Multi-source heterogeneous behavior data-based full-life-cycle personnel ability dynamic portrait generation method

Abstract

The invention discloses a full-life-cycle personnel ability dynamic portrait generation method based on multi-source heterogeneous behavior data, and relates to the technical field of data processing and intelligent recommendation, and the method comprises the steps: collecting and uniformly marking multi-source behavior data in a learning and business system, extracting dominant scores, recessive behaviors and business efficiency characteristics, and generating a dynamic portrait of the full-life-cycle personnel ability dynamic portrait; constructing a multi-dimensional feature vector; correcting the dominant score based on the recessive behavior, obtaining a real ability evaluation value, and dynamically generating an ability portrait in combination with a forgetting rule and business efficiency; and further performing difference analysis with post capability requirements, identifying skill gaps, and forming an adaptive learning resource recommendation path. By fusing multi-source behavior data and recessive behavior characteristics, accurate evaluation of the real capability of the user is realized, a forgetting rule and business efficiency are introduced to dynamically describe a capability change process, and the timeliness and credibility of a capability portrait are improved; and through capability gap analysis and an adaptive recommendation mechanism, the learning efficiency and post competency are improved.

Description

A method for generating dynamic profiles of full-lifecycle personnel capabilities based on multi-source heterogeneous behavioral data Technical Field

[0001] This invention relates to the field of data processing and intelligent recommendation technology, and in particular to a method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data. Background Technology

[0002] Existing learning management systems or talent management systems typically assess and manage personnel competency levels by recording and statistically analyzing user learning behavior and assessment results. However, current personnel competency assessment schemes still primarily rely on outcome data from examinations or assessments, such as written scores or pass / fail markings. This type of evaluation usually only reflects a user's answers at a specific point in time, failing to depict changes in user behavior during the answering process, and also unable to fully reflect the stability and reliability of a user's mastery of relevant knowledge points. Therefore, in practical applications, inconsistencies between competency assessment results and actual business performance are prone to occur.

[0003] Furthermore, existing personnel competency profiles mostly employ static or cumulative modeling methods. After a user completes learning or acquires a skill tag, the system often assumes that the competency is valid indefinitely, lacking a dynamic modeling mechanism for how competencies change over time. This type of technical solution fails to effectively consider the objective laws of human memory and skill mastery decaying over time. In the absence of continuous business practice or retraining for users, the timeliness and accuracy of competency profiles are difficult to guarantee.

[0004] On the other hand, in the process of job matching or learning resource recommendation based on competency profiles, existing technologies mostly adopt keyword or rule-based matching methods. They lack quantitative modeling methods for the differences between user competency structure and job competency requirements, making it difficult to perform fine-grained alignment and gap analysis of personnel competency and job requirements in a unified feature space. This results in limited targeting and adaptability of recommendation results.

[0005] Therefore, existing technologies still have room for improvement in areas such as multi-source behavioral data fusion, capability assessment credibility correction, capability dynamic evolution modeling over time, and job matching and recommendation path optimization based on vector space. A new technical solution needs to be proposed to improve the authenticity, dynamism, and application effectiveness of personnel capability profiles. Summary of the Invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, this invention provides a method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data, which solves the problems of insufficient authenticity of personnel capability assessment results, lack of dynamic evolution modeling of existing capability profiles over time, and insufficient targeting of existing learning resource recommendations and job matching in the prior art.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0009] In a first aspect, embodiments of the present invention provide a method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data, including: collecting multi-source heterogeneous behavioral data of users in different systems, and uniformly labeling the multi-source heterogeneous behavioral data according to preset knowledge point or business module mapping rules;

[0010] The multi-source heterogeneous behavioral data is cleaned and time-series aligned, and explicit scoring features, implicit behavioral features, and business performance features are extracted to construct a multi-dimensional feature vector.

[0011] Based on the multidimensional feature vector, a knowledge mastery confidence correction model is constructed. The hesitation index corresponding to each knowledge point is calculated according to the implicit behavioral features. The hesitation index is used to generate a confidence coefficient through nonlinear mapping. The confidence coefficient is then used to weight and correct the test score corresponding to the explicit scoring features to obtain the true ability assessment value corresponding to each knowledge point.

[0012] Using the actual ability assessment value as the initial ability state, an ability decay model is constructed based on the forgetting law, and the collected business efficiency characteristics are used as dynamic adjustment parameters to correct the ability decay process. The remaining mastery of each knowledge point at the current moment is calculated, thereby generating the user's dynamic ability vector.

[0013] The dynamic capability vector is compared with the standard capability vector corresponding to the target position or target capability requirements to identify skill gaps. Based on the size of the skill gap, the degree of capability decay, and the business relevance weight, an adaptive learning resource recommendation path is generated.

