Self-learning path generation method based on adaptive multi-objective optimization locust algorithm

Abstract

This invention discloses a self-learning path generation method based on an adaptive multi-objective optimization locust algorithm, addressing the static nature of learning path planning in existing technologies. The method includes: initialization and generation of a first path population; multi-objective fitness evaluation and ranking of the first path population; leader selection and structural crossover generation of a second dynamic path population; local perturbation generation of a third fine-tuned dynamic path population; fitness evaluation and elite retention of the third fine-tuned dynamic path population; and determination of evolution termination conditions and output of the optimal path set. This approach offers advantages such as simple logic and dynamic reliability, and possesses high practical and promotional value in the field of self-learning path generation.

Description

Technical Field

[0001] The present invention relates to the technical field of self-learning path generation, in particular to a self-learning path generation method based on an adaptive multi-objective optimization locust algorithm. Background Art

[0002] Online education platforms are currently widely used in high school academic learning, with particular emphasis on personalized learning recommendations, content push, and path planning. However, mainstream learning path recommendation systems mostly rely on rule-based matching, collaborative filtering, or static tag recommendation methods, which struggle to meet students' diverse cognitive states and dynamic learning progress requirements.

[0003] For example, the public technology of "Publication No. CN118761879A, Named "Method and System for Recommending Personalized Learning Paths for Students Based on Big Data Analysis" obtains the teaching attribute data of the students, obtains the initial recommended learning path based on the teaching attribute data and the students' expectations, obtains the students' growth efficiency rate based on the evaluation scores and records it as the first contribution rate, and obtains the second contribution rate based on the evaluation scores and recommended teaching resources; calculates the average of the first contribution rate and the second contribution rate to obtain the predicted contribution rate; modifies the reference learning path based on the predicted contribution rate and the students' expectations to obtain the first modified learning path; feeds back the predicted contribution rate and the students' expectations to the training teachers and modifies them to obtain the second modified learning path and record it as the recommended learning path.

[0004] Another example is the public technology "Publication No. CN119128275A, entitled "A Method and System for Recommending Educational Resources Based on Artificial Intelligence." It first builds an educational resource engine, then generates a preliminary user profile by collecting basic user information, and dynamically updates the user profile based on the user's personalized needs. It then interacts with the educational resource engine based on a data reliability transmission method, calls educational resource data, and uses collaborative filtering and content-based recommendation to weightedly integrate the recommended content, providing users with personalized learning resources and generating learning path planning for them. During the user's learning process, the system monitors the user's learning behavior and optimizes the recommendation model based on user feedback. Another example is the public technology "Publication No. CN119988732A, entitled "A Big Data-Based Intelligent Management System for Continuing Education Courseware Resources."

[0005] The above technologies use rule-based matching, collaborative filtering or static tags for recommendation, which has the following problems: First, the learning path recommendation lacks a dynamic update mechanism and is difficult to adapt to the real-time evolution of students' cognitive status; second, the learning path is optimized for a single goal, such as "shortest path" or "minimum number of questions", ignoring multi-dimensional goals such as mastery, cognitive jumps, and interest loss; third, the model has weak generalization ability and is difficult to transfer to different subjects or students at different levels of foundation.

[0006] Therefore, it is urgent to propose a self-learning path generation method based on the adaptive multi-objective optimization locust algorithm that is simple in logic, dynamic and reliable. Summary of the Invention

[0007] In view of the above problems, the object of the present invention is to provide a self-learning path generation method based on an adaptive multi-objective optimization locust algorithm. The technical solution adopted by the present invention is as follows:

[0008] The self-learning path generation method based on the adaptive multi-objective optimization locust algorithm includes the following steps:

[0009] Knowledge points and key modules are numbered using natural number coding. A controllable cognitive load clustering model is introduced to cluster the knowledge points to be learned. Based on the knowledge point-module importance coefficient matrix, knowledge points are nested and scheduled, and initial learning path individuals are generated to form the first path population.

