, nor have they fully considered multi-dimensional performance evaluation factors, and they rely too heavily on objective data. Summary of the Invention

[0006] The present invention aims to at least partially solve one of the technical problems in the related art.

[0007] Therefore, the first objective of this invention is to propose an intelligent performance evaluation method for project groups.

[0008] The second objective of this invention is to provide an intelligent performance evaluation device for project groups.

[0009] The third objective of this invention is to provide an electronic device.

[0010] The fourth objective of this invention is to provide a computer-readable storage medium.

[0011] The fifth objective of this invention is to provide a computer program product.

[0012] To achieve the above objectives, a first aspect of the present invention proposes an intelligent performance evaluation method for project portfolios, comprising: S1 collects text data, image data, and structured data in project group management scenarios, and automatically identifies the sub-project code and role dimension to which the data belongs through the data base, so as to achieve multi-level data alignment and standardized integration; S2 uses a large language model to perform semantic decoding on text data and generate binary event feature vectors, uses a convolutional neural network to perform target detection and behavior recognition on image data and generate behavior category feature vectors, and uses a standardization algorithm to perform dimensionless processing on structured data and generate standardized feature vectors. S3 maps the three types of feature vectors to a unified dimension through a fully connected layer and then concatenates them to separate static features and temporal features, which are then input into a multilayer perceptron and a long short-term memory network for feature extraction. S4, based on the static feature score and temporal feature score output by the multilayer perceptron, adopts a trainable weighted fusion strategy to generate a comprehensive score, and inputs the score results and business data into the large language model to generate interpretable comments. The model parameters are dynamically updated through manual review to achieve closed-loop optimization.

[0013] Optionally, the process of collecting text data, image data, and structured data in the project group management scenario, and automatically identifying the sub-project code and role dimension to which the data belongs through a data foundation to achieve multi-level data alignment and standardized integration, also includes: The data base automatically identifies the sub-project code, construction section, and role dimension to which the data belongs, and aligns the multi-source data with the project group organizational structure at multiple levels based on preset hierarchical mapping rules. Implement a data traceability mechanism for the collected multi-source data, recording the source timestamp, collection device identifier, and associated responsible person information for each piece of data to ensure data traceability.

[0014] Optionally, the step of semantically decoding the text data using a large language model and generating binary event feature vectors further includes: The semantic relevance of each text to each event template is calculated using a pre-defined template dictionary. When the relevance score exceeds the threshold of 0.6, the corresponding binary event features are concatenated with the original semantic embedding vector to form a fused feature vector.

[0015] Optionally, the step of mapping the three types of feature vectors to a unified dimension through a fully connected layer and then concatenating them, separating static features from temporal features, and inputting them separately into a multilayer perceptron and a long short-term memory network for feature extraction, further includes: When mapping the dimensions of the three types of feature vectors through a fully connected layer, a trainable linear transformation parameter matrix is ​​used to process the text feature vector, behavioral feature vector, and structured feature vector respectively, so that the dimensions of the mapped feature vectors meet the input requirements of MLP and LSTM.

[0016] Optionally, the static feature score and temporal feature score based on the output of the multilayer perceptron network are used to generate a comprehensive score by employing a trainable weighted fusion strategy, which further includes: The fusion weight parameters are obtained using the gradient descent algorithm. and Dynamic optimization is performed, among which The weights corresponding to the static feature scores, The weights corresponding to the time-series feature scores, and This is to minimize the error between the overall score and the results of manual review.

[0017] Optional, also includes: The final confirmed assessment results will be used to generate multi-dimensional visualization charts, including task completion trend charts, cross-project collaboration heatmaps, and R&D innovation radar charts. These charts and structured reports will be pushed to the project group management platform to enable multi-terminal display and real-time updates of the assessment results.