[0014] As a preferred embodiment of the method for generating dynamic profiles of full-lifecycle personnel capabilities based on multi-source heterogeneous behavioral data according to the present invention, the method includes: uniformly labeling the multi-source heterogeneous behavioral data according to preset knowledge point or business module mapping rules, including:

[0015] Based on a pre-established knowledge point identifier or business module identifier system, corresponding knowledge point identifiers or business module identifiers are assigned to different data contents in the multi-source heterogeneous behavioral data.

[0016] The multi-source heterogeneous behavioral data corresponding to the assessment questions are labeled as data associated with the knowledge point identifiers corresponding to the assessment questions;

[0017] Multi-source heterogeneous behavioral data corresponding to specific business operations or business function modules are labeled as data associated with the business module identifier;

[0018] The annotated multi-source heterogeneous behavioral data is then organized into a data set indexed by knowledge point identifiers or business module identifiers.

[0019] As a preferred embodiment of the method for generating dynamic profiles of personnel capabilities throughout the entire lifecycle based on multi-source heterogeneous behavioral data as described in this invention, the explicit scoring features include: the original score value corresponding to a single knowledge point, the proportion of the score of a single knowledge point to the full score of that knowledge point, and the historical average score or weighted score value of the same knowledge point in multiple assessments; the implicit behavioral features include: the statistical feature value corresponding to the dwell time on a single question, the cumulative value or normalized value of the number of times the option is modified, the statistical value of the number of times the browser focus is removed, and the trajectory length, speed, or fluctuation amplitude feature corresponding to the mouse sliding trajectory coordinate sequence; the business efficiency features include: the frequency or duration feature corresponding to the operation activity, the statistical feature of the frequency of knowledge base or document browsing, and the quantitative score or level mapping value of the task completion quality under simulated training scenarios.

[0020] As a preferred embodiment of the method for generating dynamic profiles of full-lifecycle personnel capabilities based on multi-source heterogeneous behavioral data according to the present invention, the method includes: constructing a knowledge mastery confidence correction model based on the multi-dimensional feature vector, and calculating the hesitation index corresponding to each knowledge point according to the implicit behavioral features, including:

[0021] For each knowledge point, the corresponding implicit behavioral features are extracted from the multidimensional feature vector, and the hesitation index of the knowledge point is calculated using the following formula:

[0022] in, This represents the user's level of hesitation regarding knowledge point k. This is the normalized value of the single-question dwell time feature corresponding to knowledge point k. This is the normalized value of the feature representing the number of times the option was modified corresponding to knowledge point k.

[0023] α is the normalized value of the number of times the browser focus is removed corresponding to knowledge point k, and β and γ are the behavior weight coefficients.

[0024] As a preferred embodiment of the method for generating dynamic profiles of full-lifecycle personnel capabilities based on multi-source heterogeneous behavioral data described in this invention, the method involves: generating a confidence coefficient from the hesitation index through nonlinear mapping, and using the confidence coefficient to weight and correct the test scores corresponding to the explicit scoring features to obtain the true capability assessment value corresponding to each knowledge point, including:

[0025] Based on the hesitation index, the corresponding confidence coefficient is calculated using a nonlinear mapping function. The expression for the confidence coefficient is as follows:

[0026] in, Let λ be the confidence coefficient corresponding to knowledge point k, and λ be the adjustment parameter.

[0027] Based on the confidence coefficient, the explicit test score is weighted and corrected to obtain the true ability assessment value corresponding to knowledge point k. The formula for calculating the true ability assessment value is as follows:

[0028] in, This represents the actual ability assessment value for knowledge point k. This represents the original score for knowledge point k on the exam paper.

[0029] Before or during the construction of the capability decay model based on the actual capability assessment value, the process also includes performing a business-related reverse verification of the actual capability assessment value. The specific steps are as follows:

[0030] When the actual capability assessment value corresponding to a knowledge point exceeds the preset capability threshold, the business performance characteristics of that knowledge point in the actual business system are retrieved, and the quality density statistical analysis of the business performance characteristics is performed.

[0031] If the proportion of low-quality behaviors or the density of low-quality results in the business performance characteristics is higher than the preset security standard, it is determined that the actual capability assessment value of the knowledge point has a model deviation, an abnormal alarm is automatically triggered, and the actual capability assessment value corresponding to the knowledge point is adjusted downward to obtain the verified capability assessment value.