[0010] Construct a multi-objective fitness function with four objective dimensions, and use the weighted sum method or Pareto non-inferiority ranking method to evaluate and sort the path fitness of the first path population;

[0011] The first N paths with the best fitness are selected from the first path population after path fitness evaluation and sorting as leader individuals, and the paths other than the leader individuals are subjected to a structural crossover operation with the leader individual to obtain a second dynamic path population;

[0012] Applying a local structural perturbation process to any learning path individual in the second dynamic path population to obtain a third fine-tuned dynamic path population; the local structural perturbation process adopts one or more of a single-point insertion operator, a single-point exchange operator, and a segment reversal operator;

[0013] Based on the multi-objective fitness function of the four target dimensions, the fitness of each learning path individual in the third fine-tuning dynamic path population is evaluated one by one, and the learning path individual with the best fitness in the current round or the first E Pareto optimal learning path individuals are retained; where E is a natural number greater than 1;

[0014] The maximum number of evolutionary iterations and the fitness convergence threshold are preset; if the maximum number of evolutionary iterations is reached or the fitness change of the local optimal path after k consecutive iterations is less than the fitness convergence threshold, the evolution process is terminated and the Pareto optimal self-learning path set is output; the k is a natural number greater than 1.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] This method uses natural numbers to encode knowledge points and key modules, and employs a controllable cognitive load clustering model and a knowledge point-module importance coefficient matrix to generate the initial path population, ensuring diverse learning paths. Furthermore, this method employs a weighted or Pareto optimal algorithm to assess the path fitness of the initial path population, comprehensively evaluating the self-study effectiveness of each path and ensuring that the recommended content is both accessible and manageable for students.

[0017] The present invention cleverly selects the top N paths with the best fitness from the first path population after path fitness evaluation and sorting as leader individuals, and performs structural cross-operation on the paths other than the leader individuals with the leader individuals to obtain a second dynamic path population, which can not only retain the high-quality self-study path structure, but also stimulate path variation to broaden students' choice space.

[0018] The present invention performs a local perturbation enhanced search on the second dynamic path population (i.e., adopts one or more of the single-point insertion operator, the single-point exchange operator, and the segment reversal operator), which retains the high-quality self-learning path structure and stimulates path variation to broaden the student's choice space.

[0019] The present invention constructs a four-dimensional multi-objective fitness function and iteratively judges the optimal path for the third fine-tuning dynamic path population, retains the optimal iterative learning path individual or the top E Pareto optimal learning path individuals at any time, and archives and inherits the best-performing learning path to ensure that the self-learning path is steadily improved and stable and reliable during the optimization process.

[0020] In summary, the present invention has the advantages of simple logic, dynamic reliability, etc., and has high practical value and promotion value in the field of self-learning path generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope of protection. For those skilled in the art, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 It is a logic flow chart of the present invention.

[0023] Figure 2 Schematic diagram of the linear sequential crossover LOX algorithm of the present invention. DETAILED DESCRIPTION

[0024] To make the purpose, technical solutions, and advantages of this application more clear, the present invention is further described below with reference to the accompanying drawings and examples. Implementation methods of the present invention include, but are not limited to, the following examples. All other embodiments obtained by persons of ordinary skill in the art based on the examples in this application without creative effort are within the scope of protection of this application.

[0025] like Figures 1 to 2 As shown, this embodiment provides a self-learning path generation method based on the adaptive multi-objective optimization locust algorithm, which is based on the knowledge point map, student cognitive status and task characteristics, and through natural number encoding, multi-objective fitness evaluation mechanism, combined path perturbation and elite retention strategy, continuously evolves and globally optimizes the learning path.

[0026] Here, the self-learning path generation method includes the following steps:

[0027] The first step is population initialization:

[0028] Here, natural numbers are used to encode the order of key modules and knowledge points; an initial path population is constructed using a "cognitive load clustering + knowledge point-module importance" strategy. Key modules are the "knowledge units / components" in learning path planning. They are essential components of a learning path and crucial for learners to achieve their learning goals. They are the core content of a learning path, typically encompassing core knowledge, skills, and abilities that learners must master.

