Learning strategy recommendation method based on knowledge graph and dynamic evaluation

Abstract

The invention discloses a learning strategy recommendation method based on a knowledge graph and dynamic evaluation, and belongs to the field of recommendation. A learning ability vector and a comprehensive knowledge state matrix are defined, an initial to-be-learned queue is generated, after first-repair knowledge point integrity check and topological sorting are carried out, personalized adjustment is carried out on a basic learning path based on the learning ability vector, and adaptive learning materials are matched for each knowledge point in the personalized learning path; and finally, generating a current-day learning task based on the daily learning time budget. After the learning task on that day is completed, pushing an adaptive test question, updating the comprehensive knowledge state matrix according to a test result, and executing an instant adjustment strategy based on the test result of a single knowledge point and an overall adjustment strategy based on a learning trend; and generating a next-day learning task according to the adjusted strategy until a preset learning target is completed. According to the method, continuous self-adaptive optimization of learning paths and resource recommendation is realized, so that the accuracy and efficiency of personalized learning are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of recommendation, and particularly relates to a learning strategy recommendation method based on knowledge graphs and dynamic evaluation. Background Technology

[0002] Personalized learning, as a core development direction in the field of educational technology, requires providing each student with a tailored learning experience. Its technological foundation typically involves knowledge graphs and recommendation algorithms. Knowledge graphs are used to structurally represent knowledge points and their prerequisite dependencies, while recommendation algorithms are responsible for matching appropriate learning content and paths with the student's current state.

[0003] In existing technologies, a common approach to achieving personalized learning is through rule-based personalized learning path generation. This method first diagnoses a student's knowledge weaknesses using pre-tests or historical scores, and then generates a linear learning sequence based on rules pre-set by teaching experts. Another common approach is collaborative filtering-based recommendation methods. This method analyzes the learning behaviors of student groups with similar learning characteristics and recommends learning materials or paths used by students of the same age or level. These methods achieve a degree of personalization in learning, providing students with learning suggestions that differ from a uniform teaching pace.

[0004] However, the aforementioned existing technical solutions still have significant limitations. First, rule-based methods heavily rely on expert experience, resulting in a rigid path generation mechanism that cannot dynamically adapt to students' real-time changing learning states and abilities. For example, they cannot perceive students' learning speed or knowledge retention rates, making it difficult to make fine adjustments to learning pace and review strategies. Second, collaborative filtering methods are essentially a group profile recommendation, resulting in lagging recommendations that cannot address differences in students' learning abilities. Furthermore, they ignore the inherent prerequisite dependencies within the knowledge graph, potentially leading to recommended learning paths that violate the scientific order of knowledge construction. In conclusion, existing technologies struggle to achieve a proactive and adaptive learning strategy planning scheme that can continuously evaluate, provide real-time feedback, and deeply integrate individual student learning abilities, knowledge states, and the inherent logic of knowledge. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a learning strategy recommendation method based on knowledge graphs and dynamic evaluation. By integrating student ability vectors, knowledge states, and knowledge graph dependencies for dynamic evaluation, and by providing real-time feedback based on learning outcomes to form a closed loop, this method achieves continuous adaptive optimization of learning paths and resource recommendations, thereby significantly improving the accuracy and efficiency of personalized learning.

[0006] The technical solution adopted in this invention is as follows:

[0007] Firstly, this invention proposes a learning strategy recommendation method based on knowledge graphs and dynamic evaluation, comprising the following steps:

[0008] Step 1: Periodically or based on trigger conditions, assess the student's learning ability vector and comprehensive knowledge status matrix. The learning ability vector includes a learning speed factor, a knowledge retention factor, and a knowledge application ability factor.

[0009] Step 2: Determine the target knowledge point set based on the preset learning objectives, calculate the priority score of each knowledge point in the set by combining the learning ability vector and the comprehensive knowledge state matrix, and generate an initial learning queue based on the priority scores;

[0010] Step 3: Perform a prerequisite knowledge point integrity check and topological sorting on the initial learning queue to generate a basic learning path, and make personalized adjustments to the basic learning path based on the learning ability vector to obtain a personalized learning path.

