High-risk industry multi-dimensional dynamic weight employee evaluation method

Abstract

The invention relates to the technical field of employee evaluation, in particular to a high-risk industry multi-dimensional dynamic weight employee evaluation method, which comprises the following steps of: obtaining quantitative scores of three-level indexes of employees and normalizing the quantitative scores by constructing a multi-level evaluation index system; inputting the standardized score vector into a Transform-based weight learning model, dynamically learning an index weight through a self-attention mechanism, carrying out step-by-step aggregation through a multi-layer perceptron, and outputting a comprehensive score of the employee on each first-level index; inputting the comprehensive score and the historical background information into a pre-established large language model, and performing fusion analysis according to a structured template to generate a personalized evaluation text; and finally, based on the score, the text and the employee information, automatically synthesizing a visual comprehensive evaluation report containing the radar map, the information bar and the text. According to the invention, objectiveness, individuation and operability of employee evaluation are realized.

Description

Technical Field

[0001] This invention relates to the field of employee evaluation technology, and in particular to a multi-dimensional dynamic weighted employee evaluation method for high-risk industries. Background Technology

[0002] In complex and high-risk industries such as energy, manufacturing, and transportation, employee competency assessment has long faced three core problems: a single assessment dimension, strong subjectivity, and a lack of personalized feedback. Traditional performance appraisal methods fail to comprehensively, objectively, and interpretably reflect employees' overall performance across multiple dimensions, including professional skills, emergency response, communication and collaboration, and leadership potential. This leads to inaccurate talent identification, inefficient training investment, and unclear employee development paths, ultimately impacting organizational safety operations and talent pipeline development.

[0003] Existing technical solutions for employee evaluation cannot simultaneously achieve "objective and dynamic weight learning" and "personalized and interpretable semantic analysis" within a single framework. Neither static weighting systems based on expert experience nor objective weighting models that rely on data dispersion can effectively capture the dynamic and nonlinear intrinsic relationships between complex evaluation indicators. This results in rigid weight settings that fail to truly reflect the differences in the ability structure of employees in different positions and at different development stages. At the same time, even when large language models are introduced to assist in report generation, the analysis content generated is superficial and monotonous because the input is only the isolated final score and lacks deep integration with the structured indicator system and the employee's specific historical behavioral data. It cannot provide truly personalized, understandable, and actionable improvement guidance. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a multi-dimensional dynamic weighted employee evaluation method for high-risk industries, aiming to improve the technical problems of existing technologies such as single evaluation dimensions, subjective weight setting, and lack of interpretability and personalized feedback in the results.

[0005] This invention provides the following technical solution: a multi-dimensional dynamic weighted employee evaluation method for high-risk industries, the method comprising the following steps: A multi-level evaluation indicator system is constructed, which includes primary indicators, secondary indicators and tertiary indicators. Quantitative scoring data of target employees on tertiary indicators is obtained, and the quantitative scoring data on tertiary indicators is normalized to form a standardized tertiary scoring vector. The standardized three-level scoring vector is input into a weight learning model based on a Transformer encoder. Through the self-attention mechanism in the weight learning model, the nonlinear correlation and dynamic weight between each level of indicators are automatically learned and a weighted feature representation is generated. The weighted feature representation is aggregated level by level using a multilayer perceptron to calculate and output the comprehensive score of the target employee on each first-level indicator. The comprehensive score of the target employee on each primary indicator, along with the target employee's historical background information, is input into the pre-built large language model. Based on the preset structured prompt template, the large language model is driven to perform fusion analysis on the comprehensive score and historical background information. Based on the fusion analysis results, the large language model generates personalized natural language evaluation text that includes analysis of the scoring results and actionable improvement suggestions; The system receives the personalized natural language assessment text and, based on the target employee's comprehensive score on each primary indicator, the personalized natural language assessment text, and the target employee's basic information, automatically generates and outputs a visual comprehensive assessment report containing a radar chart, a basic information column, and the assessment text.