[0029] Here, the expression of the controllable cognitive load clustering model is: in, Indicates the The load of knowledge points; Clusters representing knowledge points; represents the cognitive load threshold; The number of clusters representing knowledge points; Indicates the number of knowledge points.

[0030] In addition, Knowledge point load The expression is:

[0031]

[0032]

[0033] in, Indicates the The cognitive difficulty of each knowledge point is in the range of [0,1]; Indicates the pre-dependency level; Indicates the Average time to master each knowledge point; Indicates the The cognitive difficulty of each knowledge point The corresponding normalized weighting coefficient; Indicates the pre-dependency level The corresponding normalized weighting coefficient; Indicates the The average time to master each knowledge point The corresponding normalized weighting coefficient. Here, the cognitive difficulty of knowledge points can be quantified using traditional methods such as knowledge level and cognitive depth, knowledge graph and item response theory (IRT), learning behavior data, and difficulty grading based on knowledge graphs.

[0034] Here, the expression of the knowledge point-module importance coefficient matrix is:

[0035]

[0036] in, Indicates the Knowledge point for the The importance coefficient matrix of the module, ; Indicates the The learning time of each knowledge point; Indicates the The width of the learning window for each knowledge point; Represents the cognitive jump distance between knowledge points and modules.

[0037] The first path population is encoded in a nested sequence, and its expression is:

[0038]

[0039] in, Represents the first Cluster learning module; Represents the first Cluster learning module The corresponding knowledge point subset is obtained by the controllable cognitive load clustering algorithm, so that the cognitive load in each cluster does not exceed the preset threshold.

[0040] The second step is path fitness evaluation:

[0041] This embodiment uses a four-objective function to measure the mastery improvement, learning burden, cognitive jump and interest maintenance of the path. Specifically, the four-dimensional multi-objective fitness function The expression is: in, represents the mastery rate maximization function; represents the minimum function of total learning burden; represents the structure span function; represents the interest decay function.

[0042] For four-dimensional multi-objective fitness function To find the minimum value, it is expressed as:

[0043] ;

[0044] Here, the mastery rate maximization function The expression is:

[0045]

[0046] in, Indicates the number of knowledge points At the upper limit of the learning path, ; Indicates the The mastery level of each knowledge point before learning the path; Indicates the The mastery level of each knowledge point after learning the path. Here, the quantification of the mastery level before path learning adopts traditional pre-test assessment, system evaluation, etc. The pre-test assessment evaluates the learner's knowledge level before the start of the learning path by designing a set of test plans related to the learning objectives; the system evaluation is based on the learner's historical records or input academic qualifications and other information, combined with the preset knowledge point test threshold, to determine the set of knowledge points currently mastered by the learner. In addition, the mastery level after path learning is also evaluated by traditional post-test assessment, etc. This post-test assessment usually includes the following forms: standardized tests (that is, designing a set of standardized test plans that cover all key knowledge points and skills in the learning path), project assignments (assessing the learner's comprehensive ability through actual project assignments), etc.

[0047] Here, after the path is completed, the total improvement in the student's mastery of each knowledge point is calculated (negative is used to minimize the goal).

[0048] Here, the total learning burden minimum function The expression is:

[0049]

[0050] in, Indicates the number of knowledge points At the upper limit of the learning path, ; Indicates Knowledge point The required study time.

[0051] Here, the structure span function The expression is: in, Indicates the number of knowledge points At the upper limit of the learning path, ; Indicates the importance weight between knowledge points; Indicates the Knowledge point and The cognitive distance of knowledge points.

[0052] Here, the interest decay function The expression is:

[0053]

[0054] in, Indicates the number of knowledge points At the upper limit of the learning path, ; Indicates the content attractiveness of the current knowledge point; represents the historical completion rate; represents the exponential decay coefficient. The content attractiveness of a knowledge point can be assessed using traditional learning behavior data analysis, such as Item Response Theory (IRT). This analysis uses a learning management system (LMS) or other learning platform to collect learner behavior data and analyze learners' engagement and interaction with different knowledge points. Item Response Theory (IRT), a method used to assess learners' mastery of knowledge points, can also be used to assess the attractiveness of knowledge point content. By analyzing learners' performance on knowledge point-related tests, we can infer the learner's appeal to the knowledge point.