[0011] Step 4: For each knowledge point in the personalized learning path, match suitable learning materials that match the learning ability vector from the learning resource library, and generate daily learning tasks based on the daily learning time budget.

[0012] Step 5: After completing the daily learning tasks, push out suitable test questions, update the comprehensive knowledge status matrix according to the test results, and execute the real-time adjustment strategy based on the test results of individual knowledge points and the overall adjustment strategy based on learning trends; generate the next day's learning tasks according to the adjusted strategies.

[0013] Step 6: Repeat step 5 until the learning objectives preset in step 2 are achieved.

[0014] Furthermore, the comprehensive knowledge state matrix constructs a state vector for each knowledge point, including the mastery probability of the knowledge point, betweenness centrality, degree centrality, time interval since the last learning, and intent intensity.

[0015] Furthermore, in step 1, the triggering condition is detecting that the student's learning has entered a plateau period; the judgment condition for entering the plateau period is: statistically analyzing the daily increase in the average mastery probability of all knowledge points learned by the student within a consecutive preset number of days. If the daily increase is lower than a preset threshold, it is determined that the student has entered a plateau period.

[0016] Furthermore, the priority score is calculated using the following formula:

[0017] ;

[0018] in, For knowledge point k iPriority score, For knowledge point k i The probability of mastering For knowledge point k i betweenness centrality, It is the time decay coefficient. It is the distance from the last time the knowledge point was learned, k. i The time interval, It is the intention enhancement coefficient. For students to understand knowledge point k i Intent intensity, It is a capability adaptation function, which means that when the learning speed factor is lower than the lower limit of the speed threshold, knowledge points with high degree centrality are given priority enhancement.

[0019] Furthermore, step 3, which involves checking the completeness of prerequisite knowledge points and performing topological sorting on the initial learning queue, includes:

[0020] For each knowledge point in the initial learning queue, if the mastery probability of its prerequisite knowledge points is lower than the mastery threshold and it is not included in the queue, it is dynamically inserted into the initial learning queue according to the priority score and prerequisite relationship of the prerequisite knowledge points to obtain the expanded learning queue.

[0021] Initialize an empty ordered path list and iteratively execute the following steps until the expanded learning queue is empty: Take the first knowledge point from the head of the expanded learning queue; determine if all prerequisite knowledge points for that knowledge point are ready, where ready means that the prerequisite knowledge point already exists in the ordered path list or its mastery probability reaches the mastery threshold; if all prerequisite knowledge points are ready, remove the knowledge point from the expanded learning queue and add it to the end of the ordered path list; if there are unread prerequisite knowledge points, move the knowledge point from the head to the tail of the expanded learning queue.

[0022] The resulting ordered list of paths is used as the base learning path.

[0023] Furthermore, the process of personalizing the basic learning path based on the learning ability vector includes:

[0024] If the learning speed factor is higher than the upper limit of the speed threshold, then the related knowledge points in the knowledge graph that have no direct prerequisite dependencies will be arranged for parallel learning.

[0025] If the learning speed factor is lower than the lower limit of the speed threshold, a strict sequential learning strategy is adopted, and comprehensive knowledge points are broken down into sub-knowledge points;

[0026] If the knowledge retention factor is lower than the retention threshold, review knowledge points will be inserted into the basic learning path at preset intervals.

[0027] Furthermore, the learning materials in the learning resource library are characterized by feature vectors, which include the level of detail, the level of abstraction, the number of examples, and the level of interaction; and suitable learning materials that match the learning ability vectors are matched from the learning resource library according to preset ability-feature matching rules.

[0028] Furthermore, the capability-feature matching rules include:

[0029] For students whose learning speed factor is below the lower limit of the speed threshold, we will prioritize recommending learning materials with high levels of detail and a large number of examples.