[0018] To achieve the above objectives, a second aspect of the present invention provides an intelligent performance evaluation device for project clusters, comprising: The first execution module is used to collect text data, image data and structured data in the project group management scenario, and automatically identify the sub-project code and role dimension to which the data belongs through the data base to achieve multi-level data alignment and standardized integration; The second execution module is used to perform semantic decoding on text data and generate binary event feature vectors through a large language model, perform target detection and behavior recognition on image data through a convolutional neural network and generate behavior category feature vectors, and perform dimensionless processing on structured data through a standardization algorithm and generate standardized feature vectors. The third execution module is used to map the three types of feature vectors to a unified dimension through a fully connected layer and then concatenate them, separating static features and temporal features and inputting them into a multilayer perceptron and a long short-term memory network for feature extraction. The fourth execution module is used to generate a comprehensive score based on the static feature score and temporal feature score output by the multilayer perceptron network, using a trainable weighted fusion strategy. The score results and business data are then input into a large language model to generate interpretable comments. The model parameters are dynamically updated through manual review to achieve closed-loop optimization.

[0019] To achieve the above objectives, a third aspect of the present invention provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of the first aspects.

[0020] To achieve the above objectives, a fourth aspect of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of the first aspects.

[0021] To achieve the above objectives, a fifth aspect of the present invention provides a computer program product that, when executed by a processor, implements the method described in any one of the first aspects.

[0022] The technical solutions provided by the embodiments of the present invention bring at least the following beneficial effects: (1) Improve the adaptability and completeness of the assessment system: By building a data foundation with the ability to identify project group structure, it supports the standardized integration of multiple types of data and realizes the refined assessment of multiple roles and multiple projects in complex engineering scenarios.

[0023] (2) Enhance the accuracy and interpretability of the scoring results: By combining large language models and deep learning algorithms and integrating multimodal data input, interpretable scoring results and structured comments are output, which effectively improves the transparency and management usability of the assessment.

[0024] (3) Realize the intelligentization of the performance appraisal process: the appraisal task is automatically pushed, the scoring results are manually reviewed and the reasons are recorded, forming a process of "AI scoring - manual review - report issuance", which significantly improves the credibility and enforceability of the appraisal results.

[0025] (4) Support model self-learning and continuous system optimization: The final scoring results are fed back to the AI ​​training system to realize the adaptive update of the scoring model and ensure the stability and technological advancement of the system in dynamic management scenarios.

[0026] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0027] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A one-way text flowchart illustrating a project portfolio performance intelligent assessment method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of multiple logic steps for an intelligent performance evaluation method for project groups provided in an embodiment of the present invention. Figure 3 This is a flowchart illustrating the execution process of the AI-assisted scoring algorithm provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of the structure of an intelligent performance evaluation device for project groups provided in an embodiment of the present invention. Detailed Implementation

[0028] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0029] Existing technical approaches primarily focus on building assessment systems and scoring methods for single projects, making it difficult to effectively adapt to the complex organizational structures and data characteristics of project portfolios. Furthermore, they have not yet effectively leveraged the auxiliary capabilities of artificial intelligence to process and analyze management data in different formats. Specific technical bottlenecks are as follows: (1) The performance evaluation rule base is simple and the calculation method is simple: Most systems use a linear weighted model to score the preset indicators, which makes it difficult to reflect the interaction and dynamic change characteristics between indicators and cannot meet the complex performance evaluation needs under the multi-dimensional collaboration of project groups.

[0030] (2) Insufficient support for data types, making it difficult to integrate structured and unstructured information: Existing systems rely heavily on structured tabular data and cannot effectively parse unstructured data such as construction logs, images, and audio-visual materials, resulting in incomplete assessment criteria and low level of intelligence.

[0031] (3) Lack of feedback loop mechanism makes it difficult to achieve model iterative optimization: Existing methods usually take the score as the output endpoint and lack a learning mechanism based on AI score and manual correction results, making it difficult to achieve continuous improvement of system capabilities and enhancement of model adaptability.

[0032] (4) The output of assessment results is monotonous and lacks interpretability and diverse reporting functions: Most systems only provide a single score and fail to output detailed comments and suggestions, which is not conducive to managers' decision-making.

[0033] (5) The assessment process is not systematic enough and cannot meet the needs of the whole process management of the project group: The current system assessment process design is mainly based on single projects and fails to cover the assessment needs of multiple levels, multiple roles and multiple responsibilities of the project group. The process is not standardized and traceable enough.