[0055] Here, a weighted or Pareto optimal algorithm is used to evaluate the path fitness of the first path population. This comprehensively assesses the self-study effectiveness of each path, ensuring that the recommended content is both easy to grasp and manageable for students. Path fitness is a quantitative metric used to assess the performance of a path in meeting specific objectives or constraints.

[0056] Step 3: Leader selection and structure intersect:

[0057] This embodiment dynamically selects high-quality paths in iterations as leaders and uses LOX crossover to introduce high-quality path segments into the population. This preserves the high-quality self-study path structure while stimulating path variation to broaden students' choice space. Specifically:

[0058] Based on the salp swarm algorithm, the first path population after path fitness evaluation and sorting is dynamically shrunk, and its expression is:

[0059]

[0060] in, Indicates the proportion of leaders in the current iteration round; represents the initial leader ratio; represents the target leader ratio at the final convergence; Indicates the current iteration number; represents the maximum number of iterations, Represents the threshold parameter that controls the curvature of fitness change.

[0061] In any iteration, based on the path fitness evaluation and the comprehensive fitness value of multiple objectives of the learning path individuals in the sorted first path population, the global optimal learning path individual is dynamically updated, including the following steps:

[0062] If the current In the iteration Learning Path Individual The comprehensive fitness value Smaller than the global optimal learning path individual of the previous iteration The comprehensive fitness value of the current In the iteration Learning Path Individual As the first The global optimal learning path individual in the iteration Otherwise, the original global optimal learning path individual will be retained, and its expression is:

[0063] ;

[0064] After evaluating and ranking the path fitness, several globally optimal learning path individuals in the first path population are ranked by fitness, and the first one ranked in the top three is selected. In the iteration Learning Path Individual As a leader path, and with the current In the iteration Learning Path Individual Perform a linear sequential crossover operation to generate the In the iteration Learning Path Individual , whose expression is:

[0065]

[0066] in, Indicates the current In the iteration Individual learning paths; ;LOX() represents the leader-guided crossover operator function.

[0067] Step 4: Local perturbation enhanced search:

[0068] Three perturbation operators, namely single-point insertion, exchange, and segment reversal, are performed on the path. Here, by fine-tuning the learning order, the path is helped to avoid local traps and the flexibility and adaptability of the self-learning path are increased.

[0069] Here, the single point insertion operator includes the following steps:

[0070] Learning path individuals in the second dynamic path population Select a location and location Insert knowledge points , insert knowledge points The expression of the learning path individual after is:

[0071]

[0072] in, Indicates location knowledge points, .

[0073] In addition, the single-point exchange operator includes the following steps:

[0074] Learning path individuals in the second dynamic path population Position in Knowledge points and location Knowledge points To exchange, the expression is:

[0075]

[0076] in, Represents the individual learning path after executing the single-point exchange operator.

[0077] The segment reversal operator of this embodiment includes the following steps:

[0078] Learning path individuals in the second dynamic path population Middle section To reverse the order, the expression is:

[0079] Among them, Rev() represents the reversal function.

[0080] Step 5: Elite Retention Mechanism:

[0081] The optimal path of each generation or the top E Pareto optimal paths are preserved intact, and the best-performing learning paths are archived and inherited to ensure that the self-learning paths steadily improve and become stable and reliable during the optimization process.

[0082] Specifically, the expression of the optimal iterative learning path individual at any time is:

[0083]

[0084] in, Indicates the Iterative learning path population; represents the candidate learning path population generated by crossover, perturbation, and update operations; Indicates the The multi-objective fitness evaluation function of individual learning paths, .

[0085] Alternatively, retain the first E Pareto optimal learning path individuals, which can be expressed as:

[0086]

[0087] in, Indicates that under Pareto non-dominated sorting, The set of top E high-quality learning path individuals selected in the generation; Indicates selecting the first E non-inferior paths; Indicates the Generation of high-quality learning path individuals; Pareto non-dominance represents Pareto non-domination.