[0006] Preferably, the construction process of the multi-level evaluation index system includes: Define primary indicators, which include hard skills dimensions for assessing employees' core business capabilities, soft skills dimensions for assessing teamwork and communication, and development potential dimensions for assessing employees' growth potential. For each of the primary indicators, multiple secondary indicators are decomposed and defined, wherein the secondary indicators are used to provide a specific description of the capability dimension represented by the primary indicator. For each of the aforementioned secondary indicators, multiple tertiary indicators are further decomposed and defined, wherein the tertiary indicators are atomic-level evaluation items that can be directly observed, measured, or scored; Based on the definition of the three-level indicators, a quantitative scoring rule is designed and determined for each of the three-level indicators to obtain the quantitative scoring data of the target employees.

[0007] Preferably, the process for forming the standardized three-level scoring vector includes: Obtain the raw quantitative score data of the target employee on all three-level indicators; For each level three indicator, the original quantitative scoring data is linearly transformed to a value in the range [0,1] according to the preset scoring dimensions and threshold of the indicator. The normalized values ​​in the [0,1] interval corresponding to all three-level indicators are arranged in a preset order of the three-level indicators and combined to form the standardized three-level scoring vector.

[0008] Preferably, the generation process of the weighted feature representation includes: Each scalar element in the standardized three-level rating vector is mapped to a high-dimensional embedding space through the linear projection layer in the weight learning model to obtain the initial embedding vector for each three-level index. The initial embedding vector of each of the three-level indicators is input into the Transformer encoder layer in the weight learning model; Within the Transformer encoder layer, the association weights between the initial embedding vectors of each tertiary metric are calculated using the self-attention mechanism. Based on these association weights, all initial embedding vectors are weighted, fused, and nonlinearly transformed to output a weighted feature representation for each tertiary metric.

[0009] Preferably, the process for calculating and outputting the comprehensive score of the target employee on each primary indicator includes: The weighted feature representations of all tertiary indicators belonging to the same secondary indicator are concatenated to form the aggregated feature vector of that secondary indicator; The aggregated feature vector of each secondary indicator is input into the first multilayer perceptron for regression calculation, and the preliminary comprehensive score of the secondary indicator is output. The preliminary comprehensive scores of all secondary indicators belonging to the same primary indicator are combined to form the score vector of that primary indicator; The scoring vector of each primary indicator is input into the second multilayer perceptron for regression calculation, and the final comprehensive score of that primary indicator is output.

[0010] Preferably, the target employee's historical background information includes at least one of the following: Historical performance evaluation records; The training programs participated in, their duration, and the assessment results; Records of the handling process and results of past accidents or emergencies; History of job transfers and changes in responsibilities; Record of roles and contributions in team collaboration projects.

[0011] Preferably, the process of driving the large language model to perform fusion analysis of the comprehensive score and historical background information includes: The target employee's comprehensive score on each primary indicator, the target employee's historical background information, and the target employee's basic information are filled and organized according to the format and fields defined in the structured prompt template to construct a structured input context. The structured input context is fed into the pre-built large language model; The large language model performs joint reasoning and analysis on the comprehensive score and historical background information in the input context based on the instructions embedded in the structured prompt template.

[0012] Preferably, the process of jointly reasoning and analyzing the comprehensive score and historical background information in the input context includes: Identify the primary indicators in the overall score that are below a preset threshold or significantly deviate from the team's average level, and mark them as dimensions to be improved; Retrieve historical behavior or event records associated with the dimension to be improved from the historical background information; Based on the scoring performance of the dimension to be improved and the associated historical behavior or event records, infer the reasons for the scoring results of that dimension; Based on the definition of the dimension to be improved, the inferred reasons, and the pre-stored industry best practice knowledge, targeted improvement suggestions are generated.

[0013] Preferably, the process of receiving personalized natural language assessment text generated by the large language model, which includes analysis of the scoring results and actionable improvement suggestions, includes: Receive the raw natural language text stream output by the large language model; The original natural language text stream is formatted to conform to the preset requirements for the layout, length, and tone style of the evaluation report text paragraphs; The formatted text is then labeled as personalized natural language assessment text corresponding to the target employee.

[0014] Preferably, the process for generating the visual comprehensive evaluation report includes: The pre-stored visualization engine is invoked to generate the radar chart based on the comprehensive score of the target employee on each primary indicator; Extract the basic information of the target employee and generate the basic information column according to the preset layout; The personalized natural language evaluation text is embedded into a preset text display area; The radar chart, the basic information bar, and the text display area containing the evaluation text are combined and rendered according to a preset report template to generate the file of the visual comprehensive evaluation report. Output the visualized comprehensive evaluation report to the specified terminal or storage path.