[0030] For students whose knowledge retention factor is below the retention threshold, priority will be given to recommending learning materials containing comparisons of multiple solutions and knowledge summaries;

[0031] For students whose knowledge application ability factor is below the application threshold, priority will be given to recommending learning materials with low abstraction and high interactivity.

[0032] Furthermore, the real-time adjustment strategy based on the test results of a single knowledge point includes:

[0033] If a knowledge point is well understood after testing, it will be removed from the personalized learning path.

[0034] If a knowledge point is partially mastered, its priority is lowered and the personalized learning path is updated so that it appears in subsequent learning tasks.

[0035] If a knowledge point is not mastered, a reinforcement learning process is initiated, which includes: identifying weak areas, determining and reviewing the prerequisite knowledge points that have not been mastered, and recommending more detailed or more basic learning materials for that knowledge point.

[0036] Furthermore, the overall adjustment strategy based on learning trends includes:

[0037] If a student's performance on multiple consecutive knowledge points is lower than expected, the system will automatically increase the detail and number of examples in the learning materials recommended for all knowledge points to be learned, and insert general guidance materials.

[0038] In a second aspect, the present invention provides an electronic device, including a memory and a processor;

[0039] The memory is used to store computer programs;

[0040] The processor is configured to implement the learning strategy recommendation method based on knowledge graphs and dynamic evaluation as described above when executing the computer program.

[0041] Thirdly, the present invention proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the learning strategy recommendation method based on knowledge graphs and dynamic evaluation as described above.

[0042] The beneficial effects of this invention are:

[0043] This invention constructs a dynamic, multi-dimensional student model by periodically or trigger-based assessment of students' learning ability vectors and comprehensive knowledge state matrices, thus providing a precise data foundation for personalized learning strategy recommendations. When generating the initial learning queue based on preset learning objectives, priority score calculation incorporates the student's current knowledge state, ability characteristics, and the structural features of the knowledge graph, ensuring the scientific and targeted nature of the learning sequence. Furthermore, by performing prerequisite knowledge point integrity checks and topological sorting on the queue, the generated basic learning path strictly adheres to the dependencies between knowledge points, avoiding learning obstacles caused by a lack of foundational knowledge. Personalized adjustments are then made based on the learning ability vector, making the path rhythm more aligned with the student's actual learning pace. For example, knowledge points are broken down for students with slow learning speeds, or review nodes are inserted for students with low retention rates, thereby improving the rationality and adaptability of the learning path.

[0044] In the resource matching and task execution phases, each knowledge point in the personalized learning path is matched with suitable learning materials that align with the student's ability vector. Daily tasks are generated based on the daily time budget, ensuring a precise match between learning content and student abilities, as well as controllable learning progress. Crucially, after completing the daily tasks, the knowledge state matrix is ​​updated by reviewing the questions. Immediate adjustments based on the review results of individual knowledge points and overall adjustments based on overall learning trends are implemented to generate the next day's tasks. This closed-loop feedback mechanism enables real-time monitoring and dynamic optimization of the learning process. By repeating this process until the learning objectives are achieved, the entire solution not only ensures the continuous adaptability of the learning strategy but also effectively addresses changes in student learning status, preventing path rigidity and ultimately significantly improving learning efficiency and the achievement rate of learning objectives. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating a learning strategy recommendation method based on knowledge graphs and dynamic evaluation proposed in this invention.

[0046] Figure 2 This is a flowchart illustrating the process of generating the initial learning queue;

[0047] Figure 3 This is a flowchart illustrating the process of generating a personalized learning path;

[0048] Figure 4This is a flowchart illustrating the process of matching suitable learning materials for each knowledge point in a personalized learning path.

[0049] Figure 5 This is a flowchart illustrating the process of dynamically adjusting learning paths and recommended learning materials based on test results;

[0050] Figure 6 This is a schematic diagram of the electronic device terminal structure of the present invention. Detailed Implementation

[0051] The present invention will be further described and illustrated below with reference to specific embodiments. The embodiments described are merely examples of the content of this disclosure and do not limit the scope of the invention. The technical features of each embodiment in the present invention can be combined accordingly, provided that there is no mutual conflict.