[0032] As a preferred embodiment of the method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data as described in this invention, the method includes: constructing a capability decay model based on forgetting patterns and using collected business performance characteristics as dynamic adjustment parameters to correct the capability decay process, including: the capability decay model adopts a capability decay function that incorporates forgetting patterns, and the formula for the capability decay function is as follows:

[0033] in, This represents the remaining mastery level of the corresponding knowledge point at the current moment. t is the actual ability assessment value corresponding to the knowledge point, t is the time interval from the completion of the calculation of the actual ability assessment value, and M is the memory strength parameter.

[0034] As a preferred embodiment of the method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data as described in this invention, the method includes: identifying skills gaps, including:

[0035] Construct a unified multidimensional feature space and calculate the difference vector between the dynamic capability vector and the standard capability vector corresponding to the target position or target capability requirements:

[0036]

[0037] in, For dynamic capability vectors, This refers to the standard capability vector corresponding to the target position or target capability requirements. It is a difference vector.

[0038] As a preferred embodiment of the method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data as described in this invention, the method includes: generating adaptive learning resource recommendation paths based on skill gap size, capability decay degree, and business relevance weights, including:

[0039] In the process of generating the adaptive learning resource recommendation path, the learning resource recommendation path is constructed as a directed acyclic graph structure, and the directed acyclic graph structure has the following dynamic adjustment mechanism:

[0040] When a user is taking a test or learning assessment at a path node in the directed acyclic graph, and the confidence coefficient of that path node is lower than a preset threshold, the system will automatically revert the recommended path to the preceding basic node corresponding to that node.

[0041] When a user is taking a test or learning assessment at a path node in the directed acyclic graph, and the confidence coefficient of that path node is higher than a preset threshold, the system automatically skips subsequent basic nodes that have the same content or ability level as that node.

[0042] After a user completes the learning behavior in the learning resource recommendation path, a reinforcement learning algorithm is used to update the weight parameters of each influencing factor in the recommendation priority score in reverse, using the improvement of the user's corresponding ability dimension before and after learning as the reward function.

[0043] In a second aspect, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, they implement the steps of the method for generating dynamic profiles of full lifecycle personnel capabilities based on multi-source heterogeneous behavioral data as described in the first aspect of the present invention.

[0044] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the method for generating a dynamic profile of a person's capabilities throughout the entire lifecycle based on multi-source heterogeneous behavioral data as described in the first aspect of the present invention.

[0045] The beneficial effects of this invention are as follows: This invention integrates multi-source heterogeneous behavioral data to construct a knowledge mastery confidence correction model, correcting explicit scores and implicit behavioral characteristics to improve the accuracy of ability assessment. By dynamically adjusting ability decay through forgetting patterns and business efficiency, the ability profile evolves dynamically over time and with business practice. Based on vector space analysis, the gap between user abilities and job standards is analyzed, quantifying skill gaps and generating adaptive learning resource recommendation paths, achieving precise matching and dynamic optimization of ability structure and job requirements. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 is a flowchart of the method of the present invention. Detailed Implementation

[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0049] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0050] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0051] Example 1

[0052] Referring to Figure 1, which illustrates the first embodiment of the present invention, this embodiment provides a method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data, including:

[0053] S1: Collect multi-source heterogeneous behavioral data of users in different systems, and uniformly label the multi-source heterogeneous behavioral data according to preset knowledge point or business module mapping rules.

[0054] Furthermore, based on a pre-established knowledge point identifier or business module identifier system, corresponding knowledge point identifiers or business module identifiers are assigned to different data contents in the multi-source heterogeneous behavioral data.

[0055] The multi-source heterogeneous behavioral data corresponding to the assessment questions are labeled as data associated with the knowledge point identifiers corresponding to the assessment questions;

[0056] Multi-source heterogeneous behavioral data corresponding to specific business operations or business function modules are labeled as data associated with the business module identifier;

[0057] The annotated multi-source heterogeneous behavioral data is then organized into a data set indexed by knowledge point identifiers or business module identifiers.

[0058] It should be noted that the LMS system backend database is connected via ODBC / JDBC interface, and the online exam scores, course video viewing time, chapter test pass rate and skill certificate records of the target object are periodically extracted and defined as explicit ability data.

[0059] JavaScript event listeners or event tracking scripts are pre-defined in the DOM nodes of the front-end page to capture data such as mouse movement trajectory, hover position and duration, time spent on a single question page, number of option clicks / recalls, and browser window focus switching in real time, and define them as process behavior data.

[0060] By connecting to the enterprise's internal business systems via RESTful API, and extracting actual business output data (Dbiz) based on a pre-defined job KPI mapping table, a business verification dataset is formed.

[0061] Based on a pre-established knowledge point identification system or business module identification system, each record in the multi-source data is assigned a corresponding knowledge point identification or business module identification: data associated with the assessment questions is labeled as the corresponding knowledge point; data associated with business operations / functional modules is labeled as the corresponding business module.