[0088] Step 6: Terminate the judgment:

[0089] This embodiment presets a maximum number of iterations and a fitness change steepness threshold; if the maximum number of iterations is reached or the fitness difference of k consecutive iterations is less than the fitness change steepness threshold, the iteration is stopped and the Pareto optimal self-learning path set is output.

[0090] The above embodiments are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any changes that adopt the design principles of the present invention and any changes made through non-creative work on this basis should fall within the scope of protection of the present invention.

Claims

1. A self-learning path generation method based on an adaptive multi-objective optimization locust algorithm, characterized in that: The following steps are involved: Use natural number coding to number knowledge points and key modules; introduce a controllable cognitive load clustering model to cluster the knowledge points to be learned; Based on the knowledge point-module importance coefficient matrix, the knowledge points are scheduled for module nesting, and initial learning path individuals are generated to form the first path population; Construct a multi-objective fitness function with four objective dimensions, and use the weighted sum method or Pareto non-inferiority ranking method to evaluate and sort the path fitness of the first path population; The first N paths with the best fitness are selected from the first path population after path fitness evaluation and sorting as leader individuals, and the paths other than the leader individuals are subjected to a structural crossover operation with the leader individual to obtain a second dynamic path population; Applying a local structural perturbation process to any learning path individual in the second dynamic path population to obtain a third fine-tuned dynamic path population; the local structural perturbation process adopts one or more of a single-point insertion operator, a single-point exchange operator, and a segment reversal operator; Based on the multi-objective fitness function of the four target dimensions, the fitness of each learning path individual in the third fine-tuning dynamic path population is evaluated one by one, and the learning path individual with the best fitness in the current round or the first E Pareto optimal learning path individuals are retained; where E is a natural number greater than 1; The maximum number of evolutionary iterations and the fitness convergence threshold are preset; if the maximum number of evolutionary iterations is reached or the fitness change of the local optimal path after k consecutive iterations is less than the fitness convergence threshold, the evolution process is terminated and the Pareto optimal self-learning path is output; the k is a natural number greater than 1.

2. The self-learning path generation method based on the adaptive multi-objective optimization locust algorithm according to claim 1 is characterized in that: The expression of the controllable cognitive load clustering model is: in, Indicates the The load of knowledge points; Clusters representing knowledge points; represents the cognitive load threshold; The number of clusters representing knowledge points; Indicates the number of knowledge points; The said Knowledge point load The expression is: in, Indicates the The cognitive difficulty of each knowledge point is in the range of [0,1]; Indicates the pre-dependency level; Indicates the Average time to master each knowledge point; Indicates the The cognitive difficulty of each knowledge point The corresponding normalized weighting coefficient; Indicates the pre-dependency level The corresponding normalized weighting coefficient; Indicates the The average time to master each knowledge point The corresponding normalized weighting coefficient.

3. The self-learning path generation method based on the adaptive multi-objective optimization locust algorithm according to claim 2 is characterized in that: The expression of the knowledge point-module importance coefficient matrix is: in, Indicates the Knowledge point for the The importance coefficient matrix of the module, ; Indicates the The learning time of each knowledge point; Indicates the The width of the learning window for each knowledge point; Represents the cognitive jump distance between knowledge points and modules.

4. The self-learning path generation method based on the adaptive multi-objective optimization locust algorithm according to claim 3 is characterized in that: Also includes: The first path population is encoded in a nested sequence, and its expression is: in, Represents the first Cluster learning module; Represents the first Cluster learning module The corresponding subset of knowledge points.

