Intelligent learning path recommendation method and system based on dynamic state

Abstract

The embodiment of the invention provides an intelligent learning path recommendation method and system based on a dynamic state. The method is applied to the technical field of intelligent learning recommendation, and comprises the following steps: calculating a skill adaptation weight, a load weight and an emergency degree weight in real time based on a dynamic state of a target employee, including a current skill improvement condition, a workload and a task emergency degree; combining the skill matching degree, the learning strength and the task association degree of each learning unit in the candidate learning path, and performing comprehensive evaluation by using a multi-weight scoring function; and sorting the paths according to the comprehensive score, generating a personalized recommended learning path set and sending the personalized recommended learning path set to an employee terminal, thereby realizing intelligent learning path recommendation with dynamic adaptation and accurate matching. According to the scheme, personalized learning path pushing aiming at actual post requirements and working states of the employees can be realized, correlation, urgency and acceptability of learning contents are improved, learning efficiency and task adaptability are remarkably enhanced, and the employees are helped to quickly compete with post targets.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of intelligent learning recommendation, in particular to an intelligent learning path recommendation method and system based on dynamic state. BACKGROUND

[0002] Under the background of accelerating digital transformation and intensifying enterprise competition, employee skill improvement has become a core element for organizations to maintain competitiveness. However, there is a common problem of insufficient dynamic adaptability in enterprise training practice: on the one hand, employee skill needs change rapidly with business goals, project urgency and individual development paths, and traditional static training programs are difficult to respond in real time; on the other hand, employee workloads differ significantly, and forced learning under high workload may lead to decreased efficiency, while low workload periods may waste resources. In addition, task urgency directly affects learning priority, but existing methods often ignore this key dimension.

[0003] Today, fixed learning paths are recommended based on preset rules (such as job skill trees and course priority lists), and some methods adjust the rules through regular surveys (such as quarterly skill assessments).

[0004] However, existing technologies do not integrate employee real-time state data (such as daily workloads and the proportion of urgent tasks), resulting in conflicts between recommended paths and employee actual available time, or ignoring the priority learning needs of key tasks. Fixed weights cannot adapt to different scenarios (such as increasing the "task urgency" weight during project sprints and increasing the "skill adaptation" weight during skill gap filling periods), leading to a disconnect between recommended results and business goals. SUMMARY

[0005] To solve the problems of the prior art, the present disclosure provides an intelligent learning path recommendation method and system based on dynamic state. The present disclosure solves the problem that existing technologies do not integrate employee real-time state data (such as daily workloads and the proportion of urgent tasks), resulting in conflicts between recommended paths and employee actual available time, or ignoring the priority learning needs of key tasks. Fixed weights cannot adapt to different scenarios (such as increasing the "task urgency" weight during project sprints and increasing the "skill adaptation" weight during skill gap filling periods), leading to a disconnect between recommended results and business goals.

[0006] According to a first aspect of the present disclosure, an intelligent learning path recommendation method based on dynamic state is provided, comprising: obtaining a current skill improvement state, a current workload and a current task urgency value of a target employee, calculating a skill adaptation weight, a workload weight and an urgency weight according to the current skill improvement state, the current workload, the current task urgency value and a preset weight function formula; The skill matching degree, the learning intensity value and the task correlation degree of each learning unit in each candidate learning path are acquired, and the comprehensive score of each candidate learning path is calculated according to the skill adaptation weight, the work load weight, the urgency weight, the skill matching degree, the learning intensity value, the task correlation degree and a preset scoring formula; The candidate learning paths are sorted according to the comprehensive scores, a recommended learning path set of the target employee is determined, and the recommended learning path set is sent to a terminal device of the target employee.

[0007] According to a second aspect of the present disclosure, an intelligent learning path recommendation system based on dynamic states is provided for performing the method as described in the first aspect, comprising: a state evaluation module configured to acquire a current skill improvement state, a current work load and a current task urgency value of a target employee, and calculate a skill adaptation weight, a work load weight and an urgency weight according to the current skill improvement state, the current work load, the current task urgency value and a preset weight function formula; A path evaluation module is configured to acquire the skill matching degree, the learning intensity value and the task correlation degree of each learning unit in each candidate learning path, and calculate the comprehensive score of each candidate learning path according to the skill adaptation weight, the work load weight, the urgency weight, the skill matching degree, the learning intensity value, the task correlation degree and a preset scoring formula. A recommendation pushing module is configured to sort the candidate learning paths according to the comprehensive scores, determine a recommended learning path set of the target employee, and send the recommended learning path set to a terminal device of the target employee.

[0008] According to a third aspect of the present disclosure, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method as described above.

[0009] In the intelligent learning path recommendation method and system based on dynamic states as provided above, the current skill improvement state, the work load and the task urgency of the employee are dynamically acquired, the skill matching degree, the learning intensity and the task correlation degree of each learning path are combined, the comprehensive score is intelligently calculated based on a preset scoring formula and sorted for recommendation, the personalized learning path pushing for the actual job demand and the work state of the employee can be realized, the relevance, the urgency and the acceptability of the learning content are improved, the learning efficiency and the task adaptation ability are significantly enhanced, and the employee can quickly meet the job target. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present disclosure, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0011] Figure 1 A flowchart of a dynamic state-based intelligent learning path recommendation method according to an embodiment of the present disclosure is shown. Figure 2 A flowchart of a dynamic state-based intelligent learning path recommendation method according to an embodiment of the present disclosure is shown. Figure 3 A schematic block diagram of a dynamic state-based intelligent learning path recommendation system according to an embodiment of the present disclosure is shown. Figure 4 A block diagram of an exemplary electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0012] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. Note that the relative arrangement, numerical expressions, and numerical values of components and steps set forth in these embodiments do not limit the scope of the present disclosure unless otherwise specifically stated.

[0013] Those skilled in the art can understand that the terms "first", "second", and the like in the embodiments of the present disclosure are only used to distinguish different steps, devices, or modules, and do not represent any specific technical meaning, nor do they represent a necessary logical sequence between them. It should also be understood that in the embodiments of the present disclosure, "multiple" can mean two or more, and "at least one" can mean one, two, or more. It should also be understood that for any component, data, or structure mentioned in the embodiments of the present disclosure, unless explicitly limited or given a contrary implication by the context, it can generally be understood as one or more. In addition, the term "and / or" in the present disclosure is only a description of the association relationship between the associated objects, which means that there can be three relationships, for example, A and / or B can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " in the present disclosure generally represents an "or" relationship between the front and rear associated objects. It should also be understood that the description of each embodiment of the present disclosure emphasizes the differences between each embodiment, and the same or similar parts can be referred to each other, and for the sake of brevity, they will not be repeated.

[0014] It is also to be understood that the drawings are not necessarily drawn to scale, that the dimensions of the various parts can be exaggerated for the purpose of explanation, and that the technology disclosed herein is by no means limited to the specific examples of constructions as illustrated. It is also to be understood that the technology disclosed herein is not limited in its application to the details of construction and the arrangement of components set forth in the following description or illustrated in the above-described and following drawings. The technology disclosed herein is capable of other embodiments and of being practiced or of being carried out in various ways.