[0052] The accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0053] The flowchart shown in the attached diagram is merely an illustrative example and does not necessarily include all steps. For example, some steps may be broken down, while others may be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0054] like Figure 1 As shown, this invention proposes a learning strategy recommendation method based on knowledge graphs and dynamic evaluation. By assessing students' learning abilities and integrating knowledge status and knowledge graph technology, a learning improvement plan including learning materials and practice questions is formulated, thereby achieving personalized learning strategy recommendations.

[0055] A learning strategy recommendation method based on knowledge graphs and dynamic evaluation, such as... Figure 1 As shown, it includes the following steps:

[0056] S1 assesses the student's learning ability vector and knowledge state matrix.

[0057] (1.1) Definition of learning ability

[0058] By extracting students' learning speed, memory retention, problem-solving speed, and problem-solving accuracy from historical learning data, a learning ability vector is defined. The purpose is to quantify students' learning abilities and comprehensively assess their current state across all knowledge points, providing a data foundation for planning. This learning ability vector is updated periodically according to a preset cycle.

[0059] In this embodiment, the learning ability vector The definition is as follows:

[0060]

[0061] in, This is the learning speed factor, used to quantify a student's ability to absorb new knowledge per unit of time. It is calculated by the average learning time required for a student to reach a level of mastery. , It refers to the number of knowledge points that are firmly grasped within a specific time window t. t can be dynamically set according to the learning stage, such as t=30 days in a regular semester, t=7 days in a pre-exam sprint, etc. This refers to the total effective learning time, which is the time a student spends focused on learning, excluding ineffective time such as breaks and interruptions. For example, if a student masters 15 knowledge points in 30 days, the total effective learning time is 450 minutes. Knowledge points per minute.

[0062] It is a knowledge retention factor used to assess students' long-term memory capacity and determine the frequency and intensity of review strategies. It is calculated based on the degree of forgetting after a certain interval. , It is an indicator function; its value is 1 if the condition is true, and 0 otherwise. It is a knowledge point Currently, the probability is known. It is a knowledge point The current mastery probability was set one week ago, and the mastery threshold was uniformly set to 0.7 to ensure consistent assessment. For example, if a student mastered 20 knowledge points one week ago and still maintains mastery of 18, then... .

[0063] It refers to the ability to apply knowledge, used to measure the ability to transfer and apply knowledge. It influences the choice of question difficulty and the preference for different types of materials, and is assessed through students' performance on comprehensive and application questions. , It is the average accuracy rate for comprehensive application questions (involving multiple knowledge points). This is the average accuracy rate for basic questions on a single knowledge point. For example, if a student's accuracy rate for basic questions is 0.85 and their accuracy rate for comprehensive questions is 0.68, then... .

[0064] (1.2) Constructing a comprehensive knowledge state matrix

[0065] For each knowledge point k i Based on the existing mastery probability and color marking, define the knowledge point state vector. .in, It is knowledge point k i The probability of mastery (0-1) is calculated by updating historical answer data; It is knowledge point k i The betweenness centrality, calculated from the knowledge graph, reflects the importance of the knowledge point as a hub; It is knowledge point k i Degree centrality reflects the fundamental nature and broad connectivity of knowledge points; It is the distance from the last time the knowledge point was learned, k. i The time interval (days) reflects the degree of memory decay; It is the intensity of intent (0-1), calculated by analyzing behavioral data such as students' search history and the concentration of incorrect questions.