[0034] (6) The integration of artificial intelligence and human judgment is not high, which affects the fairness and acceptability of the assessment: the existing system either focuses on human scoring or relies entirely on automatic scoring by models, lacks an interactive review mechanism, and makes it difficult to establish a credible and transparent assessment mechanism.

[0035] To address several shortcomings in existing performance evaluation methods for engineering project clusters, this invention aims to provide an intelligent evaluation method and apparatus suitable for complex project cluster organizational structures. Specifically, it addresses the following key issues: how to construct a multi-dimensional performance evaluation model applicable to project cluster organizational structures to achieve differentiated evaluation for different project types and responsibilities, enhancing the universality and scalability of the evaluation system; how to integrate heterogeneous and multi-source data types (structured, text, images, etc.) within the project cluster, and effectively clean, standardize, and correlate them to support high-quality intelligent scoring model training and interpretation; and how to design an AI-assisted scoring algorithm with self-learning capabilities. The document outlines several key aspects: First, how to enable the assessment model to dynamically absorb and iteratively optimize historical data and human feedback, thereby enhancing its stability and foresight. Second, how to generate assessment results that are both interpretable and valuable for management, breaking through the limitations of traditional scoring methods and providing highly readable and actionable performance reports. Third, how to construct a comprehensive assessment task generation and review mechanism to achieve closed-loop management from task submission, AI scoring, human review to result archiving, ensuring process compliance and result traceability. Fourth, how to organically integrate artificial intelligence and human review to improve scoring efficiency while ensuring the rationality, transparency, and internal organizational acceptance of assessment results.

[0036] This invention is based on engineering management logic and uses artificial intelligence methods as tools to systematically improve the intelligence, adaptability and credibility of project portfolio performance evaluation, serving the scientific decision-making and resource allocation of large-scale engineering organizations.

[0037] The process of the technical solution of this invention is as follows: Figure 1 and Figure 2 As shown, the specific details of each step are as follows: S1 collects textual, image, and structured data from project group management scenarios, and automatically identifies the sub-project codes and role dimensions to which the data belongs through the data base, achieving multi-level data alignment and standardized integration.

[0038] In one embodiment of the present invention, a data foundation is used to collect and integrate various basic engineering management data within a project group. Specifically, this includes textual data (e.g., project construction logs, quality inspection reports, equipment operation records, meeting minutes, etc.), image data (e.g., on-site construction photos, video surveillance data, drone inspection footage, etc.), and structured data (e.g., personnel attendance records, on-site location information, construction safety inspection data, project numbers, project types, engineering levels, geographical distribution, academic disciplines, etc.). The data foundation possesses project group structure recognition capabilities, automatically identifying the sub-project number, construction section, role dimension, etc., to which the data belongs. This enables multi-level alignment and standardized integration of data from multiple sources, ensuring the authenticity, traceability, and summarization of the basic data.

[0039] Furthermore, based on the roles of personnel within the project group, personnel are divided into multiple assessment role groups (including project managers, on-site construction personnel, safety management personnel, quality management personnel, equipment maintenance personnel, technical management personnel, etc.); and the responsibility scope of each role is dynamically associated with its respective sub-project group, automatically matching the corresponding performance assessment item list to achieve personalized assessment of sub-projects, and pushing assessment tasks to the corresponding reviewers. From an integrated perspective, this enables penetrating management of all sub-projects within the project group.

[0040] S2 performs semantic decoding on text data using a large language model and generates binary event feature vectors, performs target detection and behavior recognition on image data using a convolutional neural network and generates behavior category feature vectors, and performs dimensionless processing on structured data using a standardization algorithm and generates standardized feature vectors.

[0041] In one embodiment of the present invention, a multimodal deep learning scoring model is constructed and trained based on past project group management data and historical performance scoring results (past manual scoring data), and then intelligently analyzes the current project group management data to generate a comprehensive scoring result including performance rating, specific score, and assessment comments.

[0042] Reference Figure 3 The use of AI-assisted scoring models involves two steps: model training and model usage.