[0015] The present invention has the following beneficial effects: 1. In this invention, a weight learning model based on a Transformer encoder is constructed, and a self-attention mechanism is used to automatically learn the nonlinear correlation and dynamic weights between multi-level evaluation indicators. This replaces the method of relying on manual experience to set static weights, overcomes the influence of subjective bias and cognitive bias on the evaluation results, and ensures the data-driven nature and cross-scenario consistency of the scoring process and results.

[0016] 2. In this invention, a pre-built large language model is driven by a preset structured prompt template to integrate and analyze quantitative scores and employee historical background information, generating personalized natural language assessment text that includes specific cause analysis and targeted improvement suggestions. This design enables the assessment conclusion to not only include numerical scores, but also provides semantic explanations and action guidelines, solving the problem that traditional assessment methods have single output results and lack of guidance value.

[0017] 3. In this invention, by integrating a visualization engine, multidimensional scores, personalized analysis texts and basic employee information are automatically combined into a comprehensive evaluation report containing radar charts, text and information bars, and one-click output is supported. This design integrates multiple links such as indicator quantification, weight learning, intelligent analysis and report generation into an end-to-end automated process, which significantly improves the efficiency of evaluation work and provides intuitive and comprehensive information support for management decisions. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the multidimensional dynamic weighted employee evaluation method for high-risk industries proposed in this invention. Figure 2 This is a schematic diagram of the multi-dimensional personnel evaluation framework of the multi-dimensional dynamic weighted employee evaluation method for high-risk industries proposed in this invention. Figure 3 This is a schematic diagram illustrating the evaluation report effect of the multi-dimensional dynamic weighted employee evaluation method for high-risk industries proposed in this invention. Detailed Implementation

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] In a first embodiment of the present invention, the present invention provides a multi-dimensional dynamic weighted employee evaluation method for high-risk industries, such as... Figures 1-3 As shown, it includes the following steps: A multi-level evaluation indicator system is constructed, which includes primary, secondary, and tertiary indicators. Quantitative scoring data of target employees on tertiary indicators is obtained, and the quantitative scoring data on tertiary indicators is normalized to form a standardized tertiary scoring vector.

[0021] Furthermore, the process of constructing a multi-level evaluation indicator system includes: Define primary indicators, which include hard skills dimensions for assessing employees' core business capabilities, soft skills dimensions for assessing teamwork and communication, and development potential dimensions for assessing employees' growth potential. For each primary indicator, decompose and define multiple secondary indicators under it, where the secondary indicators are used to describe the capability dimensions represented by the primary indicator in detail. For each secondary indicator, it is further decomposed and defined into multiple tertiary indicators, where the tertiary indicators are atomic-level evaluation items that can be directly observed, measured or scored; Based on the definition of the three-level indicators, we designed and determined the quantitative scoring rules for each of the three-level indicators in order to obtain the quantitative scoring data of the target employees.

[0022] Furthermore, the process for forming a standardized three-level rating vector includes: Obtain the raw quantitative score data of the target employees on all three-level indicators; For each level three indicator, the original quantitative scoring data is linearly transformed to a value in the range [0,1] based on the preset scoring dimensions and thresholds of that indicator. The normalized values ​​in the [0,1] interval corresponding to all three-level indicators are arranged in the preset order of the three-level indicators and combined to form a standardized three-level scoring vector.

