Dynamic evaluation and feedback system for employee training effect

Abstract

The invention relates to the crossing field of enterprise human resource management and information technology, and specifically discloses a dynamic evaluation and feedback system for employee training effects. The system comprises a behavior sequence modeling module, a dynamic effect evaluation module and an intelligent feedback decision module. The behavior sequence modeling module is used for obtaining the working behavior data after employee training from a business system and generating a quantitative state vector. And the dynamic effect evaluation module constructs a behavior mode evolution trajectory according to the continuous state vectors, and dynamically evaluates the training effect by analyzing the geometric characteristics of the trajectory. And the intelligent feedback decision module generates a personalized intervention instruction through a feedback strategy utility network according to the evaluation result. All the modules are sequentially connected to form a closed loop, and the decision is continuously optimized by utilizing subsequent working behavior data. According to the invention, the continuous evaluation and accurate feedback of the training effect in the conversion process in the actual work are realized.

Description

Technical Field

[0001] This invention relates to the intersection of enterprise human resource management and information technology, and more specifically, to a dynamic evaluation and feedback system for employee training effectiveness. Background Technology

[0002] Employee training is a crucial activity for enterprises to enhance human capital and organizational effectiveness. Accurately evaluating training effectiveness and providing effective feedback based on those evaluations are key management steps to ensure that training investment yields the expected return and achieves continuous employee skill development. Traditionally, the measurement of training effectiveness usually stops at immediate assessments at the end of the training. However, the value of training lies in whether its content can be effectively applied by employees in real-world work scenarios and translated into sustainable performance improvement. Therefore, developing a system that can track and intervene in the transformation process of training effectiveness on the workplace is of great significance.

[0003] Currently, common training effectiveness evaluation techniques suffer from several limitations. Regarding evaluation data, most methods rely on post-training questionnaires, knowledge tests, or operational assessments in simulated environments. The data obtained through these methods is isolated from employees' real and complex work situations, exhibiting lag and context-detachment, failing to accurately reflect the transfer and application of knowledge and skills in dynamic and ever-changing real-world tasks. In terms of evaluation dimensions, existing technologies focus primarily on discrete knowledge point mastery or single task completion results, struggling to deeply analyze systemic changes in employee work behavior patterns, such as the standardization of operational procedures, optimization paths for problem-solving strategies, and the evolution of collaborative habits—deeper processes of skill internalization. Regarding evaluation feedback linkage, most systems lack automated feedback mechanisms deeply integrated with evaluation results. Evaluation conclusions are typically presented in static report form, with a long cycle for translating them into concrete improvement measures, relying on manual interpretation, and making it difficult to provide differentiated and timely intervention guidance based on individual employee weaknesses. This leads to a disconnect between the "evaluation" and "development" stages, failing to form a closed-loop management system.

[0004] In summary, existing technologies have shortcomings in achieving continuous dynamic evaluation of training effectiveness based on real-world work scenarios, as well as in establishing personalized feedback and intervention mechanisms that intelligently link evaluation results. Summary of the Invention

[0005] In order to overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide a dynamic evaluation and feedback system for employee training effectiveness to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a dynamic evaluation and feedback system for employee training effectiveness, comprising: The behavior sequence modeling module is used to obtain operation logs and task result data generated by employees in actual work after training from the enterprise business system, and standardize the data into a behavioral event sequence composed of "action-object-result" triples arranged in chronological order. The module uses a preset fixed period as a time window, analyzes the behavioral event sequence within each time window, extracts statistical features reflecting behavioral habits and topological features reflecting work logic structure, and fuses and encodes the two types of features to generate a multi-dimensional quantitative state vector. This vector represents a digital summary of the employee's work mode within the time window. The dynamic effect evaluation module, connected to the behavior sequence modeling module, receives quantized state vectors from multiple consecutive time windows and connects these vectors in chronological order to construct a behavior pattern evolution trajectory in the feature space. This module performs geometric characteristic analysis on the trajectory, measuring the convergence of the behavior pattern by calculating the dispersion decay rate of the trajectory point set, measuring the trend of evolution direction by performing multi-stage shape matching between the employee trajectory and a predefined ideal skill evolution path, and measuring the smoothness of the evolution process by analyzing the stability of trajectory direction changes. Based on the comprehensive analysis of convergence, trend, and smoothness, the module dynamically judges the transformation status and stability of the training effect, and generates a feedback trigger signal containing a specific problem description when an adverse evolution trend is identified. The intelligent feedback decision-making module, connected to the dynamic effect evaluation module, is used to respond to the feedback trigger signal. Internally, this module maintains a feedback strategy utility network organized in a graph structure, where nodes represent historically accumulated behavioral problem patterns and validated intervention strategies. The weights of the connecting edges record historical utility estimates for solving specific problems using specific strategies. The module's decision-making process is as follows: based on the problem description in the trigger signal, it matches problem nodes in the network and selects the intervention strategy node with the highest weight connected to it, thereby generating a clearly defined and actionable personalized intervention instruction. The behavior sequence modeling module, dynamic effect evaluation module, and intelligent feedback decision-making module are connected in sequence to form a continuously operating closed loop. The employee's response to the intervention instructions will be captured by the behavior data collected in the next cycle and used to update the utility network, thereby realizing the adaptive optimization of the system evaluation and feedback strategy.