[0043] (A) During the model training phase, a training data pool is constructed using historical data samples' actual performance scores as the prediction target and past project group management data as input. By minimizing the scoring error, the parameters and weights of the aforementioned deep learning model are optimized to ensure optimal scores for each item. And the overall score (Score) achieves optimal prediction accuracy.

[0044] (B) In the model usage phase, based on the parameters obtained from model training, the current project group management data is input to predict the evaluation scores for each dimension of the person being evaluated, as well as the overall score.

[0045] The specific construction method of the model is as follows: (S21) Project Group Data Input: First, introduce "sub-project codes" into the data base to classify the management data of different sub-projects. Second, input the text data, image data, and structured data mentioned in step S1.

[0046] (S22) Data preprocessing: (S221) Text Data Processing: For multi-source text data (including construction logs, meeting minutes, quality reports, etc. related to the assessed personnel) in project portfolio management scenarios, the DeepSeek V3 large language model is used to achieve semantic decoding and targeted feature extraction. The specific processing flow is as follows: First, the raw text data is segmented, denoised, and standardized in terms of notation to ensure that all input text has a consistent format and high-quality input. Then, each standardized text is processed... Encoding is performed using DeepSeek V3 to obtain a high-dimensional semantic embedding vector, denoted as:

[0047] in, Representing text semantic vectors, For embedding spatial dimensions.

[0048] For all text embedding vectors, this embodiment of the invention includes a built-in template dictionary covering key performance events such as "project risks," "process anomalies," "safety hazards," "partnerships," "project performance," "information feedback," "technical non-standards," and "schedule delays," and calculates the relevance of each text to each event template. The specific calculation is as follows:

[0049] in, For the first An event template is embedded. When Greater than the set threshold When the value is 0.6, the text is considered to be... Activated the first Class characteristics.

[0050] The activation features of each text will be concatenated with the original embedding vector as a binary feature vector to form the final feature output:

[0051] in, , No. Dimension indicates the first Whether the event-like characteristics have been detected.

[0052] To address the issue of inconsistent amounts of associated text among different assessors, this system employs a pooling strategy for feature normalization. All of the assessments for each person (common The text is then subjected to average pooling (or max pooling), i.e.:

[0053] Through pooling, all examinees ultimately obtain feature vectors of uniform length. This ensures that the feature dimensions of the input data are consistent with those of the downstream deep learning model, facilitating batch training and efficient inference. Finally, all aggregated feature vectors are uniformly standardized using Z-score to ensure consistent input data dimensions and improve the stability and generalization ability of the model training.

[0054] (S222) Image data: ResNet50 convolutional neural network (CNN) is used to perform target detection and behavior recognition on image data (including photos, surveillance video frames, drone inspection images, etc.), such as wearing safety helmets, crossing the line, etc., and output feature tensors.

[0055] First, image data standardization preprocessing is performed, which involves normalizing the size, standardizing the color, and removing noise from the original image data, adjusting each frame of the image to the uniform format required by the model input (i.e., 224×224 pixels, RGB three channels).

[0056] Next, target detection and behavior recognition are performed by inputting the preprocessed image into the ResNet50 network. Let the original image be... After processing by the convolution and residual modules of ResNet50, a high-dimensional feature tensor is obtained. :

[0057] in, (C is the number of channels, and H and W are the feature map space dimensions).

[0058] Subsequently, behavior detection and classification are performed. In the object detection stage, a combination of Faster R-CNN and a ResNet50 backbone is used to output candidate bounding boxes and their corresponding class probabilities. Taking G detected objects as an example, the output is:

[0059] in, Indicates the first The location box of the target. It is the probability vector of its category (such as "wearing a safety helmet", "working across the line").

[0060] Based on this, this embodiment of the invention uses facial recognition for identity verification, automatically identifying and confirming the "person being assessed" for each detected human target. Specifically, for each... Region, extract feature vectors and facial feature vectors The latter is compared with the feature database of the person being evaluated. If the similarity score exceeds the preset threshold, the target is determined to be the "person being evaluated" and the next step is carried out.