[0015] In order to make the objects, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be described clearly and completely below with reference to the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only some of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present disclosure.

[0016] Figure 1 A flowchart of a dynamic state-based intelligent learning path recommendation method provided by an embodiment of the present disclosure is shown. The method of the embodiment of the present disclosure aims to realize accurate detection of large goals and small goals of pictures.

[0017] In S101, a current skill improvement state, a current work load and a current task urgency value of a target employee are obtained, and a skill adaptation weight, a work load weight and an urgency weight are calculated according to the current skill improvement state, the current work load, the current task urgency value and a preset weight function calculation formula.

[0018] The target employee can be an object employee for which a learning path is currently recommended by the system.

[0019] The current skill improvement state can be a learning progress, a mastery degree or a change trend of the employee on the skills related to his / her post.

[0020] The current work load can be a task amount, a task pressure or a working hour burden of the employee in a current time period.

[0021] The current task urgency value can be a time sensitivity or a priority involved in a task currently processed by the employee.

[0022] The skill adaptation weight can be an importance adjustment coefficient of the current skill improvement state to a “skill matching degree” index in the recommended path.

[0023] The work load weight can be an importance adjustment coefficient of a current work load to a "learning intensity" index in a learning path.

[0024] The urgency weight can be an importance adjustment coefficient of a task urgency degree to a "task relevance" index in a learning path.

[0025] The system constructs a growth trend curve of each skill dimension of the target employee in the past period of time based on the learning record log, skill test result, course completion, and task feedback data of the target employee, adopts a time series modeling method (such as LSTM or exponential weighted moving average), and identifies the growth rate change of the key skills of the current post through fitting and slope analysis of the skill growth curve, so as to obtain the current skill improvement state of the target employee.

[0026] The system extracts the task assignment number, total task working hours, task priority, task state (in progress / not started), meeting arrangement, and non-planned occupied time of the employee in the current period from the employee's working hour management system, task allocation platform, and instant messaging system.

[0027] Then, the system adopts a task load estimation model (such as an analytic hierarchy process AHP plus task pressure calculation) to normalize these raw data, and calculates the current work load in the current time window in combination with the calendar working time. If the employee is currently executing multiple high-priority tasks in parallel, the load value will be automatically increased.

[0028] The training process of the task load estimation model is as follows: A large number of completed task records and employee behavior data are collected from the historical working hour system, task platform, and collaborative office platform to construct a training sample set. Each sample contains task assignment number, task estimated working hours, actual working hours, task priority level, task state, meeting and occupied time, and the actual task completion efficiency or feedback satisfaction of the employee in the time period is taken as an evaluation index of the rationality of the load. The system adopts the analytic hierarchy process (AHP) to model the importance of each input feature through subjective or expert scoring, and constructs a feature weight matrix; then, combined with the mapping relationship between different workloads and task performance in the historical samples, gradient boosting tree (GBDT) or support vector machine regression (SVR) algorithm is used to train a nonlinear model to score and predict the task pressure (i.e. the concentration of tasks per unit time). In the model training process, the weight coefficients and parameter combinations are optimized through grid search, and the consistency of the predicted task load value and the actual load performance is evaluated on the validation set. When the model converges stably, it can be deployed in the real-time system.

[0029] Then the total number of tasks responsible by the target employee is identified and obtained, a task index set is established, and each task is processed. For each task, the information such as the project to which it belongs, the type, and the priority level is extracted, and the task weight coefficient is determined in combination with the weight configuration of the task in the performance evaluation or organizational goal. Subsequently, based on the average completion time of the historical similar tasks, the task complexity level, the required skill label, etc., the predicted required time of the task is predicted through the trained machine learning model. Then, according to the difference between the deadline set for the task and the current system time, the remaining available time is calculated, a task urgency scoring function is constructed, and the above content is substituted into the function to obtain the current task urgency value of the target employee. The scoring function can adopt the following form: wherein, is the current task urgency value; m is the number of tasks; j is the task index; is the preset task weight coefficient; is the predicted required time of the task; is the remaining available time of the task.

[0030] After obtaining the current skill improvement state, the current work load, and the current task urgency value, the current skill improvement state, the current work load, and the current task urgency value are substituted into the preset weight function calculation formula to calculate the skill adaptation weight, the work load weight, and the urgency weight.

[0031] The model training process is as follows: A historical task dataset covering rich task dimensions is constructed, and the data sources include enterprise project management systems, task allocation records, and performance systems. For each historical task, key features such as task type, task complexity level, required skill label, task description keyword, and task responsible person information are extracted, and the average completion time of its historical similar tasks is calculated as one of the enhanced features. The actual completion time is used as the target variable for supervised learning. In order to improve the expression ability of the model input, category features such as task type and skill label are one-hot encoded or embedded, and NLP models such as TF-IDF or BERT are used to extract semantic vectors for task description. For numerical features such as task complexity level and average completion time, normalization or standardization processing is performed. After data preparation, the data is divided into training set and validation set, and regression algorithms such as random forest regression, XGBoost regression, or deep neural network are used to model the mapping relationship between input features and actual completion time. Cross-validation and grid search or Bayesian optimization are used to adjust the model hyperparameters during training, and mean absolute error (MAE) or root mean square error (RMSE) is used as the loss function to evaluate the model performance. When the model performs stably and the error is controllable on the validation set, the model parameters are fixed, and the model is deployed in the actual task scenario to predict the expected time required for new tasks, providing support for subsequent task scheduling and learning path recommendation.

[0032] On the basis of the above technical solutions, optionally, the preset weight function calculation formula is: Among them, is the skill adaptation weight; is the preset skill weight adjustment coefficient; is the preset skill threshold parameter; is the current skill improvement state; Among them, is the workload weight; is the preset load weight adjustment coefficient; is the current workload; Among them, is the urgency weight; is the preset urgency weight adjustment coefficient; is the current task urgency value.

[0033] In this scheme, control the "steepness" of the function, the larger the value to The more sensitive the response is, the more sensitive the system is. The initial setting can be based on expert experience. Or use historical data (such as employee skill growth curve and recommendation hit rate) to fit the parameters (for example, use least squares fitting or grid search optimization).

[0034] The baseline value of the skill state, the midpoint of the sigmoid curve. You can take the mean or median of the historical average skill improvement state value. It can also be dynamically adjusted through A / B testing to optimize the recommendation results.

[0035] is the adjustment coefficient that controls the growth rate of the workload weight function. The initial value is usually selected in the interval [0.5, 2]. You can use regression fitting of the recommendation acceptance rate under multiple groups of different loads to select β so that has good discrimination under moderate load.

[0036] influences the amplification degree of task urgency to The greater the value, the stronger the driving effect of on You can set it according to the time sensitivity of tasks in the organization, such as high-pressure positions with higher γ. Similarly, you can use machine learning (such as logistic regression) to supervise the learning optimization of historical task response effects.

[0037] S102, acquire the skill matching degree, learning intensity value and task correlation degree of each learning unit in each candidate learning path, and calculate the comprehensive score of each candidate learning path according to the skill adaptation weight, workload weight, urgency weight, skill matching degree, learning intensity value, task correlation degree and preset scoring formula.