[0066] In this embodiment, intent strength is used to quantify a student's learning need for a particular knowledge point, and the formula is as follows:

[0067]

[0068] in, It represents the frequency of note-taking per unit of time, reflecting the level of learning activity. T is the length of the time window (usually set to 7 days). It represents the number of times knowledge point K is noted within the time window T. The frequency of questions asked about knowledge point K reflects the level of difficulty in understanding. It refers to the number of questions asked about knowledge point K. It refers to the number of notes taken for knowledge point K. It is the time decay factor (the smaller the time decay, the more urgent the demand). This is the attenuation coefficient, used to control the attenuation rate affected by time (initial value 0.1), t current It is the current time, t last This is the time of the most recent operation on knowledge point K. It is the weighting coefficient.

[0069] The final complete integrated knowledge state matrix is ​​as follows:

[0070]

[0071] The AI ​​agent automatically retrieves the latest evaluation results from the database periodically (e.g., after each task) or when a learning plateau is detected (e.g., progress stagnation for three consecutive days). In this embodiment, the determination of progress stagnation can be defined as: calculating the average daily increase in the mastery probability of all knowledge points learned by the student over three consecutive days; if the daily increase is lower than a threshold, progress is considered stagnant.

[0072] S2, based on the preset learning objectives, generates an initial learning queue.

[0073] like Figure 2 As shown, it includes the following steps:

[0074] (2.1) Confirm whether the improvement plan is a long-term goal or a short-term goal.

[0075] In this embodiment, the system presets short-term and long-term goals for the user to choose from. For example, the short-term goal is to upgrade all red and yellow knowledge points in the current course to green. The long-term goal is to master all the knowledge points of the entire course by the end of this semester. By determining the specific goals to be achieved in this improvement plan, a queue of knowledge points that need to be learned or reviewed is selected.

[0076] (2.2) Generate planning objectives

[0077] The plan objectives are selected from the knowledge point queue corresponding to the improvement plan. In this embodiment, the selection is achieved by designing recommended objective types, which include gap-filling objectives, preparatory objectives, and sprint objectives. Here, gap-filling objectives refer to consolidating weak knowledge points; preparatory objectives refer to learning the prerequisite knowledge for next week's teaching content; and sprint objectives refer to preparing for the core test points.

[0078] This invention transforms abstract goals into a set of target knowledge points within a knowledge graph, based on a knowledge graph containing student mastery and prerequisite relationships, and knowledge point attribute tags. For example, if a student has a weak grasp of the function chapter, and that student chooses a gap-filling objective, the objective will be mapped as follows: ,in This is the set of knowledge points that need to be learned for this objective.

[0079] (2.2) Priority sorting

[0080] Based on the comprehensive knowledge state matrix, the knowledge points required for this objective are prioritized, and the priority score formula is as follows:

[0081]

[0082] in, It is knowledge point k i Priority score, It is a time decay factor; forgetting intensifies over time. This is the time decay coefficient, which is usually taken as 0.1; It is an intentional enhancement factor that respects students' learning interests. This is the intention enhancement coefficient, which is usually taken as 0.5; It is a capability adaptation function. In this embodiment, the specific implementation of the capability adaptation function is as follows: , This is the enhancement coefficient for basic knowledge points, usually taken as 0.3; This is the threshold for slow learning speed. When the learning speed is slow, the basic knowledge points are strengthened. It is usually set to 0.02.

[0083] When prioritizing the target knowledge point set, the state matrix S and learning ability vector output in step 1 are combined. As input, calculate the priority score of each knowledge point in the set of knowledge points required to learn this objective. Sort by score in descending order and output the priority queue. .For example:

[0084] Function properties: P=0.6, C B =0.8, Δt=2, I=0.3, calculate the priority score Priority=0.73;

[0085] Quadratic function: P=0.4, C B =0.7, Δt=5, I=0.1, calculate the priority score Priority=0.65;

[0086] Function definition: P=0.8, C B =0.9, Δt=1, I=0.2, calculate the priority score Priority=0.45;

[0087] The final queue of knowledge points to be learned is called the initial queue to be learned: Q = [function properties, quadratic functions, function definitions].