[0061] Action recognition targets local region features and uses fully connected layers and classification layers to classify the behavior of each detected target region, with scores as follows:

[0062] in, and The classification parameters obtained during training, This represents the probability distribution of the assessed individual in each type of behavior or state. To achieve structured and automated judgment, the system addresses each safety or operational standard category. ( =1,2,…, All of them have a discrimination threshold set. (Take 0.6). The probability output by softmax. When the system determines that the person being assessed exhibits a behavior of that category (assigning a value of 1), it determines that the person does not exhibit that behavior (assigning a value of 0). Through this method, a final summary vector can be generated. :

[0063] in, This indicates that the person being assessed exhibits behavior of category K. This indicates that it does not exist. This summary vector... It is an automatic identification and quantitative assessment of the multi-dimensional behavioral states of the person being assessed, based on images from a specific chapter. It provides the comprehensive feature vector of the person being assessed over a given time period. The summation of the feature vectors of all images for the person being assessed is calculated using the following formula:

[0064] Finally, all the aggregated feature vectors are uniformly standardized using Z-score to ensure that the input data dimensions are consistent.

[0065] (S223) Structured Data Processing: In this embodiment of the invention, for the structured data of the individuals being assessed during the project management process (such as individual attendance days, working hours, performance scores, training status, equipment operation records, etc.), to improve the scientific nature of data analysis and the effectiveness of subsequent model training, the Z-score standardization method is uniformly used to process all individual feature data into dimensionless form. The specific steps are as follows: All individual structured data were processed using Z-score normalization to ensure that the mean of each feature was 0 and the standard deviation was 1. This eliminated the influence of different dimensions between features and improved the stability of subsequent multimodal fusion and model training. The normalization calculation formula is as follows:

[0066] in, For the person being assessed in the first Standardized values ​​of feature fields These are the original values. and denoted as the mean and standard deviation of the j-th feature field, respectively.

[0067] After the above standardization process, each individual being assessed possesses a set of dimensionless structured feature vectors. It can be directly used as input for subsequent multimodal feature fusion and intelligent performance evaluation models.

[0068] S3 maps the three types of feature vectors to a unified dimension through a fully connected layer and then concatenates them to separate static features and temporal features, which are then input into a multilayer perceptron and a long short-term memory network for feature extraction.

[0069] To comprehensively reflect the multi-dimensional performance of individuals being assessed in project portfolio management, this embodiment of the invention integrates three types of features—structured data features, text data features, and image data features—to form a standardized fused feature vector, which is divided into static factors ( ) and time series factors ( This serves as input for the subsequent model analysis phase. The specific process is as follows: After text feature processing, the text feature embedding vector of each individual being assessed is obtained. This includes high-dimensional semantic embedding and binary event labels. Image data undergoes object detection and behavior recognition to obtain a behavior category summary vector. After the structured data is organized and aggregated, a standardized structured feature vector for each individual is obtained. .

[0070] Since the original dimensions and numerical distributions of the three types of features may differ, they need to be mapped to a unified dimension first through a fully connected layer (linear transformation). .

[0071]

[0072]

[0073] in, and For mapping parameters, , , The feature vectors are standardized and aligned from three different categories. These standardized and aligned feature vectors are then concatenated in sequence to form the final fused feature representation of the individual being assessed.

[0074] Where “[ ; ]” represents a vector concatenation operation. This is a multimodal fusion feature vector that comprehensively reflects an individual's performance status across text, images, and structured data.

[0075] In this process, static and temporal features are separated and concatenated. Static factors that do not change over time (such as personal background information, project affiliation, project size, engineering level, etc.) are directly concatenated. For data with time-series attributes (such as unsafe behaviors at different times, attendance, number of training sessions attended, number of meetings attended, etc.), the time-series features are concatenated as a sequence and input into the subsequent time-series modeling network. .

[0076] Furthermore, this embodiment of the invention analyzes the static structural features of a project based on a standard Multilayer Perceptron (MLP) model. A Long Short-Term Memory (LSTM) network is used to extract performance evolution trends over time, suitable for handling time-series features such as periodic progress, quality fluctuations, and safety incident records. Both models output predicted score vectors. and The details are as follows: (1) Multilayer perceptron (MLP) analysis of static features: static feature vector The input is a multilayer perceptron (MLP), which consists of several fully connected layers and activation functions. The specific calculation process is as follows: Level 1:

[0077] Level 2:

[0078] And so on, until the last floor, for a total of 4 floors.