[0007] Furthermore, the specific operations of the behavior sequence modeling module for feature extraction and fusion of behavior event sequences within each time window include: First, performing statistical feature extraction by traversing the event sequence to calculate the frequency of predefined key operation types, the average time of task completion, and the count of errors or abnormal events, forming a statistical feature vector; simultaneously, performing structural feature extraction by abstracting behavior event types into graph nodes, constructing a directed weighted graph by analyzing the sequential transition relationships between events in the sequence, and extracting topological indicators reflecting the workflow structure from this graph, including the size of the largest connected subgraph, the density of community aggregation in the graph, and the length of the critical path, forming a topological feature vector; finally, concatenating the standardized statistical feature vector with the topological feature vector or fusing them through linear transformation to generate the final quantized state vector for that window.

[0008] Furthermore, the calculation of the directional index in the dynamic effect evaluation module is specifically achieved through multi-stage trajectory comparison: the system pre-stores one or more ideal skill evolution paths from the initial state to the target state, and each path is divided into several continuous stages representing different proficiency levels; at the same time, the actual behavioral pattern evolution trajectory of employees is adaptively segmented based on the significant change points between their state vectors; subsequently, a sequence alignment algorithm is used to find the optimal correspondence between the employee trajectory segments and the ideal path stages, allowing for speed differences; finally, by calculating the average similarity of all corresponding stages to the internal state vector set, and considering the stage importance for weighting, a directional index reflecting the correctness of the employee's skill evolution direction is comprehensively obtained.

[0009] Furthermore, the construction and query mechanism of the feedback strategy utility network is as follows: the behavioral problem pattern nodes in the network are defined by the featured problem description vectors output by the dynamic evaluation module, and the intervention strategy nodes are defined by the metadata of specific feedback measures; when historical data shows that a certain strategy is effective in solving a certain type of problem, a connection is established between the corresponding nodes and the edge weights are initialized; at each decision, the system calculates the similarity between the current problem feature vector and the problem nodes in the network to activate the most relevant node, and queries all its outgoing edges, selecting the strategy node pointed to by the edge with the highest weight as the output; the network updates the edge weights by continuously recording the effect feedback after each decision, so that the network's decision-making ability evolves over time.

[0010] Furthermore, the edge weights in the feedback strategy utility network adopt an update rule with experience generalization ability: after an intervention strategy is executed, the system tracks the improvement of core behavioral indicators related to the initial problem within a subsequent preset observation period, and normalizes the improvement to the utility gain value of this intervention; first, the weights of the edges corresponding to the strategy used are directly updated using this gain value, and a smooth update method is adopted to integrate new and old experiences; second, in order to accelerate the learning process, a portion of the gain value is distributed according to the decay coefficient to the edges of the same strategy corresponding to other problem nodes adjacent to the current problem node in the feature space, so that a successful intervention experience can simultaneously improve the system's ability to solve similar problems.

[0011] Furthermore, it includes an environmental context awareness module, which operates independently of the employee behavior analysis workflow. Its function is to monitor and record the timeline of generalized external environmental states that may affect work performance, including system performance indicators and organizational policy events. When the dynamic effect evaluation module detects abnormal fluctuations in employee behavior trajectories, the environmental context awareness module uses time correlation analysis to determine whether the fluctuations occur synchronously with environmental state events. If a high correlation is determined, an explanatory signal is generated and sent to the dynamic effect evaluation module. Based on this signal, the dynamic effect evaluation module can temporarily adjust the sensitivity of the evaluation algorithm or mark abnormal fluctuations, thereby distinguishing between external environmental interference and internal skill changes and improving the robustness of the evaluation conclusions.