[0062] The labeled data is organized according to knowledge points or business modules to form a multidimensional data set that can be used for subsequent analysis.

[0063] S2: Perform data cleaning and time-series alignment on the multi-source heterogeneous behavioral data, and extract explicit score features, implicit behavioral features and business efficiency features respectively to construct a multi-dimensional feature vector.

[0064] Furthermore, explicit scoring characteristics include: the original score value corresponding to a single knowledge point, the proportion of the score of a single knowledge point to the full score of that knowledge point, and the historical average score or weighted score value of the same knowledge point in multiple assessments; implicit behavioral characteristics include: the statistical characteristic value corresponding to the dwell time on a single question, the cumulative or normalized value of the number of times the option is modified, the statistical value of the number of times the browser focus is removed, and the trajectory length, speed, or fluctuation amplitude characteristics corresponding to the mouse swipe trajectory coordinate sequence; business efficiency characteristics include: the frequency or duration characteristics corresponding to the operational activity, the statistical characteristics of the frequency of knowledge base or document browsing, and the quantitative score or level mapping value of the task completion quality in simulated training scenarios.

[0065] It should be noted that for the collected behavioral data, a time threshold judgment logic was set: records with a single interaction interval of <50ms or >30min were judged as noise and removed. KNN interpolation was used for missing values, with adaptive completion based on the average of personnel at the same position and level. Temporal difference calculations were performed on mouse trajectory, click logs, and other data to extract behavioral indicators: Tact: actual dwell time on a single question; Clost: number of focus loss times; Cmod: number of answer modifications. These indicators were aggregated into a behavioral feature vector Vbehavior. Explicit score features, implicit behavioral features, and business performance features were standardized using Z-Score, with the natural day as the smallest time granularity. The cleaned and standardized multi-source data were aligned along the time axis to construct a user-time-skill three-dimensional data cube, providing an input sequence for capability decay calculation.

[0066] S3: Construct a knowledge mastery confidence correction model based on the multidimensional feature vector, calculate the hesitation index corresponding to each knowledge point according to the implicit behavioral features, generate a confidence coefficient from the hesitation index through nonlinear mapping, and use the confidence coefficient to weight and correct the test score corresponding to the explicit scoring feature to obtain the true ability assessment value corresponding to each knowledge point.

[0067] Furthermore, for each knowledge point, the corresponding implicit behavioral features are extracted from the multidimensional feature vector, and the hesitation index of the knowledge point is calculated using the following formula:

[0068] in, This represents the user's level of hesitation regarding knowledge point k. This is the normalized value of the single-question dwell time feature corresponding to knowledge point k. This is the normalized value of the feature representing the number of times the option was modified corresponding to knowledge point k.

[0069] α is the normalized value of the number of times the browser focus is removed corresponding to knowledge point k, and β and γ are the behavior weight coefficients.

[0070] Furthermore, based on the hesitation index, the corresponding confidence coefficient is calculated using a nonlinear mapping function, and the expression for the confidence coefficient is as follows:

[0071] in, Let λ be the confidence coefficient corresponding to knowledge point k, and λ be the adjustment parameter.

[0072] Based on the confidence coefficient, the explicit test score is weighted and corrected to obtain the true ability assessment value corresponding to knowledge point k. The formula for calculating the true ability assessment value is as follows:

[0073] in, This represents the actual ability assessment value for knowledge point k. This represents the original score for knowledge point k on the exam paper.

[0074] Before or during the construction of the capability decay model based on the actual capability assessment value, the process also includes performing a business-related reverse verification of the actual capability assessment value. The specific steps are as follows:

[0075] When the actual capability assessment value corresponding to a knowledge point exceeds the preset capability threshold, the business performance characteristics of that knowledge point in the actual business system are retrieved, and the quality density statistical analysis of the business performance characteristics is performed.

[0076] If the proportion of low-quality behaviors or the density of low-quality results in the business performance characteristics is higher than the preset security standard, it is determined that the actual capability assessment value of the knowledge point has a model deviation, an abnormal alarm is automatically triggered, and the actual capability assessment value corresponding to the knowledge point is adjusted downward to obtain the verified capability assessment value.

[0077] It should be noted that α, β, and γ are preset based on historical big data statistics or expert experience: for example, α=0.5, β=0.3, γ=0.2, indicating that the dwell time has the greatest impact on the degree of hesitation, followed by the number of modifications, and the number of times the focus is moved out is the least.