[0023] Specifically, a multi-level evaluation indicator system is constructed, comprising top-down primary, secondary, and tertiary indicators. Primary indicators at least cover hard skills for assessing core business competence, soft skills for assessing teamwork and communication, and development potential for assessing employee growth. For each primary indicator, it is decomposed and defined into multiple subordinate secondary indicators. These secondary indicators provide a concrete description of the competence dimension represented by the primary indicator. Then, for each secondary indicator, it is further decomposed and defined into multiple subordinate tertiary indicators. These tertiary indicators are atomic-level evaluation items that can be directly observed, measured, or scored. For example, for the secondary indicator "emergency response capability" under the "hard skills" dimension, its tertiary indicators may include "participation in emergency drills," "accident response speed," and "mastery of emergency response plans." Based on the defined tertiary indicators, quantitative scoring rules for each tertiary indicator need to be designed and determined. These scoring rules clarify the data source, collection method, and quantitative standards. For example, "participation in emergency drills" can be quantified through attendance records, and "accident response speed" can be calculated from work order system logs. Obtain the raw quantitative score data of the target employee on all three-level indicators. This process, based on the aforementioned scoring rules, collects data from relevant business systems, performance records, or manual input. The raw quantitative score data is then normalized to form a standardized three-level score vector. Specifically, for each three-level indicator, based on the preset scoring dimensions and thresholds for that indicator, the raw quantitative score data is mapped to a value in the [0,1] interval through a linear transformation. The formula for this linear transformation is: ; in, This represents the raw score data for a certain tertiary indicator, where min and max are the preset lower and upper thresholds for that indicator, respectively. For values ​​that fall within the [0,1] interval after normalization, repeat this operation for all level 3 indicators; The normalized values ​​corresponding to all tertiary indicators in the [0,1] interval are combined according to a preset order of the tertiary indicators to form a one-dimensional vector, denoted as . : ; in, This indicates the total number of third-level indicators. This vector represents the normalized score value of the i-th tertiary indicator. As input to the subsequent weight learning model; Through the above steps, a structured, multi-level evaluation index system can be constructed, and a standardized data preprocessing process can be implemented, achieving the technical effect of systematically and quantitatively representing the capabilities of complex employees.

[0024] The standardized three-level scoring vector is input into a weight learning model based on a Transformer encoder. Through the self-attention mechanism in the weight learning model, the nonlinear relationship and dynamic weight between the indicators at each level are automatically learned and weighted feature representations are generated. The weighted feature representations are aggregated level by level using a multilayer perceptron to calculate and output the comprehensive score of the target employee on each first-level indicator.

[0025] Furthermore, the generation process of the weighted feature representation includes: Each scalar element in the standardized three-level rating vector is mapped to a high-dimensional embedding space through a linear projection layer in the weight learning model to obtain the initial embedding vector for each three-level indicator. The initial embedding vector for each level 3 metric is input into the Transformer encoder layer in the weight learning model; Within the Transformer encoder layer, the association weights between the initial embedding vectors of each tertiary metric are calculated through a self-attention mechanism. Based on the association weights, all initial embedding vectors are weighted, fused, and nonlinearly transformed to output the weighted feature representation of each tertiary metric.

[0026] Furthermore, the process for calculating and outputting the target employee's comprehensive score on each primary indicator includes: The weighted feature representations of all tertiary indicators belonging to the same secondary indicator are concatenated to form the aggregated feature vector of that secondary indicator. The aggregated feature vector of each secondary indicator is input into the first multilayer perceptron for regression calculation, and the preliminary comprehensive score of the secondary indicator is output. The preliminary comprehensive scores of all secondary indicators belonging to the same primary indicator are combined to form the score vector of that primary indicator. The scoring vector of each primary indicator is input into the second multilayer perceptron for regression calculation, and the final comprehensive score of that primary indicator is output.