[0038] The candidate learning path can be a set of multiple optional learning paths selected by the system for a target employee based on their job responsibilities, ability profile, learning history and current task requirements from the knowledge base or course library. These paths are usually composed of multiple learning units, each path representing a possible and relatively coherent skill improvement route for achieving the job competence or task adaptation goal.

[0039] A learning unit can refer to the basic component module in a learning path, usually corresponding to a specific course, knowledge point, task training or practice module. For example, a learning unit can be "Java programming foundation course" or "stress test operation drill", etc. It is the smallest independent learning content block in the path.

[0040] Skill match degree can refer to the degree of fit between the knowledge and skills involved in the learning unit or path and the current ability of the target employee. It reflects the pertinence of a learning unit to the current ability gap of the employee, and is usually calculated by matching the employee's ability profile with the knowledge point label. The higher the value, the more "short board" it fills.

[0041] Learning intensity value can refer to the estimated value of the time, attention load, cognitive complexity, etc. required to complete a learning unit or an entire learning path. For example, the time required to read materials + video length + exercise complexity can be combined into a learning intensity value. It reflects the degree of burden on the employee to complete the learning task.

[0042] Task relevance can refer to the direct relevance of the content in the learning unit or path to the current tasks or job responsibilities of the employee. The system can evaluate it through semantic embedding similarity, historical co-occurrence frequency, etc. between task semantic labels and learning content labels. The higher the relevance, the more "current task" the learning content fits.

[0043] The comprehensive score can refer to the total score of each candidate path calculated by the system according to the preset scoring formula, which combines the dynamic weights of multiple dimensions (such as skill adaptation weight, work load weight, task urgency weight) and the attribute indicators of the candidate path (such as skill match degree, learning intensity value, task relevance). The score is used to sort the learning paths and recommend the optimal path.

[0044] The system first extracts the complete structure information of each candidate learning path from the learning resource library, including all learning units in the path and their associated knowledge point labels, learning resource types (such as videos, documents, practical training), resource duration, operation complexity, skill improvement level, etc. Meta data. Then, for the target employee, the system uses its existing employee ability profile to extract the employee's current skill level distribution, experience, learning preferences and historical learning records, and establish a vectorized expression.

[0045] When obtaining the skill match degree, the system uses the Vector Space Model (Vector Space Model, VSM) or BERT embedding vector similarity calculation method to compare the similarity between the knowledge point labels associated with the learning unit and the skill labels of the employee's ability profile. Usually calculate the cosine similarity or dot product similarity. The higher the similarity, the more suitable the learning unit is for the employee's current ability improvement direction. The matching degrees of multiple learning units can be weighted and averaged to obtain the overall skill match degree of the path.

[0046] When obtaining the learning intensity value, the system extracts the resource duration, content density, cognitive complexity classification (such as Bloom cognitive dimensions), and other characteristics of the learning unit based on the resource type and content structure of the learning unit, and uses a learning cost model to evaluate the intensity. For example, the video duration is calculated for video resources combined with the keyword density of the subtitles; for document type resources, the text word count and term density are counted, and adjusted combined with past learner feedback (such as completion time and dropout rate). The system can use regression models (such as XGBoost, SVR) to fit and predict the intensity.

[0047] When obtaining the task relevance, the system extracts the task semantic label, key operation point and task target description of the current task of the employee, and performs semantic embedding of the task description and the knowledge point label of the learning unit through pre-trained language models such as BERT and RoBERTa. Calculate the semantic vector similarity of the two, and at the same time, combine the co-occurrence statistics of historical tasks and learning units to evaluate the relevance of each learning unit to the current task using collaborative filtering or co-occurrence matrix construction. Finally, the task relevance at the path level is obtained by weighted average.

[0048] Then, the skill adaptation weight, the workload weight, the urgency weight, the skill matching degree, the learning intensity value, and the task relevance are substituted into the preset scoring formula to calculate the comprehensive score of each candidate learning path.

[0049] On the basis of the above technical solutions, optionally, the preset scoring formula is: wherein, is the comprehensive score of each candidate learning path; n is the total number of learning units of each candidate learning path; i is the index of the learning unit of each candidate learning path; is the skill matching degree of the learning unit of each candidate learning path; is the skill adaptation weight; is the workload weight; is the learning intensity value of each learning unit in each candidate learning path; is the urgency weight; is the task relevance of each learning unit in each candidate learning path.

[0050] S103, rank each candidate learning path according to the comprehensive score to determine the recommended learning path set of the target employee, and send the recommended learning path set to the terminal device of the target employee.

[0051] The recommended learning path set can refer to a combination of several optimal learning paths selected after ranking according to the current state (such as skills, tasks, and workload) of the target employee and the comprehensive score of each candidate path.

[0052] The terminal device can refer to a hardware device used by the target employee to receive and display the learning path recommendation information.

[0053] The scoring data table can be processed using the Python-based pandas data processing framework, sorted by the score in descending order using.sort_values(by='score', ascending=False), and the top N paths with the highest scores are retained as the recommended results. Here, N can be set by system policy, such as "recommend Top5 paths" or "all paths with a comprehensive score greater than a certain threshold".

[0054] Subsequently, the path reordering strategy function is used for final screening optimization. For example, when multiple path scores are close, an exploratory parameter (such as the epsilon-greedy strategy in the Bayesian multi-arm tiger machine algorithm) can be introduced to recommend a small number of diverse paths to prevent falling into a local optimum. This process can be achieved through sklearn.linear_model in Scikit-learn combined with numpy.random random sampling strategy.

[0055] After sorting, the system organizes the top-ranked path set into a structured JSON format recommendation learning path set, which includes the unique identification of each path, the composition of learning units, the knowledge points of each learning unit, the estimated completion time, resource links, and recommended reason explanations.

[0056] Then, the recommendation set will be encapsulated and distributed through the API interface calling mechanism of the backend. The system calls the enterprise's message push service (such as based on Apache Kafka, RabbitMQ, or directly calls the enterprise WeChat, DingTalk open platform API) to push the recommended learning path set to the target employee's terminal device.

[0057] In the embodiments of the present application, the current skill improvement state, the current work load and the current task urgency value of the target employee are obtained, the skill adaptation weight, the work load weight and the urgency weight are calculated according to the current skill improvement state, the current work load, the current task urgency value and a preset weight function calculation formula, the skill matching degree, the learning intensity value and the task correlation degree of each learning unit in each candidate learning path are obtained, the comprehensive score of each candidate learning path is calculated according to the skill adaptation weight, the work load weight, the urgency weight, the skill matching degree, the learning intensity value, the task correlation degree and a preset scoring formula, each candidate learning path is sorted according to the comprehensive score, the recommended learning path set of the target employee is determined, and the recommended learning path set is sent to the terminal device of the target employee. Through the above intelligent learning path recommendation method based on dynamic state, the current skill improvement state, the work load and the task urgency of the employee are dynamically obtained, the skill matching degree, the learning intensity and the task correlation of each learning path are combined, the comprehensive score is intelligently calculated based on the preset scoring formula, and the sorting recommendation is performed, the personalized learning path pushing for the actual post demand and the work state of the employee can be realized, the relevance, the urgency and the acceptability of the learning content are improved, and thus the learning efficiency and the task adaptation ability are significantly enhanced, and the employee is helped to quickly meet the post target.