[0078] When the Strue of knowledge point k exceeds the preset capability threshold (e.g., 80 / 100 points), the business verification step is triggered: retrieve the business performance feature Dbiz corresponding to the knowledge point, including task completion quality, operation accuracy, knowledge base query frequency, etc.; perform quality density statistics on the business performance features, such as calculating the proportion of low-quality behaviors or the density of low-quality results.

[0079] If the density of low-quality behaviors or results exceeds the preset safety standard (e.g., 20%), it is determined that there is a model bias in the ability of that knowledge point; the system automatically triggers an anomaly alarm and adjusts Strue accordingly.

[0080] S4: Using the actual ability assessment value as the initial ability state, a capability decay model is constructed based on the forgetting law, and the collected business efficiency characteristics are used as dynamic adjustment parameters to correct the capability decay process. The remaining mastery of each knowledge point at the current moment is calculated, thereby generating the user's dynamic capability vector.

[0081] Furthermore, the ability decay model employs an ability decay function that incorporates forgetting patterns, and the formula for the ability decay function is as follows:

[0082] in, This represents the remaining mastery level of the corresponding knowledge point at the current moment. t is the actual ability assessment value corresponding to the knowledge point, t is the time interval from the completion of the calculation of the actual ability assessment value, and M is the memory strength parameter.

[0083] It should be noted that, to consider the reinforcing effect of business practice on capabilities, dynamic memory strength is introduced:

[0084]

[0085] in, Basic memory strength represents the ability to retain memory based solely on default memory retention during learning activities. φ is a business performance characteristic value, reflecting the actual operational quality, task completion rate, and reuse frequency of users in the business system for knowledge point k; φ is an adjustment coefficient used to quantify the enhancing effect of business performance on memory strength.

[0086] This dynamic mechanism ensures that users retain the strength of their knowledge points when performing frequent or high-quality tasks in actual business operations. Increase, thereby slowing down R tk The decay rate simulates the effect of "learning by doing and consolidating memory".

[0087] S5: Perform gap analysis between the dynamic capability vector and the standard capability vector corresponding to the target position or target capability requirements to identify skill gaps, and generate an adaptive learning resource recommendation path based on the size of the skill gap, the degree of capability decay, and the business relevance weight.

[0088] Furthermore, in the process of generating the adaptive learning resource recommendation path, the learning resource recommendation path is constructed as a directed acyclic graph structure, which has the following dynamic adjustment mechanism:

[0089] When a user is taking a test or learning assessment at a path node in the directed acyclic graph, and the confidence coefficient of that path node is lower than a preset threshold, the system will automatically revert the recommended path to the preceding basic node corresponding to that node.

[0090] When a user is taking a test or learning assessment at a path node in the directed acyclic graph, and the confidence coefficient of that path node is higher than a preset threshold, the system automatically skips subsequent basic nodes that have the same content or ability level as that node.

[0091] After a user completes the learning behavior in the learning resource recommendation path, a reinforcement learning algorithm is used to update the weight parameters of each influencing factor in the recommendation priority score in reverse, using the improvement of the user's corresponding ability dimension before and after learning as the reward function.

[0092] It should be noted that, in addition to the user's dynamic capability vector and the job's standard capability vector, features such as user learning preferences, learning history, and behavioral activity can also be introduced as weighting factors to generate personalized paths.

[0093] When constructing a Directed Acyclic Graph (DAG), the system can use factors such as ability level, knowledge point dependencies, and resource difficulty level as node attributes to form a weighted directed acyclic graph. This ensures that the recommended path conforms to learning logic while also meeting personalized needs. When backtracking to previous basic nodes, the backtracking depth can be selected based on time intervals or historical performance. For example, it can backtrack to the most recent node where the mastery level was below a threshold, rather than fixing the previous node, to avoid excessive repetition. The system can automatically identify repeatedly performed and repetitive basic knowledge points, merging them into a single node or skipping similar content, thus compressing and accelerating learning resources.

[0094] The DAG allows for conditional branching nodes, enabling the selection of different paths based on user performance or business practice results. This allows learning paths to adaptively switch to address different weaknesses or business scenarios. The system incorporates a multi-dimensional reward function into the reinforcement learning algorithm, considering not only the magnitude of ability improvement but also learning efficiency (learning duration), business application effectiveness (task completion quality), and user preference matching. Reverse updates of recommendation weight parameters can be performed in stages, such as weekly or batch-by-task iterations, ensuring stable model convergence and real-time adaptation to new behavioral data. The system can statistically analyze historical path performance to create a "learning path performance database," supporting subsequent user group-based recommendations and cold-start strategies for new users.

[0095] Example 2

[0096] This embodiment is the second embodiment of the present invention. This embodiment provides a method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data. In order to verify the beneficial effects of the present invention, a simulation experiment is conducted for scientific demonstration.