[0079] Output score vector:

[0080] in, , , ..., and , , ..., All of these are model parameters, and ReLU(·) represents the activation function.

[0081] (2) Analysis of Long Short-Term Memory (LSTM) Networks with Temporal Features: Time-series feature sequences The input is fed into an LSTM network, which can capture the dynamic evolution over time. The calculation process is as follows: Each time step t:

[0082] in, For the first Input features at each time step, In hidden state, This is the unit state.

[0083] Final score vector output: Take the hidden state at the last moment. And output through a fully connected layer:

[0084] S4, based on the static feature score and temporal feature score output by the multilayer perceptron, adopts a trainable weighted fusion strategy to generate a comprehensive score, and inputs the score results and business data into the large language model to generate interpretable comments. The model parameters are dynamically updated through manual review to achieve closed-loop optimization.

[0085] This invention embodiment uses static feature scoring ( ) and time series feature score ( The weighted sum is then used to obtain the final performance score:

[0086] in, and To integrate weight parameters, satisfy and Weight parameters and The AI-assisted scoring model is automatically optimized during model training. It is trained and predicted separately for four aspects: task completion, organizational coordination, cross-project collaboration, and R&D and innovation. Specifically, the outputs of the AI-assisted scoring are as follows: task completion rating and score (Score_t, Class_t), organizational coordination rating and score (Score_o, Class_o), cross-project collaboration rating and score (Score_c, Class_c), R&D and innovation rating and score (Score_r, Class_r), and overall rating and score (Score, Class). Ratings are based on score ranges, and the overall score is calculated using weighted averages determined by experts.

[0087] Finally, the input data, including scores for each dimension, rating levels, overall score and level, and relevant business descriptions, is fed into the DeepSeek V3 language model. The model provides corresponding prompts, and through its semantic understanding and generation capabilities, automatically generates logically clear, content-rich, and business-relevant performance reviews. These reviews include, but are not limited to, evaluations of each dimension, strengths, weaknesses, and suggestions for improvement.

[0088] The system generates assessment results and outputs the final score and report, including the following four core contents: task completion rating and specific score (Score_t, Class_t), organization and coordination rating and specific score (Score_o, Class_o), cross-project collaboration rating and specific score (Score_c, Class_c), R&D and innovation rating and specific score (Score_r, Class_r), comprehensive rating and specific score (Score, Class), and written evaluation.

[0089] Furthermore, this embodiment of the invention also supports manual review. Specifically, for each score, rating, or evaluation item on the assessment interface, two text boxes are set. One text box is labeled "AI Evaluation," and the other text box is not specifically labeled and is used to fill in the final actual score, rating, or evaluation content. After the reviewer enters the assessment interface, the AI ​​score, rating, or evaluation will be automatically filled into the corresponding text box (i.e., the text box labeled "AI Evaluation"). The reviewer can manually review the relevant materials of the person being assessed, make subjective judgments, and compare and judge with the results in the "AI Evaluation" column. If it is consistent with the results in the "AI Evaluation" column, the "Accept" button can be clicked, and the results in the "AI Evaluation" column will be automatically filled into the other text box; if it is inconsistent, the actual score, rating, or evaluation content can be filled in the other text box. An "Explanation" window is provided at the bottom of the assessment interface to collect the reasons for the reviewer's modification of the AI ​​score. Finally, the completed review results must be digitally signed and certified before they can be submitted to the system, realizing the standardization and traceability of the review process.

[0090] The final confirmed assessment results are automatically generated into a structured performance report, including a task completion trend chart, a cross-project collaboration heatmap, and a research and development innovation radar chart. These charts and the structured report are then pushed to the project group management platform to enable multi-terminal display and real-time updates of the assessment results.