[0012] Furthermore, the skill internalization assessment unit integrated into the behavior sequence modeling module evaluates the depth of training effectiveness through cross-task scenario behavior pattern analysis: This unit collects behavioral graph models generated by the same employee in multiple different but related work tasks; by comparing the core topology of different graph models, it calculates a skill stability index, which reflects the repeatability of core operating patterns in different situations; it calculates a skill reuse rate index, which reflects the degree to which key operating units become habitual response patterns of employees; and it assesses a skill adaptability index, which analyzes whether the adjustments made by employees to their behavioral structure when facing task variations conform to the logic of efficient problem-solving; by combining these three dimensions, it generates a depth assessment score that goes beyond the performance of a single task and reflects whether skills are internalized and transferable.

[0013] Furthermore, when generating the final instruction, the intelligent feedback decision-making module achieves personalization and contextualization through a feedback content assembly engine: based on the selected intervention strategy objective, the engine retrieves relevant explanatory materials, positive and negative cases, and micro-task templates from a structured knowledge atom library with semantic tags; simultaneously, the engine accesses the employee's historical interaction records to obtain their preference data for different types of content (such as videos and text), as well as their completion status for exercises of different difficulty levels; finally, combined with the specific work task context that the employee is currently handling, the selected knowledge elements and contextual parameters are dynamically assembled to generate a highly customized feedback document or task list that can directly guide the next step of action.

[0014] Furthermore, the system supports the evaluation of composite training encompassing multiple skill dimensions. This is achieved by independently configuring an ideal skill evolution path and corresponding feature extraction rules for each key skill dimension. During directional assessment, the system automatically selects the ideal path for the dominant dimension based on the employee's current primary task type for comparison, or uses parallel computing to analyze the degree of conformity between the employee's behavioral trajectory projection on the feature subspace of each dimension and the corresponding ideal path. Then, based on task requirements or development goals, the directional scores of each dimension are weighted and integrated to output an evaluation report that reflects both overall progress and reveals the balanced development of each skill dimension.

[0015] Furthermore, the system includes a group evolution map analysis module, which extends the system from individual assessment to team analysis. This module aggregates the quantified state vectors of all team members at the same time period, maps them to a low-dimensional space using dimensionality reduction techniques, and forms a visual map of the team's skill state distribution. It quantifies the clustering density of team skills by calculating the dispersion of member state points relative to the team's center point; it quantifies the convergence of team development by calculating the consistency of recent behavioral evolution direction vectors among members; and it further identifies subgroup structures in the map using clustering algorithms, revealing potential skill differentiation patterns within the team. The output of this module is used to evaluate the overall training effectiveness of the team, identify collaboration weaknesses, and plan team-level development interventions.

[0016] The technical effects and advantages of this invention are as follows: Compared to existing technologies that rely on isolated test data after training, this invention constructs a behavioral sequence modeling module to continuously collect operational logs generated by employees in actual work after training directly from the enterprise's business system, and transforms them into time-series behavioral event sequences and quantified state vectors. This system does not depend on simulated data from the training session, but rather automates the processing and analysis of employees' native behaviors in their daily work, ensuring that the assessment is based on real work output. This approach integrates the assessment process with the workflow, reflecting the application status of skills in actual tasks in real time, providing a high-fidelity and continuously updated data source for effectiveness evaluation, and solving the problems of lagging and unrealistic assessment data.

[0017] Compared to existing technologies that focus solely on static outcome indicators, this invention utilizes a dynamic effect evaluation module to construct a behavioral pattern evolution trajectory from a continuous-cycle quantified state vector, and then performs geometric analysis on its convergence, trend, and smoothness. This method does not score individual behavioral outcomes, but rather quantifies the stability, directional correctness, and consistency of employee behavioral patterns as they evolve towards a pre-defined ideal state. It reveals the process of employees moving from "knowing" to "being proficient" and then to "stability" from a dynamic evolutionary perspective, deeply evaluating "how it evolves" and "what the trend is," thereby achieving a measurement of the depth of absorption, consolidation, and transfer of training effects, overcoming the problems of traditional evaluation dimensions being singular and static.