[0097] To further verify the effectiveness and superiority of the "Dynamic Profile Generation System and Method for Full Life Cycle Personnel Capabilities Based on Multi-Source Heterogeneous Behavioral Data" proposed in this invention in practical applications, this embodiment collects historical data from annual company training assessments and self-learning on the learning platform for retrospective testing and comparative analysis.

[0098] Data source: Analysis was conducted on the company's annual training and assessment data, as well as historical data of trainees' self-study on the company's learning platform, totaling data records of 326 trainees.

[0099] Dataset size: Includes approximately 15,000 online exam answer logs (including micro-behaviors such as mouse trajectories), 180,000 self-learning behaviors, and 538 monthly performance evaluation reports.

[0100] Experimental Groups:

[0101] Control group: The traditional learning management system uses the profile generation logic, that is: ability value = exam score, and the score is permanently valid. The recommendation algorithm is based on the simple "recommend before the exam" logic.

[0102] Experimental group: The method described in this invention is adopted, namely: enabling the hesitation correction model, enabling the Ebbinghaus dynamic decay model combined with business efficiency and the vector space recommendation algorithm.

[0103] To quantitatively evaluate system performance, the following five core evaluation metrics are defined:

[0104] a. Accuracy of competency assessment: Using the expert scores in the employee's subsequent quarterly "technical defense" or "practical assessment" as the "true value", the Pearson correlation coefficient between the profile score generated by the system and the expert scores is calculated.

[0105] b. Skills Timeliness Error (MAE): The mean absolute error of (system prediction score - surprise test score) is calculated when an employee stops learning a certain skill (such as SQL injection) for 90 days.

[0106] c. Accelerated Learning Cycle (ACC): A comparison of the efficiency of the learning path based on gap analysis and pruning mechanisms in vector space with traditional learning paths is conducted. The average time taken for new employees to become fully competent in their roles (meeting all dimensions of the employee profile) is statistically analyzed. The formula for calculating the entire learning cycle is as follows:

[0107]

[0108] in, The total number of sample users participating in the test. Let i be the time point at which the i-th user begins receiving the recommended path. Let be the time point at which the i-th user passes the final competency assessment.

[0109] d. Course Completion Rate (CCR)

[0110] In existing technologies (control group), recommended content often contains a large amount of redundant information already known to the user or non-urgent needs, leading to low user learning motivation, high dropout rates, and typically low completion rate (CCR) scores. In this invention (experimental group), thanks to dynamic gap analysis and precise pruning technology, the system only pushes content to users in areas where they urgently need to fill knowledge gaps. Users perceive the practicality of the content, thus maintaining high learning engagement. The course completion rate is calculated using the following formula:

[0111]

[0112] in, This refers to the total number of learning units (such as video courses, documents, and exercises) that the system pushes to users. The number of learning units that the user has actually completed (e.g., video playback progress > 95% or passing unit tests).

[0113] e. Redundancy of knowledge points Knowledge redundancy refers to the proportion of knowledge points in the personalized learning path generated by the system that belong to the user's already mastered skill domain (i.e., do not require repeated learning). This metric measures the effectiveness of the recommendation algorithm in eliminating invalid content and achieving accurate recommendations. The formula for calculating knowledge point redundancy is as follows:

[0114]

[0115] in, This refers to the number of knowledge points in the recommended path that overlap with the user's existing knowledge (i.e., redundant points). A set of recommended learning paths, The set of user profiles for their acquired knowledge / skills is as follows.

[0116] The comparison of the "true and false" effect of the real ability assessment addresses the common phenomena of "cramming" and "guessing" in online assessments by comparing the correlation between the scores generated by the two systems and the scores given by offline experts in practical operation.

[0117]

[0118] The accuracy of predicting "anti-forgetting" using dynamic profiles was compared among 73 trainees, and their skill retention was tracked 90 days after the end of the training.

[0119] Control group results: The system showed that the ability score remained at 88.5 points without decay, while the actual average score of the surprise test was 59.3 points, with a prediction error (MAE) of 29.2. The system seriously overestimated the current ability of employees, leading to misjudgment by managers.

[0120] Experimental results: The system showed that the ability score decayed to 71.2 points (the model automatically calculated the forgetting), the actual average score of the surprise test was 65.8 points, and the prediction error MAE=5.4.

[0121] A comparative experiment was constructed to objectively evaluate the performance advantages of the learning path recommendation method proposed in this invention in real-world business scenarios. The experiment selected a traditional rule-based filtering recommendation algorithm as a benchmark control group, focusing on quantitative evaluation of three core indicators: knowledge point redundancy, course completion rate, and average competence cycle. By analyzing the experimental data, the aim was to verify the technical effectiveness of this invention in eliminating invalid computational load, improving recommendation accuracy, and shortening the user competence development path, thereby demonstrating that this solution has significant efficiency improvements and resource-saving advantages compared to existing technologies.