[0027] Specifically, the standardized three-level rating vector The input is fed into the weighted learning model. First, each scalar element in the vector is mapped to a high-dimensional embedding space through a linear projection layer in the model. This operation is performed by a learnable projection matrix. The implementation dimension is determined according to the model design. For a vector containing k tertiary indicators, the initial embedding vector of each tertiary indicator is obtained after mapping. These vectors together constitute the sequence representation of the model input. The initial embedding vector sequence for each tertiary indicator is input into the Transformer encoder layer of the weight learning model. Inside the encoder, the self-attention mechanism automatically calculates the association weight between any two initial embedding vectors in the sequence according to a formula. This association weight reflects the semantic and contribution relationship between the corresponding tertiary indicators. Subsequently, the encoder performs a weighted summation and nonlinear transformation on all initial embedding vectors based on the calculated association weights, specifically implemented through a feedforward neural network. This process outputs a new vector representation of each tertiary indicator enhanced with contextual information, denoted as the weighted feature representation. This step can be represented by the following formula: ; in, This represents the computational process involving self-attention and feedforward networks. The dimension is consistent with the dimension of the model's hidden layer; For any secondary indicator, the weighted characteristic of all its subordinate tertiary indicators is represented as follows. A concatenation operation is performed to form the aggregated feature vector of the secondary indicator. Then, the aggregated feature vector of each secondary indicator is input into a lightweight first-level multilayer perceptron for regression calculation. This multilayer perceptron typically contains a hidden layer with a ReLU activation function and a linear output layer. Its output is the preliminary comprehensive score of the secondary indicator, denoted as... For any primary indicator, calculate the preliminary comprehensive score of all its subordinate secondary indicators. Combine them to form the scoring vector for this primary indicator. Finally, the scoring vector for each primary indicator is... The inputs are fed into a lightweight second-level multilayer perceptron for regression calculation. The structure of this second-level multilayer perceptron is similar to that of the first-level multilayer perceptron, and its output is the final comprehensive score for that primary indicator, denoted as . This process can be characterized by the following formula: ; ; in, This indicates a splicing operation. (.)and (.) represent the regression calculations performed by the first and second multilayer perceptrons, respectively. Owned by subordinates The combined vectors ultimately output by the system provide a comprehensive score for the target employee across all preset primary indicator dimensions. Through the above steps, by introducing a weight learning model based on a Transformer encoder and a hierarchical aggregation mechanism, the technical effect of data-driven, dynamic weighting, and precise quantification of employees' multidimensional capabilities is achieved.

[0028] The comprehensive score of the target employee on each primary indicator, along with the employee's historical background information, is input into the pre-built large language model. Based on the preset structured prompt template, the large language model is driven to perform integrated analysis of the comprehensive score and historical background information.

[0029] Furthermore, the target employee's historical background information includes at least one of the following: Historical performance evaluation records; The training programs participated in, their duration, and the assessment results; Records of the handling process and results of past accidents or emergencies; History of job transfers and changes in responsibilities; Record of roles and contributions in team collaboration projects.

[0030] Furthermore, the process of integrating and analyzing comprehensive scores with historical context information using the large language model includes: The target employee's comprehensive score on each primary indicator, the target employee's historical background information, and the target employee's basic information are filled and organized according to the format and fields defined in the structured prompt template to form a structured input context. Input the structured input context into a pre-built large language model; The large language model uses instructions embedded in structured prompt templates to perform joint reasoning and analysis on the comprehensive score and historical background information in the input context.

[0031] Furthermore, the process of jointly reasoning and analyzing the comprehensive score and historical background information in the input context includes: Identify primary indicators in the overall score that are below a preset threshold or significantly deviate from the team average level, and mark them as dimensions to be improved; In the historical background information, retrieve historical behavior or event records that are associated with the dimension to be improved; Based on the scoring performance of the dimension to be improved and the associated historical behavior or event records, infer the reasons for the scoring results of that dimension; By combining the definition of the dimension to be improved, the inferred reasons, and the pre-existing industry best practice knowledge, targeted improvement suggestions are generated.

[0032] Specifically, based on a preset structured prompt template, the system fills in the corresponding fields defined in the template with the target employee's comprehensive score on each primary indicator, the employee's historical background information, and basic information. The historical background information can be selected from one or more of the following: historical performance evaluation records, training records, accident handling logs, job change history, or team contribution records. Basic information includes the employee's name, employee ID, position, and department. After filling in the information, the system organizes this information into a coherent, structured natural language text according to the paragraph order and grammatical structure defined in the template, forming the final input context. For example, the template might include: "Employee [Name] has served in [Position] for [Number of Years] years. The scores for each dimension in this evaluation are as follows: [List the comprehensive score]. The employee's historical performance includes: [List relevant historical background information]. Please analyze based on the above information."