[0018] Compared to existing technologies where assessment and feedback are often disconnected, this invention connects dynamic assessment results with historical intervention experience through a feedback strategy utility network built into the intelligent feedback decision-making module. When the assessment identifies adverse trends, the module automatically searches and makes decisions within the network that records the historical utility of various strategies, based on the specific behavioral problem pattern, generating personalized intervention instructions. The assessment conclusions are automatically transformed into actionable improvement plans, and the strategy utility data in the network is continuously optimized by tracking changes in behavioral patterns after intervention, forming a closed loop that makes feedback intervention more targeted and sustainable. Attached Figure Description

[0019] Figure 1 This is the overall flowchart of the employee training effectiveness dynamic evaluation and feedback system of the present invention.

[0020] Figure 2 This is a flowchart of the behavioral sequence feature extraction and quantization state generation process of the present invention.

[0021] Figure 3 This is a diagram illustrating the application of the intelligent feedback decision-making and strategy utility network of the present invention. Detailed Implementation

[0022] The technical solutions of 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.

[0023] Example 1 As attached Figures 1 to 3 The present invention is a dynamic evaluation and feedback system for employee training effectiveness. The technical solution of this system will be described in detail below with reference to the specific implementation process.

[0024] The system comprises multiple software functional modules running on a computing device. The main software functional modules include a behavior sequence modeling module, a dynamic effect evaluation module, and an intelligent feedback decision-making module.

[0025] Before the system starts running, initialization configuration is required. The initialization configuration includes: setting an evaluation cycle. , This is a time duration value, such as 7 days; it establishes interface parameters for data interaction with external business systems; it loads a predefined data file, which includes coordinate data of the ideal skill evolution path and initial weight values ​​of each connection edge in the feedback strategy utility network. After initialization, the system enters the working state.

[0026] The system's workflow is a cyclical process, with each cycle executing the following steps sequentially.

[0027] Step 1: Behavioral Data Collection and Standardization Processing The behavior sequence modeling module periodically reads operation log data generated by target employees after training from a specified business system through a configured data interface. Each operation log is converted into a structured behavior event. A behavior event contains the following fields: action type, operation object, execution result, and timestamp. For example, an operation log is mapped as: {Action: "Submit", Object: "Approval Application", Result: "Approved", Timestamp: "2023-10-27 10:05:22"}. All generated behavior events are arranged in chronological order according to their timestamps, forming a time-sorted behavior event sequence.

[0028] Step 2: Time Window Segmentation and Behavioral Pattern Feature Extraction The system uses a fixed duration As a window, the continuous sequence of behavioral events is divided into multiple time periods for processing. For each event subsequence within a time window, feature extraction is performed to generate a quantized state vector representing the working mode of that time period. .vector The generation involves the following two parallel computation processes.

[0029] Step 1: Statistical Feature Extraction. Iterate through all behavioral events within the current time window and calculate a set of predefined statistics. These statistics include: the frequency of occurrence of specific operation event types. The average time (AvgDuration) to complete a single task. ErrorCount represents the number of abnormal or warning events recorded by the system during the operation. These statistics are combined to form a statistical feature vector. .

[0030] Step 2: Structural Feature Extraction. This step involves constructing and analyzing a graph model. The types of behavioral events that occur are used as candidate nodes for the graph model.

[0031] In a specific implementation, to improve model efficiency, event types that occur more frequently than a preset threshold can be selected as nodes to participate in graph construction.

[0032] Analyze the sequential relationship between event types in the event subsequence. If the event types After it occurs, the event type If it frequently occurs in a short period of time, then it is at the node. and nodes Establish a line between them point to The directed edges. The weight of the directed edges. Set as an event pair The number of times they occur in this order. Based on the events throughout the entire time window, a directed weighted graph is constructed. .

[0033] Subsequently, from the directed weighted graph Extract its topological features, including: the number of nodes contained in the largest connected subgraph. ; The average clustering coefficient of nodes in the graph; the hop count of the critical path obtained through graph algorithms (e.g., calculating the longest path after identifying critical nodes using the PageRank algorithm). These topological eigenvalues ​​are combined to form a topological eigenvector. .

[0034] Step 3: Quantization of state vector generation statistical feature vectors and topological eigenvectors Standardization was performed separately to eliminate differences in magnitude between different feature values.