[0058] On the basis of the above technical solutions, optionally, after the recommended learning path set is sent to the terminal device of the target employee, the method further comprises: If the current task urgency value exceeds a preset task urgency threshold, the correlation degree data of each candidate learning path in the recommended learning path set and the current task of the target employee is obtained; The candidate learning path corresponding to the correlation degree data exceeding the preset correlation threshold is taken as a target learning path, the task deadline corresponding to the target learning path and the available learning resource information are obtained, and the preset learning upper limit load data of the target employee is obtained; Based on the task deadline, the available learning resource information and the preset learning upper limit load data, the content of the target learning path is reconstructed to generate a compressed learning path, and the compressed learning path is integrated into the recommended learning path set.

[0059] In the present solution, the preset task urgency threshold can be a task urgency judgment standard value set in advance by the system according to the task management strategy and the experience rule.

[0060] The current task can be a set of work tasks that the target employee is currently performing or will soon process in a current time window. These tasks can include multiple concurrent projects, have different priorities, deadlines, skill requirements and other information, and are usually recorded by task management systems, project platforms, scheduling tools and the like.

[0061] The correlation data can represent the degree of semantic / content / target relevance between each candidate learning path and the current task.

[0062] The preset correlation threshold can represent the demarcation value of whether the system-set learning path is highly relevant to the current task.

[0063] The target learning path can refer to the path selected from the candidate paths, which is highly correlated (exceeding the set correlation threshold) with the current urgent task, and subsequent compression, reconstruction, replacement, etc. It is the core path unit that needs to be adjusted and optimized.

[0064] The task deadline can be the remaining completion time of the key task served by the target learning path, usually represented as the system-set end time of the task minus the current time, used to determine whether there is time to complete the original path.

[0065] The available learning resource information can refer to the collection of micro-lessons, videos, text materials, and answer resources in the current system that can be allocated for content reconstruction. Each resource should have metadata (such as duration, skill point coverage, recommended load, etc.) and a structural mapping with knowledge points.

[0066] The preset learning upper limit load data can refer to the maximum learning intensity or cognitive load limit that the target employee can bear per unit time within the current time period, taking into account factors such as task pressure, daily available time, and historical learning tolerance.

[0067] The compressed learning path can refer to the minimum set of key knowledge units required for the task extracted from the target learning path through structured pruning and intelligent reconstruction, which shortens the total learning time, reduces the learning intensity, and improves the efficiency of the lightweight learning path without reducing the task adaptability.

[0068] If the system detects that the current task urgency value of the employee exceeds the preset task urgency threshold set in the system, based on the task list currently being processed by the employee, the knowledge point tag matching algorithm is used to analyze the correlation between each candidate path in the recommended learning path set and the current task, and construct the semantic correlation data of each path and the task. This analysis combines three pieces of information: one is the knowledge point metadata covered by the candidate path (provided by the learning resource tag system), the second is the ability item and skill point information corresponding to the task requirements (derived from the enterprise internal or external knowledge graph), and the third is the semantic text embedding vector of the task target and path content (calculated by BERT, SBERT, etc. Language model to calculate similarity). Through the above information fusion, the correlation score between the path and the task is calculated, and compared with the preset correlation threshold defined in the system, and all candidate learning paths with correlation scores exceeding the threshold are selected as the target learning path.

[0069] Subsequently, the system performs task-driven compression reconstruction on each target learning path. First, the task deadline corresponding to the task served by the path is obtained, i.e., the preset completion time of the task minus the current system time, and it is used as the learning resource allocation time limit; second, all available learning resource information related to the target path in the current system is obtained, including video, text and picture courseware, micro-lecture, test and other content units, and the metadata of these resources such as knowledge point annotation, expected learning time, cognitive load label and usage state are extracted; in addition, the learning upper limit load data of the target employee in the current time period, i.e., the maximum tolerable learning time or cognitive load value, is also obtained, which is usually estimated by the system in combination with the task pressure, work calendar, historical learning behavior and other factors of the employee.

[0070] After obtaining the above constraints, the system first constructs a learning unit dependency graph of the target learning path, i.e., mapping the knowledge points in the path into a graph structure, with nodes as knowledge points and edges as dependency relationships. Then, through graph traversal and reverse positioning of task-related knowledge points, the set of core knowledge nodes related to the task is determined. Next, the first candidate learning resource set covering these core knowledge nodes is selected from the available learning resource information, and whether the total learning time is less than the remaining time limit of the task deadline is calculated to exclude unfeasible resource combinations, and the second candidate learning resource set that meets the time constraint is obtained.

[0071] On this basis, the system evaluates the cognitive load and learning intensity distribution of the second candidate resource, compares it with the learning upper limit load data of the employee, and finally determines the target learning resource set that meets all the constraints. Finally, according to its structural position in the dependency graph, the learning path is reorganized to construct a compressed learning path covering the core knowledge points of the task. The path structure is simple, time controllable, and load adaptive, which can effectively improve the learning adaptability in high task pressure scenarios.

[0072] Finally, the system inserts or replaces the generated compressed learning path into the original recommended learning path set to form a new recommended set and performs real-time synchronization update.

[0073] In this scheme, by combining the task urgency, time limit, learning resource availability and employee learning load, intelligent compression and reconstruction of the recommended learning path are realized, which not only improves the adaptability of the learning content to the current task, but also significantly improves the learning efficiency and resource utilization rate in high pressure scenarios, ensuring that the employee masters the key knowledge points in a limited time and helps to complete the task with high quality.

[0074] On the basis of the above technical scheme, based on the task deadline, available learning resource information and preset learning upper limit load data, the target learning path is subjected to content reconstruction processing to generate a compressed learning path, including: obtain knowledge point dependency data corresponding to each learning unit in the target learning path, and construct a learning unit dependency graph according to the knowledge point dependency data; determine a core knowledge node in the learning unit dependency graph according to the learning unit dependency graph and the association degree data of the target learning path and the current task of the target employee; filter a first candidate learning resource from the set of available learning resources according to the learning unit dependency graph and the core knowledge node; calculate the cumulative learning time of the first candidate learning resource according to the task deadline, and take the first candidate learning resource that does not exceed the remaining available time as a second candidate learning resource; comprehensively evaluate the cognitive load and learning intensity of the second candidate learning resource according to the preset learning upper limit load data, and determine a target learning resource; determine the topological relationship between the target learning resource and the learning unit dependency graph, reorganize and compress the path according to the target learning resource and the topological relationship, and generate a compressed learning path.

[0075] In this scheme, the knowledge point dependency data can refer to data representing the logical order, the order of mastering relationship between different learning units or knowledge points. For example, "A skill needs to be learned before B skill" is a kind of dependency relationship. This data can come from knowledge graph, course design structure, expert annotation rules, etc., and is used to construct the logical skeleton of the learning path.

[0076] The learning unit dependency graph can be a directed graph constructed with learning units as nodes and knowledge point dependency data as edges, representing the order relationship between learning units. Through this graph, it can be determined which content is the basic knowledge and which is the advanced content, which is helpful for retaining the key path in subsequent path pruning.