[0033] The system inputs the structured input context, as described above, into a pre-trained large language model (e.g., the GPT series models) via an application programming interface. The model receives this context as input prompts and, based on pre-defined instructions in the prompt template guiding its analysis and generation, performs deep joint reasoning on the comprehensive score and historical background information within the input context. This reasoning process includes the following sub-steps: First, the model identifies primary indicators in the comprehensive score that are below a preset threshold or significantly deviate from the historical average score of the employee's team, marking these indicators as "dimensions to be improved." Second, the model retrieves specific historical behavior or event records associated with these "dimensions to be improved" from the provided historical background information. For example, if... If the score for the "emergency response capability" dimension is low, the model retrieves historical accident handling records or related training and assessment results. Then, based on the low score in this dimension and the specific historical records retrieved, the model makes causal inferences to analyze the possible reasons for the poor score. For example, the inferred reasons may be "responding beyond the time limit twice in the past three emergency drills" or "failure to complete the emergency response procedure training for this year". Finally, the model combines the definition of the "dimensional to be improved" itself, the inferred specific reasons, and the industry best practice knowledge base pre-stored in the system to generate one or more specific and actionable improvement suggestions. For example, based on the above inference, the model generates the suggestion "It is recommended to participate in the emergency response refresher training in the next quarter and complete at least two simulation drills within the quarter to improve response speed". Through the above steps, the structured comprehensive score and multi-source historical background information can be integrated through a preset template and input into a large language model for deep reasoning, realizing a fundamental transformation of the evaluation results from simple numerical scores to personalized, semantic, and executable analysis reports.

[0034] Based on the fusion analysis results, the large language model generates personalized natural language evaluation text that includes analysis of the scoring results and actionable improvement suggestions.

[0035] Furthermore, based on the fusion analysis results, the process by which the large language model generates personalized natural language evaluation text containing analysis of the scoring results and actionable improvement suggestions includes: Receive the raw natural language text stream output by the large language model; The raw natural language text stream is formatted to conform to the preset requirements for the layout, length, and tone of the evaluation report text paragraphs; The formatted text is then labeled as personalized natural language assessment text corresponding to the target employee.

[0036] Specifically, the system connects to the application programming interface of the large language model service and receives the raw natural language text stream generated by the model. This text stream is the direct output of the model after reasoning based on the input context, and usually contains an analysis paragraph of the scoring results and several improvement suggestions. However, its format, length, and wording may not yet meet the standards of the final report. The system calls the built-in text processing module to automatically format the raw text stream according to the preset evaluation report text specifications. This process includes: First, according to the preset layout requirements, the text is segmented, headings are added, and fonts and indentation are set. For example, "analysis" and "suggestions" are separated into independent paragraphs and bullet points are added. Second, the total length of the text is checked. If it exceeds the preset maximum character limit, it is reduced to the specified range by extracting a summary or deleting redundant expressions. Finally, the wording is adjusted according to the preset tone and style requirements. For example, if the style is set to "professional and formal," colloquial expressions are converted into written language; if the style is set to "motivational and guiding," encouraging language is added. This step is implemented through a rule engine or a lightweight style conversion model. The formatted text is associated with the unique identifier of the target employee and marked in the system as "personalized natural language assessment text corresponding to the target employee [employee name]". Then, the text is stored in a specified field of the database or generated as a separate text file for later use when generating visualization reports. Through the above steps, by receiving the raw text output by the large language model and performing automated formatting post-processing, the technical effect of transforming the preliminary analysis content generated by artificial intelligence into standardized text output that can be directly used in the final report is achieved.

[0037] The system receives the personalized natural language assessment text and, based on the target employee's comprehensive score on each primary indicator, the personalized natural language assessment text, and the target employee's basic information, automatically generates and outputs a visual comprehensive assessment report containing a radar chart, a basic information column, and the assessment text.

[0038] Furthermore, the process for generating a comprehensive visual assessment report includes: The pre-stored visualization engine is invoked to generate a radar chart based on the target employee's comprehensive score on each primary indicator; Extract the basic information of the target employees and generate a basic information column according to the preset layout; Embed personalized natural language assessment text into a preset text display area; The radar chart, basic information bar, and text display area containing the evaluation text are combined and rendered according to the preset report template to generate a visual comprehensive evaluation report file. Output the visual comprehensive evaluation report to the specified terminal or storage path.