[0035] The two standardized vectors are concatenated to form a composite feature vector. This composite feature vector can be directly used as the quantization state vector for the current time window. Output.

[0036] In one implementation, the synthesized feature vector can be generated by an encoder function. The encoder function, which can be a linear transformation or a shallow neural network, outputs a quantized state vector with adjusted dimensions. .

[0037] Quantization state vector It is a digital representation of the employee's work behavior patterns during that time period.

[0038] Step 4: Constructing the evolution trajectory of behavioral patterns and calculating dynamic indicators The dynamic effect evaluation module receives continuous The quantized state vector generated within each time window is denoted as... ,in This is a pre-defined positive integer. Connecting these state vectors in chronological order forms a trajectory in the high-dimensional feature space, called the behavior pattern evolution path. Geometric analysis is performed on this path, and three dynamic evaluation metrics are calculated: 1. Convergence index Cv: Calculated by the nearest convergence index at the end of the path. The degree of dispersion of the distribution of state points. This degree of dispersion is determined by calculating the trace of the covariance matrix of these points. To measure, among which Indicated by the most recent The covariance matrix is ​​calculated from the state vectors.

[0039] Convergence Cv can be defined as the value calculated within the current time window. Calculated with the previous time window The attenuation ratio, for example .

[0040] when Less than a minimum value (For example When Cv is 1, it can be directly determined that Cv has converged sufficiently.

[0041] 2. Trend Indicator Tr: This indicator links employees' actual skill development paths to a predefined ideal skill evolution path. A comparison was made. The ideal skill evolution path was described. It is pre-divided into several consecutive stages.

[0042] First, the actual path is adaptively segmented based on the magnitude of changes between its state vectors. This adaptive segmentation is achieved by detecting abrupt changes in the state vector sequence, for example, by calculating the Euclidean distance between adjacent state vectors. ,when Distance greater than the historical average times ( When the preset parameter is 1.5, then... The location serves as a segmentation point.

[0043] The optimal correspondence between the stages of two paths is found by using a sequence alignment method, which can be a dynamic time warping algorithm.

[0044] Finally, for each optimal corresponding stage, the average cosine similarity of all state vectors within the two stages is calculated, and the average similarity of all stage pairs is weighted and averaged to obtain the tendency index Tr.

[0045] 3. Smoothness index Sm: Calculated for each pair of adjacent state vectors along the path Path. The direction vector of change .

[0046] Calculate the angle between continuously changing direction vectors The included angle sequence is obtained. The smoothness Sm can be expressed as the standard deviation of the included angle sequence. The reciprocal of, that is The smaller the standard deviation, the higher the Sm value.

[0047] The system performs a comprehensive judgment on the above indicators based on preset logic. For example, it calculates a comprehensive score. ,in , , The pre-defined non-negative weight coefficients satisfy the following conditions: The specific values ​​can be preset by the administrator in the system configuration according to the evaluation focus of different positions.

[0048] When the overall score Below the set threshold Or, the directional indicator Tr is below its individual threshold for multiple consecutive periods. When the system determines that the training effectiveness conversion status requires intervention, it generates a feedback trigger signal containing the specific judgment conclusion.

[0049] The threshold , This can be determined by analyzing the boundary between normal and abnormal states in historical data, or by initial settings based on experience values ​​by domain experts.

[0050] Step 5: Decision-making based on feedback strategy using utility networks The intelligent feedback decision-making module receives the feedback trigger signal. Internally, this module maintains a feedback policy utility network, which is a graph-structured data model.

[0051] The network contains two types of nodes: behavioral problem pattern nodes. and intervention strategy nodes If the history display strategy Used to solve problems Then at node and There exists a connecting edge between them, and the weight of this edge is... The historical utility estimates of this strategy in solving the problem are recorded.

[0052] The module analyzes the trigger signal to determine the specific behavioral problem pattern at present. And retrieve semantically matching problem nodes in the network. .

[0053] Then, query all slave nodes. Starting from the connecting edges, select the weights among them. The intervention strategy node pointed to by the edge with the highest value This will serve as the output of this decision.

[0054] After the intervention strategy is implemented, the system initiates an asynchronous utility learning process. After a pre-set observation period, the system calculates the improvement in behavioral indicators related to the initial problem and normalizes this improvement into a utility gain value between 0 and 1. .