[0122]

[0123] Results Analysis: The experimental group, through gap analysis and pruning mechanisms in vector space, automatically skipped over the existing skills already mastered by employees and precisely targeted forgotten points that "urgently need reinforcement." This avoided repetitive learning and significantly shortened the talent development cycle.

[0124] Conclusion: The experimental data above show that, compared with the prior art, the method of the present invention effectively reduces a large amount of content that employees have already mastered in the learning content recommended to them, reduces the redundancy of knowledge points from 25% to 5%, and reduces the learning of meaningless repetitive knowledge points by 80%, which can effectively improve the learning efficiency of trainees.

[0125] By identifying the gap analysis system and pinpointing the content needs of learners, the system only recommends knowledge that they don't yet understand but urgently require, thus meeting their learning needs and encouraging them to complete the recommended courses. Course completion rate increased from 62% to 88%, an effective improvement of 41.9%. The reduced redundancy of knowledge points and the increased course completion rate shortened the entire training cycle for learners from 45 days to 31 days. The adaptive recommendation path based on vector space improved the efficiency of person-job matching and talent development by more than 30%.

[0126] This embodiment also provides a computer device applicable to the method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data as proposed in the above embodiment.

[0127] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0128] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for generating dynamic profiles of full-lifecycle personnel capabilities based on multi-source heterogeneous behavioral data as proposed in the above embodiments.

[0129] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data, characterized in that: include: Collect multi-source heterogeneous behavioral data of users in different systems, and uniformly label the multi-source heterogeneous behavioral data according to preset knowledge point or business module mapping rules; The multi-source heterogeneous behavioral data is cleaned and time-series aligned, and explicit scoring features, implicit behavioral features, and business performance features are extracted to construct a multi-dimensional feature vector. Based on the multi-dimensional feature vector, a knowledge mastery confidence correction model is constructed. The hesitation index corresponding to each knowledge point is calculated according to the implicit behavioral features. The hesitation index is used to generate a confidence coefficient through nonlinear mapping. The confidence coefficient is then used to weight and correct the test scores corresponding to the explicit scoring features to obtain the true ability assessment value corresponding to each knowledge point. Using actual ability assessment values ​​as the initial ability state, an ability decay model is constructed based on the forgetting curve, and collected business performance characteristics are used as dynamic adjustment parameters to correct the ability decay process. The remaining mastery of each knowledge point at the current moment is calculated to generate the user's dynamic ability vector. The dynamic ability vector is then compared with the standard ability vector corresponding to the target position or target ability requirements to identify skill gaps. Based on the size of the skill gap, the degree of ability decay, and the business relevance weight, an adaptive learning resource recommendation path is generated.

2. The method for generating dynamic profiles of full-lifecycle personnel capabilities based on multi-source heterogeneous behavioral data as described in claim 1, characterized in that, The step of uniformly labeling the multi-source heterogeneous behavioral data according to preset knowledge point or business module mapping rules includes: assigning corresponding knowledge point identifiers or business module identifiers to different data contents in the multi-source heterogeneous behavioral data based on a pre-established knowledge point identifier or business module identifier system; labeling the multi-source heterogeneous behavioral data corresponding to the assessment question as data associated with the knowledge point identifier corresponding to the assessment question; labeling the multi-source heterogeneous behavioral data corresponding to specific business operations or business function modules as data associated with the business module identifier; and organizing the labeled multi-source heterogeneous behavioral data into a data set indexed by the knowledge point identifier or business module identifier.

3. The method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data as described in claim 1, characterized in that... The explicit scoring features include: the original score value corresponding to a single knowledge point, the proportion of the score of a single knowledge point to the full score of that knowledge point, and the historical average score or weighted score value of the same knowledge point in multiple assessments; the implicit behavioral features include: the statistical feature value corresponding to the dwell time on a single question, the cumulative or normalized value of the number of times the option is modified, the statistical value of the number of times the browser focus is removed, and the trajectory length, speed, or fluctuation amplitude feature corresponding to the mouse sliding trajectory coordinate sequence; the business efficiency features include: the frequency or duration feature corresponding to the operation activity, the statistical feature of the frequency of knowledge base or document browsing, and the quantitative score or level mapping value of the task completion quality in simulated training scenarios.