[0077] The core knowledge node can refer to a knowledge point that is strongly associated with the current task and is in a key position in the dependency chain in the learning unit dependency graph. The system identifies the core part of the learning path that cannot be missed by analyzing the task relevance score of these knowledge points and their connection degree, in-degree / out-degree and other indicators in the graph.

[0078] The first candidate learning resource can refer to a content resource set that meets the requirements of covering the core knowledge node and adapting to the current learning state of the employee in all available learning resource sets, which can include video courseware, graphic materials, interactive exercises, etc.

[0079] The cumulative learning time can be the cumulative sum of the estimated learning time of each resource in the first candidate learning resource, representing the approximate time required to complete the batch of resources.

[0080] The remaining available time can refer to the length of time from the current time to the target task deadline, minus the scheduled work tasks, meetings, and other unavailable time of the employee, and the time resource that can be allocated to learning activities evaluated by the system.

[0081] The second candidate learning resource can refer to a subset of the first candidate learning resource, in which the cumulative learning duration does not exceed the remaining available time. This resource set ensures that the learning content can be completed before the task deadline.

[0082] Cognitive load represents the degree of occupation of working memory during the learning process, and learning intensity refers to the information density and difficulty per unit time. The system can obtain a cognitive intensity model by analyzing physiological sensor data (such as eye movement, electroencephalogram) or learning behavior modeling (such as dwell time, error rate) to quantify these two dimensions.

[0083] The target learning resource can refer to the final resource set used for learning content reconstruction after further evaluation of cognitive load and learning intensity in the second candidate learning resource, which is most suitable for employees to complete efficient learning under the current load.

[0084] The topological relationship can refer to the structural position and dependency path layout of the target learning resource in the original learning unit dependency graph. When the path is reorganized, the topological structure is referred to, the necessary dependency order is preserved, and secondary or redundant content is compressed to ensure that the compressed learning path still has logical coherence and knowledge coverage.

[0085] In the process of executing the compressed learning path generation, the system first needs to obtain the knowledge point dependency relationship data corresponding to each learning unit in the target learning path. This dependency data can be obtained by parsing course design metadata, teaching syllabus, expert-defined knowledge graph, or extracting concept reference relationships in course text descriptions based on natural language processing techniques (such as BERT, TextRank) to obtain annotations such as "predecessor dependency" and "containment relationship" between knowledge points. For example, "current calculation" depends on "Ohm's law", and the system models this relationship as a directed edge.

[0086] Subsequently, the system constructs a learning unit dependency graph based on the above knowledge point dependency relationship data, with learning units as nodes and dependency relationships as directed edges forming a directed graph structure. Graph construction can use graph structure tool libraries such as NetworkX to calculate the topological position, connectivity, path length, etc. of each learning unit in the graph, which is used for subsequent path pruning and evaluation.

[0087] To identify the key content in the graph, the system combines the learning unit dependency graph with the target learning path and the current task of each unit and target employee to identify core knowledge nodes with high task relevance and located in the key path of the graph (such as high PageRank value, high in-degree or located on the longest dependency path). The task relevance can be obtained by analyzing the semantic similarity between the task text and the learning unit description through a semantic matching model (such as Sentence-BERT).

[0088] After identifying the core nodes, the system filters out resources that can effectively cover these core knowledge nodes from the currently accessible available learning resource set to obtain the first candidate learning resource. Resource filtering can be based on label matching (such as skill label and knowledge point label), content abstract matching, etc. to ensure that the resource content meets the learning needs of the core knowledge points.

[0089] Then, the system calculates the cumulative learning time of these first candidate learning resources. The time can come from resource metadata (such as video length, text estimated reading time) or average time consumption data from historical user learning behavior. The system then calculates the remaining available time of the target task (from the current time to the task deadline minus the employee's scheduled non-disposable period), and defines the resource set whose learning time does not exceed the remaining time limit as the second candidate learning resource.

[0090] On this basis, the system combines the employee's preset learning upper limit load data (such as the maximum learnable time per day, cognitive load threshold) to evaluate the cognitive load and learning intensity of the second candidate learning resource. Cognitive load can be evaluated through resource structure complexity, term density, average dwell time prediction model, etc. Learning intensity is based on information density per unit time, number of exercises, etc. The system uses a multi-index weighted scoring model or clustering analysis method (such as K-means) to filter out the target learning resource set that best fits the current state of the employee.

[0091] Finally, the system needs to determine the topological relationship of the target learning resource in the learning unit dependency graph, that is, to determine the relative position of the learning unit corresponding to these resources in the dependency graph. By analyzing its upstream and downstream dependency relationships, dependency chain breakpoints, path jump conditions, the system performs path reorganization and compression, retains key dependency structures, and omits low-weight or skippable content nodes, and finally outputs a compressed learning path that meets the task urgency, employee load and time limit constraints as a replacement or supplement to the current recommended path.

[0092] In this scheme, by constructing a learning unit dependency graph, combining task urgency, learning resource availability and employee learning load, accurately identifying key knowledge points and generating a compressed learning path, efficient, task-oriented learning recommendation can be achieved under limited time and load conditions, improving learning adaptability and completion rate in emergency scenarios.

[0093] Figure 2 A flowchart of a method for recommending an intelligent learning path based on a dynamic state is provided for an embodiment of the present disclosure. The method can include the following steps: S201, obtaining a current skill improvement state, a current workload, and a current task urgency value of a target employee, and calculating a skill adaptation weight, a workload weight, and an urgency weight according to the current skill improvement state, the current workload, the current task urgency value, and a preset weight function calculation formula.

[0094] S202, obtaining a skill matching degree, a learning intensity value, and a task association degree of each learning unit in each candidate learning path, and calculating a comprehensive score of each candidate learning path according to the skill adaptation weight, the workload weight, the urgency weight, the skill matching degree, the learning intensity value, the task association degree, and a preset scoring formula.

[0095] S203, sorting each candidate learning path according to the comprehensive score, determining a recommended learning path set of the target employee, and sending the recommended learning path set to a terminal device of the target employee.

[0096] S204, if a new employee is identified, obtaining post information and personal history data of the new employee, constructing a post capability demand graph based on the post information, and constructing an employee capability portrait based on the personal history data.

[0097] The new employee can be an employee user who enters the enterprise system for the first time or enters a new post for the first time due to post adjustment, and usually lacks historical learning behavior data, which has a cold start problem.

[0098] The post information can represent the structured definition data of the current post of the employee. It can include post number (such as DEV-03), post name (such as senior backend development engineer), department to which it belongs, post level, post responsibility description, key skill requirement, general ability requirement, etc.

[0099] The personal history data can refer to the information of the employee's past education background, work experience, project experience, certificate qualification, and training record, etc., which is used to depict the initial ability basis.

[0100] The post capability demand graph can refer to a knowledge graph constructed according to the post information, which is centered on post responsibilities, has capability items as nodes, and has capability relationships as edges, and describes the combination, dependency, and hierarchical relationship of each capability required by the post.

[0101] The employee capability portrait can refer to a structured description reflecting the current capability characteristics of the employee constructed based on the employee history data, which is usually represented in the form of a vector or a label set.