[0055] Let the relevant indicator values ​​before intervention be... The indicator value after the observation period is (Assuming the lower the indicator value, the better), then the absolute improvement amount .

[0056] Utility gain ,in and The maximum and minimum possible improvements for this type of problem indicator are derived from historical data or set by experts. The utility learning process includes the following updates: 1. Direct Edge Weight Update: Update the weights of the edges corresponding to the strategies used in this iteration. An exponential moving average method is used for the update, with the following formula: .in, The learning rate coefficient is a preset constant between 0 and 1 (e.g., 0.1).

[0057] 2. Diffusion Update: Searching for Nodes Other problem nodes with similar characteristics For each similar node Update its relationship with the strategy according to the following formula. Edge weights: ,in, The diffusion attenuation factor is a factor less than 1. A positive preset constant (e.g., 0.05); This is a function that calculates the similarity between two nodes, such as cosine similarity based on node feature vectors, and returns a value between 0 and 1. This process allows single-experience results to be partially applied to similar problem scenarios.

[0058] Step 6: Generating Personalized Feedback Content The system is based on the selected intervention strategy Generate specific feedback instructions. A content assembly engine, based on a strategy... Based on the type and objective of the task, as well as the specific context in which the employee is currently performing the task, matching materials are retrieved from a tagged knowledge component library, including micro-learning videos, graphic descriptions, and task templates.

[0059] The engine combines employees' past interaction history with various materials to sort and filter the retrieved materials. Finally, according to a predefined content template, it combines the selected materials with the specific parameters of the current task to generate a complete personalized feedback message. This message is sent to the target employee via a system interface.

[0060] Step 7: System Closed Loop and Continuous Operation Once the feedback instruction is issued, the system automatically begins the next work cycle, returning to step one to collect subsequent behavioral data generated by the employee. This new data will be used to generate new quantitative state vectors and evaluation results.

[0061] If the feedback produces the expected results, subsequent evaluation indicators will show positive changes, and the utility gain G of this intervention will be calculated and used to update the feedback strategy utility network, i.e., to execute the utility learning process in step five.

[0062] By continuously executing steps one through seven, the behavior sequence modeling module, dynamic effect evaluation module, and intelligent feedback decision-making module together constitute a complete closed-loop system that continuously evaluates and provides feedback based on actual work behavior sequences. Within this closed loop, the weights in the feedback strategy utility network are continuously updated, thereby achieving adaptive optimization of the system's feedback decision-making capability.

[0063] Further details of specific embodiments The following describes several specific implementation forms of the system.

[0064] The first specific embodiment relates to environmental context awareness.

[0065] The system uses an independently operating environment context awareness module to continuously monitor time-series data of a set of external environmental indicators, such as the response latency of the business server and the number of pending tasks in the entire department. When the dynamic performance evaluation module detects abnormal fluctuations in employee behavior indicators, the environment context awareness module checks whether any environmental indicators have also changed beyond the normal range around the time of the fluctuation.

[0066] If a strong temporal correlation exists, the module sends a suppression signal to the dynamic effect evaluation module. Upon receiving this signal, the dynamic effect evaluation module can temporarily increase the feedback trigger threshold for this behavioral fluctuation, or temporarily exclude related outlier data points when calculating the smoothness index Sm.

[0067] The second specific embodiment involves an assessment of the depth of skill internalization.

[0068] The behavior sequence modeling module includes a skill internalization assessment unit. This unit is activated after the system collects data on employees handling multiple different but related task scenarios. It analyzes the behavioral graphical models generated by employees in each task scenario. .

[0069] The core skills sub-graph From the graph using community detection algorithms (such as the Louvain algorithm) The community with the highest modularity is obtained by extracting it, or by filtering the connected parts in the graph whose edge weights are greater than a preset threshold.

[0070] By comparing different tasks Calculate three internalization indicators: Skill stability, through calculation of different The average Jaccard similarity between nodes and edges is obtained; Skill reuse rate, calculated by statistics across all The proportion of frequently occurring common nodes in the total number of nodes is used to obtain the value. Skill adaptability, by analyzing employees' performance when faced with variations of similar tasks. The structural adjustments made are evaluated to determine whether they are logical, for example, by using a pre-defined set of rules to determine whether adding a node is a reasonable verification step.