[0088] S3. Based on the prior dependencies of knowledge points in the preset knowledge graph, the initial learning queue Q is transformed into an ordered learning path, and the ordered learning path is adjusted in a personalized way based on the learning ability vector to obtain a personalized learning path.

[0089] The goal of this step is to transform the initial learning queue Q output from step S2 into an ordered learning path that considers the prerequisite dependencies of knowledge points and then performs personalized optimization. The generation method of the ordered learning path is based on the topological structure of the knowledge graph and incorporates dependencies to ensure the logical correctness of the final learning path.

[0090] like Figure 3 As shown, the specific algorithm flow is as follows:

[0091] (3.1) Generation of ordered learning paths based on knowledge graphs

[0092] Phase 1: Dependency checking and expansion of the learning queue

[0093] First, the system does not directly sort the initial priority queue, but instead performs an "integrity check" to ensure that all necessary prerequisite knowledge points are included.

[0094] Specifically, read the initial learning queue Q, the knowledge graph G, and the current knowledge state matrix S. Then, sequentially traverse each knowledge point k in queue Q. i In the knowledge graph G, query all its direct prerequisite knowledge points to form a set of prerequisite knowledge points, denoted as Preq(k). i This set contains all the data that are learning k. i The knowledge points that must be mastered beforehand.

[0095] Next, check Preq(k) i For each prerequisite knowledge point p in queue Q: if the current mastery probability P(p) of prerequisite knowledge point p is less than 0.8 (i.e., not yet mastered) and p is not included in queue Q, it means the student has neither mastered the prerequisite knowledge nor is it currently scheduled for study. The system will immediately and dynamically insert this prerequisite knowledge point p into a suitable position in queue Q. The determination of this position will comprehensively consider p's own priority score and its prerequisite dependencies with other knowledge points in the queue, placing it before the knowledge points that depend on it.

[0096] Through this stage of processing, the initial queue Q is expanded into a more complete learning queue Q' containing all the necessary prerequisite knowledge points. This step is crucial, as it prevents subsequent learning from being hindered by a lack of foundational knowledge.

[0097] Phase Two: Topological sorting to generate an ordered learning path

[0098] After ensuring the integrity of the learning queue Q', the final ordered learning paths are generated. First, an empty list L is created to store the final generated ordered learning paths.

[0099] Repeat the following steps until the learning queue Q' becomes empty:

[0100] a. Take the first knowledge point k from the head of the current queue Q'.

[0101] b. Determine whether all prerequisite knowledge points for knowledge point k have been processed. "Processed" means that the prerequisite knowledge point either already exists in the generated learning path L, or its current mastery level has reached the threshold (P ≥ 0.8), and no further learning is required.

[0102] c. If the condition is met, that is, all the prerequisite knowledge points of k are ready, then remove k from queue Q' and add it to the end of the learning path L. This means that k has met the learning conditions and can be arranged at this position in the current path.

[0103] If the condition is not met, i.e., k has prerequisite knowledge that is not ready, then k is not added to path L, but is moved from the head of queue Q' to the tail of the queue. The key to this operation is to re-queue k, waiting for its prerequisite knowledge points to be learned first.

[0104] d. When the loop ends, list L stores the final ordered learning paths. This path strictly guarantees that for any knowledge point in the path, all its prerequisite knowledge points appear before it, thus forming a logically correct learning sequence.

[0105] This path generation method solves the knowledge dependency problem through a two-stage approach of first expanding and then sorting, ensuring the scientific validity and feasibility of the learning path.

[0106] (3.2) Personalized path adjustment based on learning ability vector

[0107] After generating the ordered learning path L, based on the student's learning ability vector Personalized adjustments were made to better suit students' actual learning characteristics and needs. These adjustments primarily focused on three aspects:

[0108] a. Learning speed adaptation adjustment

[0109] In this embodiment, based on the student's learning speed factor The learning pace and structure of the path are optimized.