[0039] Specifically, the system first calls a pre-stored visualization engine, such as a chart library based on ECharts, Matplotlib, or D3.js, and uses the target employee's comprehensive score on each primary indicator as data input. The visualization engine automatically draws a radar chart based on a preset chart style, such as a pentagonal radar chart. Each axis of this radar chart represents a primary indicator, and its length is proportional to the indicator's score, thus visually displaying the employee's ability profile and relative strengths across various dimensions. Simultaneously, the system extracts the target employee's basic information from the database, including name, employee ID, position, department, date of employment, and photo. Following a preset layout template, this information is filled into the corresponding fields to generate a uniformly formatted and complete employee basic information column. Furthermore, the system fully embeds the tagged, personalized natural language evaluation text corresponding to the target employee into the text display area of ​​the preset evaluation report template. This area is typically located at the bottom of the report, and its style is predefined in the template.

[0040] The system loads the generated radar chart image file, the HTML or image fragment of the basic information bar, and the text area content embedded with the evaluation text as three independent components into the preset evaluation report template. The template defines the overall layout of the report, such as placing the radar chart in the upper left area, the basic information bar in the upper right area, and the text in the bottom area. The system positions the three components in the specified areas according to the layout instructions of the template and performs composite rendering. This rendering process generates a single document file with complete format containing all visualization and text elements. The file format can be set to PDF, HTML, or image format as needed. The system will output the final visualized comprehensive evaluation report file in at least one of the following ways: first, send it to the terminal device of the designated manager for preview or printing; second, store it on the company's designated file server or cloud storage path and generate an accessible link, or directly link it to the employee's file in the human resources management system to complete the archiving of this evaluation. Through the above steps, by integrating a visualization engine with preset report templates, multidimensional scoring data, semantic analysis text, and basic employee information are automatically synthesized into a structured, visualized comprehensive report, realizing the integrated output of evaluation results from scattered data to intuitive, deliverable decision documents.

[0041] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-dimensional dynamic weighted employee evaluation method for high-risk industries, characterized in that: The method includes the following steps: A multi-level evaluation indicator system is constructed, which includes primary indicators, secondary indicators and tertiary indicators. Quantitative scoring data of target employees on tertiary indicators is obtained, and the quantitative scoring data on tertiary indicators is normalized to form a standardized tertiary scoring vector. The standardized three-level scoring vector is input into a weight learning model based on a Transformer encoder. Through the self-attention mechanism in the weight learning model, the nonlinear correlation and dynamic weight between each level of indicators are automatically learned and a weighted feature representation is generated. The weighted feature representation is aggregated level by level using a multilayer perceptron to calculate and output the comprehensive score of the target employee on each first-level indicator. The comprehensive score of the target employee on each primary indicator, along with the target employee's historical background information, is input into the pre-built large language model. Based on the preset structured prompt template, the large language model is driven to perform fusion analysis on the comprehensive score and historical background information. Based on the fusion analysis results, the large language model generates personalized natural language evaluation text that includes analysis of the scoring results and actionable improvement suggestions; The system receives the personalized natural language assessment text and, based on the target employee's comprehensive score on each primary indicator, the personalized natural language assessment text, and the target employee's basic information, automatically generates and outputs a visual comprehensive assessment report containing a radar chart, a basic information column, and the assessment text.

2. The multi-dimensional dynamic weighted employee evaluation method for high-risk industries according to claim 1, characterized in that, The construction process of the multi-level evaluation index system includes: Define primary indicators, which include hard skills dimensions for assessing employees' core business capabilities, soft skills dimensions for assessing teamwork and communication, and development potential dimensions for assessing employees' growth potential. For each of the primary indicators, multiple secondary indicators are decomposed and defined, wherein the secondary indicators are used to provide a specific description of the capability dimension represented by the primary indicator. For each of the aforementioned secondary indicators, multiple tertiary indicators are further decomposed and defined, wherein the tertiary indicators are atomic-level evaluation items that can be directly observed, measured, or scored; Based on the definition of the three-level indicators, a quantitative scoring rule is designed and determined for each of the three-level indicators to obtain the quantitative scoring data of the target employees.

3. The multi-dimensional dynamic weighted employee evaluation method for high-risk industries according to claim 1, characterized in that, The process for forming the standardized three-level scoring vector includes: Obtain the raw quantitative score data of the target employee on all three-level indicators; For each level three indicator, the original quantitative scoring data is linearly transformed to a value in the range [0,1] according to the preset scoring dimensions and threshold of the indicator. The normalized values ​​in the [0,1] interval corresponding to all three-level indicators are arranged in a preset order of the three-level indicators and combined to form the standardized three-level scoring vector.