[0071] By combining these three indicators, a quantitative score for the depth of skill internalization can be calculated.

[0072] The third specific embodiment involves multi-dimensional skills assessment adaptation.

[0073] The system configures multiple ideal skill progression paths for a comprehensive training course, such as... , Each path corresponds to an independent skill dimension. When the system is configured with multiple ideal skill evolution paths, the dynamic effect evaluation module performs trend calculations using one of the following methods: (1) Dynamic selection: Based on the task type labels involved in the employee's behavioral events within the current time window, the corresponding ideal path is matched and calculated; (2) Weighted fusion: Calculate the employee path and each ideal path Tendency Then, the required weights for each skill dimension are determined based on the current task context. Calculate the general trend ,in .

[0074] In this way, the system supports the evaluation of the effectiveness of training for multi-skilled positions or cross-skills training.

[0075] The fourth specific embodiment involves population evolution map analysis.

[0076] The system includes a population evolution map analysis interface. This interface calls a population situation analysis module, whose functions include: aggregating the quantified state vectors generated by all members within a team in the most recent evaluation period. By using dimensionality reduction algorithms (such as principal component analysis), these high-dimensional state vectors are projected onto a two-dimensional plane to generate a visualized map of team skill distribution.

[0077] This module calculates the average Euclidean distance of all member state points relative to the team's average state point (centroid), which serves as an index of the cluster density of the team's overall state points. The average cosine similarity between each pair of recent behavioral evolution direction vectors of each member is calculated as an index of the convergence of trajectories among members.

[0078] Furthermore, cluster analysis is performed on the two-dimensional projection points to automatically identify employee subgroups with similar skill patterns or development status.

[0079] The degree of synergy and risk of differentiation in team skill development can be assessed by calculating the mobility correlation of the evolution trajectory of behavioral patterns among members, or by analyzing the distance between average state points between different subgroups.

[0080] The above embodiments illustrate the operating principle and various implementation forms of the system. Other variations and implementations based on this technical solution should also be considered within the scope of protection claimed in this application.

[0081] Finally, the following points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection", and "linkage" should be interpreted broadly, and can be mechanical or electrical connections, or internal connections between two components, or direct connections. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change. Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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 dynamic evaluation and feedback system for employee training effectiveness, characterized in that, include: The behavior sequence modeling module is used to obtain behavioral data generated by employees in actual work after training from the enterprise business system, and process the behavioral data into a standardized sequence of behavioral events arranged in chronological order. This module uses a preset period as a time window, extracts and fuses features from the sequence of behavioral events within each time window, and generates a quantitative state vector representing the employee's work mode within that window. The dynamic effect evaluation module, connected to the behavior sequence modeling module, is used to construct the behavior pattern evolution trajectory based on the quantized state vectors of multiple consecutive time windows; by calculating the convergence, trend and smoothness indices of the behavior pattern evolution trajectory, the conversion level and stability of the training effect are dynamically evaluated, and a feedback trigger signal is output based on the evaluation results. The intelligent feedback decision-making module, connected to the dynamic effect evaluation module, is used to respond to the feedback trigger signal. Based on the specific behavioral problem pattern indicated by the evaluation result, this module combines a feedback strategy utility network that records historical utility data of different intervention strategies to select the intervention strategy with the highest expected utility to generate personalized intervention instructions. The behavior sequence modeling module, dynamic effect evaluation module, and intelligent feedback decision-making module are connected in sequence to form a closed-loop system for continuous evaluation and feedback based on actual work behavior sequences.

2. The employee training effectiveness dynamic evaluation and feedback system according to claim 1, characterized in that, The specific methods by which the behavior sequence modeling module extracts and fuses features of behavior event sequences within each time window include: Statistical features are extracted from the sequence of behavioral events to obtain a statistical feature vector; the statistical features include at least the frequency of specific operation event types, the average processing time of tasks, and the number of abnormal events. Structural feature extraction of behavioral event sequences specifically includes: identifying the sequential relationship between behavioral events, constructing a directed weighted graph model, and extracting topological feature vectors from the directed weighted graph model; the topological features include at least the size of the largest connected subgraph, the average clustering coefficient, and the critical path length. The statistical feature vector and the topological feature vector are standardized and then fused and encoded to generate a quantized state vector.