[0091] It should be noted that the task completion trend chart uses time-series visualization technology to display the changes in the task completion score of the assessed person at different time points in the form of a line chart or area chart, and supports dynamic switching by granularity such as week, month, and project stage. The cross-project collaboration heatmap uses a matrix heatmap to display the frequency and quality score of collaboration between the assessed person and other project members or teams, where the color intensity represents the collaboration strength, and the axes correspond to project number and person ID respectively. The R&D innovation radar chart uses a polar coordinate chart to display the assessed person's comprehensive innovation ability around the dimensions of "technological innovation, problem solving, process optimization, and knowledge contribution", and the score of each dimension is visualized by the side length of a polygon.

[0092] From a technical perspective, this step significantly improves the readability and operability of performance appraisal results through multi-dimensional visualization, enabling managers to quickly identify performance strengths, weaknesses, and potential problems. Simultaneously, the synchronized delivery mechanism of structured reports and charts ensures the consistency and completeness of appraisal information, providing data-driven decision support for project portfolio performance management. This technical solution not only enhances the timeliness and intuitiveness of performance feedback but also demonstrates the intelligent output capabilities of the AI-assisted scoring system, making it a crucial component in achieving closed-loop performance management for project portfolios.

[0093] It is particularly important to note that once all scores are determined, the embodiments of the present invention will enter the training set of the AI-assisted scoring algorithm (i.e., the "data pool" mentioned above) for subsequent model optimization training, dynamically updating the model parameters and structure of the AI-assisted scoring algorithm; thereby enabling the model to continuously adapt and improve performance as management scenarios change and data accumulates, ensuring that the system maintains its technological leadership and scoring accuracy in the long term.

[0094] To achieve the above embodiments, the present invention also proposes an intelligent performance evaluation device for project groups. Figure 4 This is a schematic diagram of the structure of an intelligent performance evaluation device for project groups provided in an embodiment of the present invention. Figure 4 As shown, the device includes: The first execution module 100 is used to collect text data, image data and structured data in the project group management scenario, and automatically identify the sub-project code and role dimension to which the data belongs through the data base to achieve multi-level data alignment and standardized integration; The second execution module 200 is used to perform semantic decoding on text data and generate binary event feature vectors through a large language model, perform target detection and behavior recognition on image data through a convolutional neural network and generate behavior category feature vectors, and perform dimensionless processing on structured data through a standardization algorithm and generate standardized feature vectors. The third execution module 300 is used to map the three types of feature vectors to a unified dimension through a fully connected layer and then concatenate them, separating static features and temporal features and inputting them into a multilayer perceptron network and a long short-term memory network for feature extraction. The fourth execution module 400 is used to generate a comprehensive score based on the static feature score and temporal feature score output by the multilayer perceptron network, and to input the score result and business data into the large language model to generate interpretable comments. The model parameters are dynamically updated through manual review of the results to achieve closed-loop optimization.

[0095] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0096] To implement the above embodiments, the present invention also proposes an electronic device, comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory to implement the method provided in the foregoing embodiments.

[0097] To implement the above embodiments, the present invention also proposes a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the methods provided in the foregoing embodiments.

[0098] To implement the above embodiments, the present invention also proposes a computer program product, including a computer program that, when executed by a processor, implements the methods provided in the foregoing embodiments.

[0099] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in this invention all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0100] It should be noted that personal information collected from users should be used for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. Furthermore, such collection / sharing should only be conducted after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign an agreement / authorization that includes authorization of relevant user information before the user uses the function. In addition, any necessary steps must be taken to protect and safeguard access to such personal information data and ensure that others with access to personal information data comply with their privacy policies and procedures.

[0101] This invention is intended to provide implementation schemes for users to selectively prevent the use or access to personal information data. That is, this disclosure is intended to provide hardware and / or software to prevent or block access to such personal information data. Once personal information data is no longer needed, risks can be minimized by restricting data collection and deleting data. Furthermore, where applicable, such personal information can be de-identified to protect user privacy.

[0102] In the foregoing descriptions of the embodiments, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0103] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0104] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of the invention pertain.