4. The method for generating dynamic profiles of full-lifecycle personnel capabilities based on multi-source heterogeneous behavioral data as described in claim 1, characterized in that, The step of constructing a knowledge mastery confidence correction model based on the multidimensional feature vector, and calculating the hesitation index corresponding to each knowledge point according to the implicit behavioral features, includes: for each knowledge point, extracting the corresponding implicit behavioral features from the multidimensional feature vector, and calculating the hesitation index of the knowledge point using the following formula: ;in, This represents the user's level of hesitation regarding knowledge point k. This is the normalized value of the single-question dwell time feature corresponding to knowledge point k. This is the normalized value of the feature representing the number of times the option was modified corresponding to knowledge point k. α is the normalized value of the number of times the browser focus is removed corresponding to knowledge point k, and β and γ are the behavior weight coefficients.

5. The method for generating dynamic profiles of full-lifecycle personnel capabilities based on multi-source heterogeneous behavioral data as described in claim 1, characterized in that, The step of generating a confidence coefficient from the hesitation index through a nonlinear mapping, and then using the confidence coefficient to weight and correct the test scores corresponding to the explicit scoring features to obtain the true ability assessment value for each knowledge point, includes: calculating the corresponding confidence coefficient based on the hesitation index using a nonlinear mapping function, wherein the expression for the confidence coefficient is as follows: ;in, Let λ be the confidence coefficient corresponding to knowledge point k, and λ be an adjustment parameter. Based on the confidence coefficient, the explicit test score is weighted and corrected to obtain the true ability assessment value corresponding to knowledge point k. The formula for calculating the true ability assessment value is as follows: ;in, This represents the actual ability assessment value for knowledge point k. The original score for knowledge point k is used. Before or during the construction of the capability decay model based on the actual capability assessment value, the method further includes business reverse verification of the actual capability assessment value. The specific steps are as follows: when the actual capability assessment value corresponding to the knowledge point exceeds the preset capability threshold, the business performance characteristics of the knowledge point in the actual business system are retrieved, and the quality density statistical analysis of the business performance characteristics is performed. If the proportion of low-quality behaviors or the density of low-quality results in the business performance characteristics is higher than the preset security standard, it is determined that the actual capability assessment value of the knowledge point has a model deviation, an abnormal alarm is automatically triggered, and the actual capability assessment value corresponding to the knowledge point is adjusted downward to obtain the verified capability assessment value.

6. The method for generating dynamic profiles of full-lifecycle personnel capabilities based on multi-source heterogeneous behavioral data as described in claim 1, characterized in that, The process of constructing a capability decay model based on the forgetting curve and using collected business performance characteristics as dynamic adjustment parameters to correct the capability decay process includes: the capability decay model adopts a capability decay function that incorporates the forgetting curve, and the formula for the capability decay function is as follows: ;in, This represents the remaining mastery level of the corresponding knowledge point at the current moment. t is the actual ability assessment value corresponding to the knowledge point, t is the time interval from the completion of the calculation of the actual ability assessment value, and M is the memory strength parameter.

7. The method for generating dynamic profiles of full-lifecycle personnel capabilities based on multi-source heterogeneous behavioral data as described in claim 1, characterized in that, The identification of skills gaps includes: constructing a unified multidimensional feature space and calculating the difference vector between the dynamic ability vector and the standard ability vector corresponding to the target position or target ability requirement. ;in, For dynamic capability vectors, This refers to the standard capability vector corresponding to the target position or target capability requirements. It is a difference vector.

8. The method for generating dynamic profiles of personnel capabilities throughout their entire lifecycle based on multi-source heterogeneous behavioral data as described in claim 1, characterized in that, The method for generating an adaptive learning resource recommendation path based on skill gap size, ability decay degree, and business relevance weights includes: During the generation of the adaptive learning resource recommendation path, the learning resource recommendation path is constructed as a directed acyclic graph (DAG), which has the following dynamic adjustment mechanisms: when a user's confidence coefficient for a path node in the DAG is lower than a preset threshold during a test or learning assessment, the system automatically reverts the recommendation path to the preceding basic node; when a user's confidence coefficient for a path node in the DAG is higher than a preset threshold during a test or learning assessment, the system automatically skips subsequent basic nodes that are duplicated in content or have the same ability level as that node; after the user completes the learning behavior in the recommended learning resource path, a reinforcement learning algorithm is used, with the improvement in the user's corresponding ability dimension before and after learning as the reward function, to update the weight parameters of each influencing factor in the recommendation priority score in reverse.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for generating dynamic profiles of full-lifecycle personnel capabilities based on multi-source heterogeneous behavioral data as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for generating dynamic profiles of full-lifecycle personnel capabilities based on multi-source heterogeneous behavioral data as described in any one of claims 1 to 8.

Images