[0110] if This indicates that students learn quickly, allowing for the parallel arrangement of highly related knowledge points in the knowledge graph that have no direct prerequisite dependencies. For example, in mathematics learning, the highly related knowledge points "function properties" and "function graphs" can be studied concurrently within the same learning period. Furthermore, the time spent on basic practice for each knowledge point can be appropriately reduced, allowing immediate progress to the next knowledge point after mastery is confirmed. In this embodiment, .

[0111] if This indicates that students learn slowly, so a strict sequential approach is necessary, adhering to the principle of mastering one thing at a time before moving on to the next, ensuring that each knowledge point is mastered before proceeding to the next. Furthermore, highly complex knowledge points can be broken down into finer-grained sub-knowledge points for step-by-step learning. In this embodiment, .

[0112] b. Knowledge retention rate adaptation adjustment

[0113] Based on knowledge retention factors The review mechanism can be customized.

[0114] if This indicates a low student retention rate. Based on the forgetting curve principle, review nodes can be inserted into the learning path. The specific review intervals can be designed in a progressive manner, for example... In this embodiment, .

[0115] Steps (3.1) and (3.2) combine the queue Q output from step S2 with the knowledge graph G and apply personalized adjustment rules to generate a personalized learning path. .

[0116] S4 matches suitable learning materials for each knowledge point in the personalized learning path.

[0117] This step matches the most suitable learning materials for each knowledge point in the personalized learning path to the student's current ability level. By establishing a standardized feature system for the learning materials and designing corresponding matching algorithms, personalized resource recommendations are achieved.

[0118] like Figure 4 As shown, it includes the following steps:

[0119] (4.1) Feature extraction of learning materials

[0120] Standardize the features of all resources in the learning resource database, construct the learning resource database, and define the resource feature vectors:

[0121]

[0122] in, The level of detail is quantified using a 1-5 scale, with level 1 representing a general overview, level 3 representing a standard explanation, and level 5 representing an extremely detailed analysis. The level of abstraction is quantified using a 1-5 scale. Level 1 represents a fully concrete presentation, Level 3 represents a semi-abstract explanation, such as a presentation combining charts and graphs, and Level 5 represents a pure theoretical derivation. This refers to the number of example problems, recording the number of typical example problems included in the material. Each example problem must include a complete solution process. The level of interaction is quantified using a 1-5 level standard. Level 1 is passive reception of pure text, such as directly asking students to read. Level 3 includes basic interaction, such as setting up multiple-choice tests. Level 5 is highly interactive, such as virtual games or experimental methods for testing or feedback. It is the estimated learning time (in minutes), which is the average time required to master the knowledge point calculated by the system based on historical student learning data.

[0123] For example, in the learning resource library, a learning resource about "function properties" is labeled [2,4,5,1,20], indicating that the resource has a level of detail of 2, an abstraction level of 4, contains 5 examples, an interactive level of 1, and an estimated learning time of 20 minutes.

[0124] (4.2) Personalized matching model

[0125] Define the matching degree function:

[0126]

[0127] in, The learning ability vector is obtained by filtering based on ability-feature matching rules. The set of available learning materials, for ease of description, is further defined as M(k) i ). It is knowledge point k i With data m j The correlation is calculated by mapping knowledge point descriptions and data content to the same semantic space, and the correlation can be calculated using cosine similarity.

[0128] The ability-feature matching rules are used to filter available learning materials for students:

[0129] For students with slow learning speeds: They prefer materials with high levels of detail and numerous examples. For example, if... This indicates that the student's learning speed is slow. We recommend materials with a level of detail of 4-5 and more than 3 examples to help the student fully understand the material through detailed explanations and abundant examples.

[0130] For students with low retention rates: They prefer materials that include multiple solutions and summaries. For example, if... This indicates that students have a low knowledge retention rate. Therefore, we should prioritize recommending materials that include comparisons of multiple solutions and knowledge summaries to enhance long-term memory.

[0131] For students with low application skills: They prefer materials with low abstraction and high interactivity. For example, if... This indicates that the students' application ability is weak. We recommend materials with an abstract level of 1-3 and an interactive level of 4-5 to build their confidence in application through concrete examples and interactive exercises.