[0102] When the system identifies a new employee (i.e., an employee without historical learning records under the current position), the system first obtains the position information and personal resume data of the employee from the enterprise human resource system. The position information includes the position name, department, position level, and responsibility text description, and the resume data includes education, work experience, project experience, training records, certificates, etc. The system extracts the core skills and knowledge points required by the position based on the pre-set position-ability matching rule library (for example: a position needs to master "Python programming, data analysis, communication and coordination" ability items) and industry standards (such as national occupation standards, enterprise job description, skill dictionary table) using keyword matching and rule template methods, and forms a position ability list.

[0103] According to the artificially set prerequisite relationship and logical dependence between skills (for example, "need to master Excel before learning data analysis"), a structured position ability requirement graph is constructed, which takes ability as a node and dependence relationship as an edge, reflecting the logical requirement structure of the position for the ability.

[0104] For personal resumes, the system uses a keyword matching algorithm (such as dictionary-based scanning and classification) to map certificate names, project keywords, job title labels, etc. in the resume to a pre-defined set of ability items. For example: "participate in the company's annual financial budget preparation" → match "financial analysis ability", "obtain PMP certificate" → match "project management ability". Through this mapping method, the system constructs the set of ability items that the employee currently possesses, and according to the experience years or the task complexity described in the resume, assigns an experience level label (such as "master", "skilled", "expert") to each ability, thereby completing the construction of the employee's ability portrait.

[0105] S205, performing ability difference analysis on the position ability requirement graph and the employee's ability portrait to obtain an ability difference graph.

[0106] The ability difference graph can refer to a structured graph data generated by comparing the position ability requirement graph and the employee's ability portrait, used to describe the key ability items and their dependence relationships that the employee currently lacks or lacks. It is the core intermediate result in personalized learning path recommendation, reflecting the skill points that the employee needs to improve and the priority learning order.

[0107] All the ability items from the job capability demand map and the employee capability profile can be standardized, including synonym normalization of the ability items using a skill library; mapping the ability names in different data sources to a unified skill label (e.g., "Excel application" and "Office table processing" are combined into "Excel operation") using entity alignment technology; and encoding all the ability items into vectors (e.g., using word embedding algorithms such as Word2Vec or BERT-based Embedding) for subsequent similarity calculation.

[0108] For the standardized skill labels, the ability nodes in the job capability demand map and the employee capability profile are compared to calculate the ability coverage relationship between them, mainly including: for each job capability node, finding whether there is a same-name or similar ability item (vector comparison based on cosine similarity or Euclidean distance) in the employee profile; if there is, comparing the ability levels (such as "master", "skilled", "understand", etc.); if the employee's ability level is lower than the job requirement level, it is recorded as a "capability difference node"; if there is no matching ability item, it is recorded as a "capability missing node".

[0109] All the ability nodes with differences and their dependency relationships are sorted into a graph structure to form a capability difference map: ability nodes: composed of the ability items that exist in the job but are not possessed or have insufficient levels by the employee; Node attributes: including ability level difference (such as "demand: master / current: understand, gap level = 2"), urgency (importance score of the ability in the job); Graph edge relationship: the ability dependency edge from the original job capability demand map (e.g., "master A → can learn B"), used to guide the subsequent learning path construction; If the capability graph is a directed acyclic graph (DAG), the topological sorting algorithm can be used to determine the learning priority.

[0110] S206, input the capability difference map into a preset learning path prediction model to generate an initial recommended learning path set adapted to the job information.

[0111] The preset learning path prediction model can be an artificial intelligence model for generating personalized learning paths based on the capability difference map, which mainly aims to automatically recommend one or more most adaptable learning paths according to the employee's current capability short board, job requirements and graph structure. The model can be implemented based on the following technical architecture: graph neural network: taking the capability difference map as the input graph structure (nodes are skill gaps, edges are dependency relationships), predicting the learning priority and path order of each node. Use mainstream structures such as GCN, GAT or GraphSAGE.

[0112] Sequence decision model (e.g., reinforcement learning): build the recommended path as a Markov decision process, considering factors such as the shortest path, learning cost, cognitive load, etc.; use DeepQ-Network or Policy Gradient to generate the optimal path.

[0113] Path generation model (e.g., Transformer-based Seq2Seq): encode the ability difference vector and job requirements into input sequences; output a sequence of learning resources (path) for end-to-end path prediction.

[0114] Rule fusion mechanism (as a supplement): in the cold start phase or low data context, combine the job standard learning path template and knowledge graph dependency rules to synthesize the path.

[0115] The initial recommended learning path set can refer to the learning path set output by the system based on the above models, which is adapted to the employee's job, minimizes the ability difference, and has strong learning resource availability. It usually has the following characteristics: contains multiple path candidates, each path is an ordered sequence of learning units (e.g., videos, documents, online courses, simulation exercises, etc.); each path covers the key nodes of the employee's to-be-supplemented abilities; path length, learning resource type, intensity, etc. are constrained by job pace and employee learning load; employees can choose independently or be pushed by managers.

[0116] The above ability difference graph can be input into the trained learning path prediction model. The model uses a graph neural network (GNN) architecture such as GraphSAGE or GAT (graph attention network) to identify the optimal learning path combination by aggregating the semantic features and structural adjacency information of the nodes. When training the model, the learning behavior sequences of historical employees in different positions, skill achievement data, and performance data are used as supervision signals, so that the model has the ability to "predict whether the path is effective".

[0117] After the model encodes the ability difference graph, it outputs one or more optimal path candidates. Each path represents an ordered combination of learning units for filling in missing skills. To fill in the actual content of these paths, the system matches learning resources based on each target skill point in the enterprise knowledge base and third-party learning platforms, uses semantic retrieval and content scoring mechanisms (such as TF-IDF vector space model, BERT embedding similarity) to filter suitable learning units, and calculates their learning duration, cognitive load, and resource availability.

[0118] Finally, the system scores and ranks the learning paths based on multiple indicators such as completeness, estimated completion time, and resource matching, forms an initial set of recommended learning paths with clear structure and reasonable order, and displays this set when the employee first logs in to the learning platform or during the onboarding process.

[0119] The training process of the preset learning path prediction model is as follows: First, a set of high-quality training samples is prepared. The training data set is composed of the learning path recommendation records of historical employees during job change, new employment or initial job transfer. Each training sample contains the input ability difference graph and the output initial recommended learning path set, which are used as the input features and supervised labels of the model respectively.

[0120] The ability difference graph is derived from the difference between the historical employee's job capability demand graph and the employee's ability portrait. The graph is represented in a graph structure, with nodes representing the ability points or skill labels to be improved, and edges representing the pre-post dependency relationship, hierarchical structure or theme association. The system encodes the node features (such as skill level, scarcity, historical coverage) and structural features (such as node degree, hierarchical depth, position in topology, etc.) in the graph through graph embedding, and uses graph representation learning algorithms such as Node2Vec and GraphSAGE to convert the graph structure into a fixed-length vector representation.

[0121] The label data, i.e. the historical initial recommended learning path set, is usually the learning path records recommended by the enterprise learning platform during the new employee onboarding phase and executed. These paths are composed of learning units, with a sequence, and each learning unit is associated with a specific knowledge point or skill label. In order to enhance the expression ability of the supervision information, the label not only includes the path itself, but also the completion rate, average score, required time, learning order, etc. of each path, which is used as the optimization target for model learning.