4. The multi-dimensional dynamic weighted employee evaluation method for high-risk industries according to claim 1, characterized in that, The process for generating the weighted feature representation includes: Each scalar element in the standardized three-level scoring vector is mapped to a high-dimensional embedding space through the linear projection layer in the weight learning model to obtain the initial embedding vector for each three-level index. The initial embedding vector of each of the three-level indicators is input into the Transformer encoder layer in the weight learning model; Within the Transformer encoder layer, the association weights between the initial embedding vectors of each tertiary metric are calculated using the self-attention mechanism. Based on these association weights, all initial embedding vectors are weighted, fused, and nonlinearly transformed to output a weighted feature representation for each tertiary metric.

5. The multi-dimensional dynamic weighted employee evaluation method for high-risk industries according to claim 1, characterized in that, The process for calculating and outputting the comprehensive score of the target employee on each primary indicator includes: The weighted feature representations of all tertiary indicators belonging to the same secondary indicator are concatenated to form the aggregated feature vector of that secondary indicator; The aggregated feature vector of each secondary indicator is input into the first multilayer perceptron for regression calculation, and the preliminary comprehensive score of the secondary indicator is output. The preliminary comprehensive scores of all secondary indicators belonging to the same primary indicator are combined to form the score vector of that primary indicator; The scoring vector of each primary indicator is input into the second multilayer perceptron for regression calculation, and the final comprehensive score of that primary indicator is output.

6. The multi-dimensional dynamic weighted employee evaluation method for high-risk industries according to claim 1, characterized in that, The target employee's historical background information includes at least one of the following: Historical performance evaluation records; The training programs participated in, their duration, and the assessment results; Records of the handling process and results of past accidents or emergencies; History of job transfers and changes in responsibilities; Record of roles and contributions in team collaboration projects.

7. The multi-dimensional dynamic weighted employee evaluation method for high-risk industries according to claim 1, characterized in that, The process of driving the large language model to fuse and analyze the comprehensive score and historical background information includes: The target employee's comprehensive score on each primary indicator, the target employee's historical background information, and the target employee's basic information are filled and organized according to the format and fields defined in the structured prompt template to construct a structured input context. The structured input context is fed into the pre-built large language model; The large language model performs joint reasoning and analysis on the comprehensive score and historical background information in the input context based on the instructions embedded in the structured prompt template.

8. The multi-dimensional dynamic weighted employee evaluation method for high-risk industries according to claim 7, characterized in that, The process of jointly reasoning and analyzing the comprehensive score and historical background information in the input context includes: Identify the primary indicators in the overall score that are below a preset threshold or significantly deviate from the team's average level, and mark them as dimensions to be improved; Retrieve historical behavior or event records associated with the dimension to be improved from the historical background information; Based on the scoring performance of the dimension to be improved and the associated historical behavior or event records, infer the reasons for the scoring results of that dimension; Based on the definition of the dimension to be improved, the inferred reasons, and the pre-stored industry best practice knowledge, targeted improvement suggestions are generated.

9. The multi-dimensional dynamic weighted employee evaluation method for high-risk industries according to claim 1, characterized in that, Based on the fusion analysis results, the process by which the large language model generates personalized natural language evaluation text that includes analysis of the scoring results and actionable improvement suggestions includes: Receive the raw natural language text stream output by the large language model; The original natural language text stream is formatted to conform to the preset requirements for the layout, length, and tone style of the evaluation report text paragraphs; The formatted text is then labeled as personalized natural language assessment text corresponding to the target employee.

10. The multi-dimensional dynamic weighted employee evaluation method for high-risk industries according to claim 1, characterized in that, The process for generating the comprehensive visualization assessment report includes: The pre-stored visualization engine is invoked to generate the radar chart based on the comprehensive score of the target employee on each primary indicator; Extract the basic information of the target employee and generate the basic information column according to the preset layout; The personalized natural language evaluation text is embedded into a preset text display area; The radar chart, the basic information bar, and the text display area containing the evaluation text are combined and rendered according to a preset report template to generate the file of the visual comprehensive evaluation report. Output the visualized comprehensive evaluation report to the specified terminal or storage path.

Images