3. The employee training effectiveness dynamic evaluation and feedback system according to claim 1, characterized in that, The specific methods by which the dynamic effect evaluation module calculates the trend indicators include: The evolution trajectory of the behavioral pattern is compared with a predefined ideal skill evolution path, which is pre-divided into multiple consecutive stages. Adaptive segmentation is performed on the evolution trajectory of the aforementioned behavioral patterns; The optimal correspondence between employee trajectory segments and ideal path stages is found through sequence alignment methods. Calculate the average similarity of the state vectors within each corresponding stage, and then perform a weighted average of the similarities of all stage pairs to obtain the trend index.

4. The employee training effectiveness dynamic evaluation and feedback system according to claim 1, characterized in that, The feedback strategy utility network stores information in a graph structure, and its nodes include behavioral problem pattern nodes and intervention strategy nodes. The weight of the connection edge between nodes represents the historical utility estimate obtained by using the corresponding intervention strategy to solve the corresponding behavioral problem. The decision-making process of the intelligent feedback decision-making module includes: activating the corresponding problem node in the feedback strategy utility network according to the current behavioral problem pattern, and selecting the intervention strategy node with the highest weight based on the weights of the edges connected to the node to generate an intervention instruction.

5. The employee training effectiveness dynamic evaluation and feedback system according to claim 4, characterized in that, The weights of the connecting edges in the feedback strategy utility network are dynamically updated, and the update process includes: After the intervention instructions are implemented, the actual utility gain of this intervention is calculated based on the degree of improvement in the evolution trajectory of employee behavior patterns during the subsequent observation period. The weights of the edges corresponding to the strategy used in this instance are updated directly based on the actual utility gain. Part of the utility gain information is diffused to the edges connected to other nodes similar to the current problem node or strategy node according to preset rules for updating.

6. The employee training effectiveness dynamic evaluation and feedback system according to claim 1, characterized in that, It also includes an environment context-aware module for performing the following operations: Continuously monitor time-series data of a set of external environmental indicators; When abnormal fluctuations are detected in the evolution trajectory of the behavior pattern, it is determined whether there is a temporal correlation between the fluctuations and changes in external environmental indicators; If a correlation exists, a suppression signal is sent to the dynamic effect evaluation module so that the dynamic effect evaluation module temporarily adjusts its evaluation sensitivity to such fluctuations.

7. The employee training effectiveness dynamic evaluation and feedback system according to claim 2, characterized in that, The behavior sequence modeling module also includes a skill internalization assessment unit, used to perform the following operations: Analyze the directed weighted graph models generated by employees in different task scenarios, and extract the core skill subgraphs from each model; Skill stability is assessed by calculating the structural similarity between different core skill subgraphs. Skill reuse rate is assessed by statistically analyzing the frequency of occurrence of common nodes across all core skill subgraphs; Skill adaptability is assessed by analyzing the rationality of structural adjustments to core skill subgraphs among similar tasks; By combining stability, reusability, and adaptability indicators, a score for the depth of skill internalization is generated.

8. A dynamic evaluation and feedback system for employee training effectiveness according to claim 1 or 4, characterized in that, When the intelligent feedback decision-making module generates personalized intervention instructions, it calls the feedback content assembly engine to perform the following operations: Based on the selected intervention strategy and the current task context, retrieve matching material from the structured knowledge element base; The retrieved materials are sorted and filtered based on employees' historical preferences; By combining selected materials with contextual parameters according to a predefined template, customized feedback content is generated.

9. The employee training effectiveness dynamic evaluation and feedback system according to claim 3, characterized in that, The system supports configuring multiple ideal skill evolution paths for composite skills training, with each path corresponding to an independent skill dimension; When performing directional calculations, the dynamic effect evaluation module dynamically selects the corresponding ideal skill evolution path based on the employee's current task type, or calculates the directionality of the employee's trajectory with each ideal path separately and then performs a weighted fusion.

10. The dynamic evaluation and feedback system for employee training effectiveness according to claim 1, characterized in that, It also includes a population evolution map analysis module, used to perform the following operations: Aggregate the quantified state vectors of all team members in the most recent evaluation period; By using a dimensionality reduction algorithm, high-dimensional state vectors are projected onto a two-dimensional plane to generate a visual map of team skill distribution. Calculate the cluster density index of the team's overall state points and the convergence index of the trajectories among members; Cluster analysis is performed on the projected points to identify employee subgroups with similar skill patterns or similar development status.

Images