[0122] In terms of model structure, Graph Neural Network (GNN) is used as the main architecture, such as Graph Attention Network (GAT) or Graph Convolutional Network (GCN), to encode the ability difference graph. The model encodes the output latent skill improvement trajectory representation, and then introduces a sequence decoder (such as Decoder based on Transformer structure or Pointer Network) to gradually generate the optimal learning unit ordering path from the skill trajectory vector.

[0123] The entire training process is conducted using a supervised learning framework, and the loss function combines path structure matching loss (such as cross-entropy loss based on sequence similarity) and path effect indicators (such as regression loss of path completion rate), ensuring that the model not only learns "what similar employees should learn", but also optimizes "how effective is it after learning".

[0124] To enhance generalization ability, data augmentation methods are also introduced to perturb the graph nodes and reasonably transform the path order to construct diverse training samples. At the same time, to improve the adaptability to cold-start positions, the model is pre-trained to introduce a general ability difference graph in the industry knowledge graph as auxiliary input.

[0125] After training, the model can accept the ability difference graph calculated from the position capability requirement graph and the employee capability portrait during the new employee's probation period, generate an initial recommended learning path set that adapts to the position, and thus realize intelligent customization recommendation on the basis of zero experience.

[0126] In this embodiment, in the cold start stage of new employees lacking historical behavior data, based on the position requirements and individual ability differences, an initial recommended learning path with high position matching degree is quickly generated, which significantly improves the efficiency of job adaptation and learning accuracy, reduces the waste of training resources, and improves the quality of talent growth.

[0127] On the basis of the above technical solutions, after generating the initial recommended learning path set adapted to the position information, the method further includes: Obtaining an industry knowledge graph, a general skill library, and professional standard specification data corresponding to the position information, fusing the industry knowledge graph, the general skill library, and the professional standard specification data with the position capability requirement graph to obtain an enhanced position capability requirement graph; Based on the enhanced position capability requirement graph and the employee capability portrait, re-performing ability difference analysis to generate an ability difference enhanced graph; Inputting the ability difference enhanced graph into a preset learning path prediction model to update the initial recommended learning path set.

[0128] In this solution, the industry knowledge graph can refer to a structured data network covering professional knowledge, skills, and their mutual relationships in a specific industry field, usually containing industry standards, job responsibilities, core skills, and technical trends, etc. information, used to enrich the external reference of position capability requirements.

[0129] The general skill library can refer to a collection of basic skills and ability keywords widely applicable to multiple positions and industries, such as communication skills, project management, team collaboration, etc., which can assist in supplementing the skill dimensions in the position capability requirement graph.

[0130] Professional standard specification data can refer to professional qualification standards and regulatory documents published by governments, industry associations or authoritative institutions, which specify the required ability level and professional requirements for a specific position, and are used as authoritative basis for regulating job ability requirements.

[0131] The enhanced job ability requirement graph can be a more comprehensive, more generalizable and more authoritative job ability requirement model constructed on the basis of the original enterprise internal job ability requirement graph, by integrating industry knowledge graph, general skill library and professional standard specification data.

[0132] The ability difference enhanced graph can be structured data obtained by difference analysis based on the enhanced job ability requirement graph and employee ability portrait, reflecting the gap between the current ability of the employee and the job requirement after enhancement, with higher accuracy and coverage, guiding more accurate learning path recommendation.

[0133] According to the job information of the target position, the system accesses and extracts the corresponding industry knowledge graph, general skill library and professional standard specification data. Among them, the industry knowledge graph is constructed by crawling and analyzing authoritative industry websites, standard issuing agencies and professional literature, using natural language processing technology (NLP) to perform entity recognition and relationship extraction on text data, and constructing a knowledge graph of job responsibilities, key skills and their associated relationships within the industry; the general skill library extracts structured skill tags from public professional skill libraries and cross-industry general ability standards, and combines word vector models (such as Word2Vec or BERT) for semantic similarity calculation and word meaning aggregation to ensure coverage of general skills required by the position; the professional standard specification data is obtained by text analysis and structured processing of government-issued professional qualification standard documents to form a standardized set of ability requirements.

[0134] Subsequently, the system uses graph database (such as Neo4j) and knowledge graph fusion algorithm based on graph fusion technology to perform structural mapping and semantic alignment between the above external data and the existing job ability requirement graph within the enterprise, solving the problems of entity duplication, attribute conflict and relationship consistency between heterogeneous data, and finally generating an enhanced job ability requirement graph with higher coverage and authority.

[0135] Next, the system compares and analyzes the enhanced job ability requirement graph with the corresponding employee's ability portrait. By constructing vector representations of the two, the system uses graph embedding technology (such as GraphSAGE, Node2Vec) to capture the semantic features of nodes and relationships, and then uses difference calculation algorithms (such as cosine similarity or Euclidean distance) to quantify the difference between employee ability and job requirement, forming a structured ability difference enhanced graph that reflects the ability gap and strengths and weaknesses.

[0136] Finally, the system inputs the enhanced capability difference graph into a preset learning path prediction model. The model is based on a deep neural network combined with a graph neural network architecture, comprehensively analyzes the difference characteristics of job requirements and employee capabilities, dynamically adjusts and optimizes the learning path recommendation strategy, and generates an updated initial recommended learning path set that is highly adapted to the job information.

[0137] In this scheme, by fusing multi-source authoritative data such as industry knowledge graph, general skill library, and professional standard specification, an enhanced job capability requirement graph is constructed, on the basis of which a more comprehensive and accurate capability difference enhanced graph is generated, and the learning path recommendation result is optimized accordingly, which can effectively improve the coverage, accuracy, and job adaptation degree of the initial recommended learning path, especially suitable for rapid customization and adjustment of new employee learning paths in complex job requirements or cross-industry transfer scenarios, significantly improving the personalized learning effect and job competence efficiency.

[0138] Figure 3 An intelligent learning path recommendation system based on dynamic state is provided for the embodiments of the present disclosure. The system comprises: The state evaluation module 301 is configured to obtain the current skill improvement state, current work load, and current task urgency value of the target employee, and calculate the skill adaptation weight, work load weight, and urgency weight according to the current skill improvement state, current work load, and current task urgency value and a preset weight function calculation formula. The path evaluation module 302 is configured to obtain the skill matching degree, learning intensity value, and task association degree of each learning unit in each candidate learning path, and calculate the comprehensive score of each candidate learning path according to the skill adaptation weight, work load weight, urgency weight, skill matching degree, learning intensity value, task association degree, and a preset scoring formula. The recommendation pushing module 303 is configured to sort each candidate learning path according to the comprehensive score, determine the recommended learning path set of the target employee, and send the recommended learning path set to the terminal device of the target employee.

[0139] As shown in Figure 4 The electronic device 400 provided by the embodiments of the present disclosure includes a processor 401, a memory 402, and a program or instruction stored in the memory 402 and executable on the processor 401. The program or instruction is executed by the processor 401 to implement each process of the encryption protection method of the deep learning large model, and achieves the same technical effect. To avoid repetition, details are not repeated here.