[0105] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0106] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any of the following techniques known in the art, or a combination thereof: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0107] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0108] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0109] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

[0110] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0111] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for intelligent performance evaluation of project portfolios, characterized in that, include: S1 collects text data, image data, and structured data in project group management scenarios, and automatically identifies the sub-project code and role dimension to which the data belongs through the data base, so as to achieve multi-level data alignment and standardized integration; S2 uses a large language model to perform semantic decoding on text data and generate binary event feature vectors, uses a convolutional neural network to perform target detection and behavior recognition on image data and generate behavior category feature vectors, and uses a standardization algorithm to perform dimensionless processing on structured data and generate standardized feature vectors. S3 maps the three types of feature vectors to a unified dimension through a fully connected layer and then concatenates them to separate static features and temporal features, which are then input into a multilayer perceptron and a long short-term memory network for feature extraction. S4, based on the static feature score and temporal feature score output by the multilayer perceptron, adopts a trainable weighted fusion strategy to generate a comprehensive score, and inputs the score results and business data into the large language model to generate interpretable comments. The model parameters are dynamically updated through manual review to achieve closed-loop optimization.

2. The method as described in claim 1, characterized in that, The data collection includes text, image, and structured data from the project group management scenario. It also automatically identifies the sub-project code and role dimension of the data through a data foundation, achieving multi-level data alignment and standardized integration. The data base automatically identifies the sub-project code, construction section, and role dimension to which the data belongs, and aligns the multi-source data with the project group organizational structure at multiple levels based on preset hierarchical mapping rules. Implement a data traceability mechanism for the collected multi-source data, recording the source timestamp, collection device identifier, and associated responsible person information for each piece of data to ensure data traceability.

3. The method as described in claim 1, characterized in that, The step of semantically decoding text data and generating binary event feature vectors using a large language model also includes: The semantic relevance of each text to each event template is calculated using a pre-defined template dictionary. When the relevance score exceeds the threshold of 0.6, the corresponding binary event features are concatenated with the original semantic embedding vector to form a fused feature vector.

4. The method as described in claim 1, characterized in that, The process of mapping the three types of feature vectors to a unified dimension through a fully connected layer and then concatenating them, separating static features from temporal features, and inputting them separately into a multilayer perceptron and a long short-term memory network for feature extraction, also includes: When mapping the dimensions of the three types of feature vectors through a fully connected layer, a trainable linear transformation parameter matrix is ​​used to process the text feature vector, behavioral feature vector, and structured feature vector respectively, so that the dimensions of the mapped feature vectors meet the input requirements of MLP and LSTM.

5. The method as described in claim 1, characterized in that, The static feature score and temporal feature score based on the output of the multilayer perceptron network are used to generate a comprehensive score by employing a trainable weighted fusion strategy, and the comprehensive score also includes: The fusion weight parameters are obtained using the gradient descent algorithm. and Dynamic optimization is performed, among which The weights corresponding to the static feature scores, The weights corresponding to the time-series feature scores, and This is to minimize the error between the overall score and the results of manual review.

6. The method as described in claim 1, characterized in that, Also includes: The final confirmed assessment results will be used to generate multi-dimensional visualization charts, including task completion trend charts, cross-project collaboration heatmaps, and R&D innovation radar charts. These charts and structured reports will be pushed to the project group management platform to enable multi-terminal display and real-time updates of the assessment results.

7. A project portfolio performance intelligent assessment device, characterized in that, include: The first execution module is used to collect text data, image data and structured data in the project group management scenario, and automatically identify the sub-project code and role dimension to which the data belongs through the data base to achieve multi-level data alignment and standardized integration; The second execution module is used to perform semantic decoding on text data and generate binary event feature vectors through a large language model, perform target detection and behavior recognition on image data through a convolutional neural network and generate behavior category feature vectors, and perform dimensionless processing on structured data through a standardization algorithm and generate standardized feature vectors. The third execution module is used to map the three types of feature vectors to a unified dimension through a fully connected layer and then concatenate them, separating static features and temporal features and inputting them into a multilayer perceptron and a long short-term memory network for feature extraction. The fourth execution module is used to generate a comprehensive score based on the static feature score and temporal feature score output by the multilayer perceptron network, using a trainable weighted fusion strategy. The score results and business data are then input into a large language model to generate interpretable comments. The model parameters are dynamically updated through manual review to achieve closed-loop optimization.

8. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method of any one of claims 1-6.

Images