5. The self-learning path generation method based on the adaptive multi-objective optimization locust algorithm according to claim 3 is characterized in that: The first N paths with the best fitness are selected from the first path population after path fitness evaluation and sorting as leader individuals, and paths other than the leader individuals are subjected to a structural crossover operation with the leader individual to obtain a second dynamic path population, including the following steps: Based on the salp swarm algorithm, the first path population after path fitness evaluation and sorting is dynamically shrunk, and its expression is: in, Indicates the proportion of leaders in the current iteration round; represents the initial leader ratio; represents the target leader ratio at the final convergence; Indicates the current iteration number; represents the maximum number of iterations, Represents the threshold parameter that controls the curvature of fitness change; In any iteration, based on the path fitness evaluation and the comprehensive fitness value of multiple objectives of the learning path individuals in the sorted first path population, the global optimal learning path individual is dynamically updated, including the following steps: If the current In the iteration Learning Path Individual The comprehensive fitness value Smaller than the global optimal learning path individual of the previous iteration The comprehensive fitness value of the current In the iteration Learning Path Individual As the first The global optimal learning path individual in the iteration Otherwise, the original global optimal learning path individual will be retained, and its expression is: After evaluating and ranking the path fitness, several globally optimal learning path individuals in the first path population are ranked by fitness, and the first one ranked in the top three is selected. In the iteration Learning Path Individual As a leader path, and with the current In the iteration Learning Path Individual Perform a linear sequential crossover operation to generate the In the iteration Learning Path Individual , whose expression is: in, Indicates the current In the iteration Individual learning paths; ;LOX() represents the leader-guided crossover operator function.

6. The self-learning path generation method based on the adaptive multi-objective optimization locust algorithm according to claim 5 is characterized in that: The single point insertion operator comprises the following steps: Learning path individuals in the second dynamic path population Select a location and location Insert knowledge points , insert knowledge points The expression of the learning path individual after is: in, Indicates location knowledge points, .

7. The self-learning path generation method based on the adaptive multi-objective optimization locust algorithm according to claim 5 is characterized in that: The single-point exchange operator includes the following steps: Learning path individuals in the second dynamic path population Position in Knowledge points and location Knowledge points To exchange, the expression is: in, Represents the individual learning path after executing the single-point exchange operator.

8. The self-learning path generation method based on the adaptive multi-objective optimization locust algorithm according to claim 5 is characterized in that: The segment reversal operator comprises the following steps: Learning path individuals in the second dynamic path population Middle section To reverse the order, the expression is: Among them, Rev() represents the reversal function.

9. The self-learning path generation method based on the adaptive multi-objective optimization locust algorithm according to claim 6, 7 or 8, characterized in that: The four-dimensional multi-objective fitness function The expression is: ,in, represents the mastery rate maximization function; represents the minimum function of total learning burden; represents the structure span function; represents the interest decay function; For four-dimensional multi-objective fitness function To find the minimum value, it is expressed as: The mastery rate maximization function The expression is: ,in, Indicates the number of knowledge points At the upper limit of the learning path, ; Indicates the The mastery level of each knowledge point before learning the path; Indicates the The mastery level of each knowledge point after path learning; The total learning burden minimum function The expression is: in, Indicates the number of knowledge points At the upper limit of the learning path, ; Indicates Knowledge point Required study time; The structure span function The expression is: in, Indicates the number of knowledge points At the upper limit of the learning path, ; Indicates the importance weight between knowledge points; Indicates the Knowledge point and The cognitive distance of knowledge points; The interest decay function The expression is: in, Indicates the number of knowledge points At the upper limit of the learning path, ; Indicates the content attractiveness of the current knowledge point; represents the historical completion rate; represents the exponential decay coefficient.

10. The self-learning path generation method based on the adaptive multi-objective optimization locust algorithm according to claim 9 is characterized in that: The expression for retaining the learning path individual with the best fitness in the current round is: ,in, Indicates the Iterative learning path population; represents the candidate learning path population generated by crossover, perturbation, and update operations; Indicates the The multi-objective fitness evaluation function of individual learning paths, ; Or retain the first E Pareto optimal learning path individuals, the expression is: ,in, Indicates that under Pareto non-dominated sorting, The set of the top E high-quality learning path individuals selected in the generation; Indicates selecting the first E non-inferior paths; Indicates the Generation of high-quality learning path individuals; Pareto non-dominance means Pareto non-dominance.

Images