[0140] It should be noted that the electronic device in the embodiments of the present disclosure includes the mobile electronic device and the non-mobile electronic device described above.

[0141] The embodiment of the application further provides a readable storage medium, which stores a program or instructions, and the program or instructions are executed by a processor to realize the processes of the cable installation process based on tension self-adaptive control system embodiment and achieve the same technical effects. To avoid repetition, details are not described herein.

[0142] The processor is the processor in the electronic device in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0143] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or system. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article or system including the element. In addition, it should be pointed out that the scope of the methods and systems in the embodiments of the application is not limited to the order of performing the functions shown or discussed, but can also include performing the functions in a substantially simultaneous manner or in reverse order according to the functions involved.

[0144] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment method can be realized by means of software and necessary general hardware platform, of course, it can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes a plurality of instructions for making a terminal (which can be a mobile phone, computer, server or network device) execute the method described in the embodiments of the present application.

[0145] The embodiments of the application are described above in combination with the drawings, but the application is not limited to the above specific embodiments, which are only illustrative and not restrictive. Those skilled in the art can make many forms under the inspiration of the application without departing from the scope of the application and the protection scope of the claims.

[0146] The above merely describes the preferred embodiments of the present application and the technical principles applied. The present application is not limited to the specific embodiments described herein, and various obvious changes, modifications and replacements made by those skilled in the art without departing from the scope of the present application shall not be excluded. Therefore, although the present application is described in more detail through the above embodiments, the present application is not limited to the above embodiments, and more other equivalent embodiments can be included without departing from the concept of the present application, and the scope of the present application is determined by the scope of the claims.

Claims

1. A dynamic state-based intelligent learning path recommendation method, characterized by, The method comprises: obtaining the current skill improvement state, the current workload and the current task urgency value of the target employee, calculating the skill adaptation weight, the workload weight and the urgency weight according to the current skill improvement state, the current workload, the current task urgency value and a preset weight function calculation formula; obtaining the skill matching degree, the learning intensity value and the task correlation degree of each learning unit in each candidate learning path, calculating the comprehensive score of each candidate learning path according to the skill adaptation weight, the workload weight, the urgency weight, the skill matching degree, the learning intensity value, the task correlation degree and a preset scoring formula; sorting each candidate learning path according to the comprehensive score to determine the recommended learning path set of the target employee, and sending the recommended learning path set to the terminal device of the target employee.

2. The method of claim 1, wherein, Wherein, after sending the recommended learning path set to the terminal device of the target employee, the method further comprises: if the current task urgency value exceeds a preset task urgency threshold, obtaining the correlation degree data of each candidate learning path in the recommended learning path set and the current task of the target employee; regarding the candidate learning path corresponding to the correlation degree data exceeding the preset correlation threshold as the target learning path, obtaining the task deadline and available learning resource information corresponding to the target learning path, and obtaining the preset learning upper limit load data of the target employee; based on the task deadline, the available learning resource information and the preset learning upper limit load data, performing content reconstruction processing on the target learning path to generate a compressed learning path, and integrating the compressed learning path into the recommended learning path set.

3. The method of claim 2, wherein, Wherein, based on the task deadline, the available learning resource information and the preset learning upper limit load data, performing content reconstruction processing on the target learning path to generate a compressed learning path, comprising: obtaining the knowledge point dependency relationship data corresponding to each learning unit in the target learning path, and constructing a learning unit dependency graph according to the knowledge point dependency relationship data; determining the core knowledge node in the learning unit dependency graph according to the learning unit dependency graph and the correlation degree data of the target learning path and the current task of the target employee; according to the learning unit dependency graph and the core knowledge node, filtering out first candidate learning resources from the available learning resource set; calculating the cumulative learning time of the first candidate learning resource according to the task deadline, and regarding the first candidate learning resource that does not exceed the remaining available time as a second candidate learning resource; comprehensively evaluating the cognitive load and learning intensity of the second candidate learning resource according to the preset learning upper limit load data to determine the target learning resource; determining the topological relationship between the target learning resource and the learning unit dependency graph, and performing path reorganization and compression according to the target learning resource and the topological relationship to generate a compressed learning path.

4. The method of claim 1, wherein, Wherein, the preset weight function calculation formula is: wherein, is a skill adaptation weight; is a preset skill weight adjustment coefficient; is a preset skill threshold parameter; is a current skill boost status; wherein, is a workload weight; is a preset load weight adjustment coefficient; is a current workload; wherein, is an urgency weight; is a preset urgency weight adjustment coefficient; is a current task urgency value.

5. The method of claim 1, wherein, wherein, the preset scoring formula is: wherein, is a comprehensive score of each candidate learning path; n is the total number of learning units of each candidate learning path; i is an index of the learning units of each candidate learning path; is a skill matching degree of the learning units of each candidate learning path; is a skill adaptation weight; is a workload weight; is a learning intensity value of each learning unit in each candidate learning path; is an urgency weight; is a task relevance degree of each learning unit in each candidate learning path.

6. The method of claim 1, wherein, wherein, after sending the recommended learning path set to the terminal device of the target employee, the method further comprises: If a new employee is identified, obtain the job information and personal resume data of the new employee, construct a job capability demand graph based on the job information, and construct an employee capability portrait based on the personal resume data; Perform capability difference analysis on the job capability demand graph and the employee capability portrait to obtain a capability difference graph; Input the capability difference graph into a preset learning path prediction model to generate an initial recommended learning path set adapted to the job information.

7. The method of claim 6, wherein, Wherein, After generating the initial recommended learning path set adapted to the job information, the method further comprises: Obtain an industry knowledge graph corresponding to the job information, a general skill library, and professional standard specification data, fuse the industry knowledge graph, the general skill library, and the professional standard specification data with the job capability demand graph to obtain an enhanced job capability demand graph; Perform capability difference analysis based on the enhanced job capability demand graph and the employee capability portrait to generate a capability difference enhanced graph; Input the capability difference enhanced graph into a preset learning path prediction model to update the initial recommended learning path set.

8. A dynamic state based intelligent learning path recommendation system characterized in that, The system comprises: A state evaluation module for obtaining the current skill improvement state, current work load, and current task urgency value of a target employee, and calculating skill adaptation weight, work load weight, and urgency weight according to the current skill improvement state, current work load, current task urgency value, and a preset weight function calculation formula; A path evaluation module for obtaining the skill matching degree, learning intensity value, and task association degree of each learning unit in each candidate learning path, and calculating the comprehensive score of each candidate learning path according to the skill adaptation weight, work load weight, urgency weight, skill matching degree, learning intensity value, task association degree, and a preset scoring formula; A recommendation pushing module for sorting each candidate learning path according to the comprehensive score, determining the recommended learning path set of the target employee, and sending the recommended learning path set to the terminal device of the target employee.

9. An electronic device, comprising: A processor, a memory, and a program or instruction stored on the memory and executable on the processor, wherein the program or instruction is executed by the processor to implement the steps of the intelligent learning path recommendation method based on dynamic state according to any one of claims 1-7.

10. A readable storage medium, characterized by, The readable storage medium stores a program or instruction, and the program or instruction is executed by the processor to implement the steps of the intelligent learning path recommendation method based on dynamic state according to any one of claims 1-7.

Images