[0132] In one specific embodiment of the present invention, the data selection decision-making process is as follows:

[0133] For each knowledge point k in the pathi Perform the following steps:

[0134] Based on the capability-feature matching rules described above, all data matching k are filtered from the database. i The relevant set of available learning materials M(k) i );

[0135] Calculate m for each data set j degree of matching with the current student's learning ability vector ;

[0136] Sort by match from highest to lowest, and select the best matching data at the top: , The final selected matching knowledge point k i The information.

[0137] (4.3) Generation of daily learning tasks

[0138] By combining personalized learning paths with appropriate resource recommendations, specific daily actionable tasks are generated. It should be noted that the knowledge points in the daily actionable tasks are a subset of the knowledge points in the personalized learning paths, and the generation of the next day's actionable tasks is related to the current test results.

[0139] First, obtain the student's available study time budget for the day. (e.g., 120 minutes), strictly following a personalized learning path. The selection process proceeds sequentially, choosing knowledge points one by one to ensure that prerequisite relationships between knowledge points are not disrupted. Starting with the first knowledge point in the current personalized learning path, the estimated learning time for its recommended materials is accumulated. Selection stops when the accumulated time approaches but does not exceed the daily time budget, ensuring that each selected knowledge point has sufficient learning time. The best matching materials are then selected for each chosen knowledge point.

[0140] For example, for knowledge point 1: properties of functions, the best matching material includes a 25-minute detailed video explanation and step-by-step analysis of 3 basic example problems; for knowledge point 2: definition of quadratic functions, the best matching material includes a 15-minute conceptual explanation article and an interactive graphing tool. The path L={properties of functions, definition of quadratic functions} is combined with the resource database, and a matching model is applied to automatically generate daily executable learning tasks.

[0141] This step, through a refined material recommendation and task generation mechanism, ensures that each student can learn the most appropriate content at the right time, in the most suitable way, maximizing learning efficiency and effectiveness.

[0142] S5. The learning effect is tested through questions, the knowledge status is updated, and the learning path and learning material recommendations are dynamically adjusted based on the test results.

[0143] This step involves evaluating the learning outcomes and making real-time, adaptive, and dynamic adjustments to the learning plan based on the results, ensuring that the learning process continues to progress efficiently.

[0144] like Figure 5 As shown, it includes the following steps:

[0145] (5.1) Intelligent Question Push Strategy

[0146] After students complete their daily learning tasks, the assessment process is automatically initiated. The questions pushed to them are not randomly selected, but are calculated in the following multi-dimensional way.

[0147] For each question q in the question bank, calculate a comprehensive score Score(q). Based on three core factors, the comprehensive score function is as follows:

[0148]

[0149] in, It refers to the coverage of knowledge points, used to measure the accuracy of a question q in verifying the target knowledge point k. i The more comprehensive the coverage of the core concepts, the higher the score. It refers to the difficulty level of the questions. It requires students' ability to apply knowledge The closer the difficulty level is to the student's ability, the higher the score for that item. This is to avoid questions that are too difficult and discourage students, or too easy and lose their testing value. Regarding the quality of the questions, this embodiment introduces an expert role to mark some classic or important questions, and obtains a fixed value through a set of preset scoring rules to mark the quality of the questions.

[0150] The number of questions tested is not fixed, but varies depending on the student's ability and the importance of the knowledge points. The calculation formula is as follows:

[0151]

[0152] in, It is knowledge point k i The betweenness centrality is used to assess the importance of knowledge points. The more critical and pivotal a knowledge point is, the higher its betweenness centrality calculated based on the knowledge graph topology, and the more questions it is assigned to ensure that core knowledge is fully tested. It is a learning speed factor. Students who learn at a slower pace will be assigned slightly more questions to provide more sufficient practice and a more stable assessment basis; conversely, students who learn at a faster pace will have their number of questions appropriately reduced to improve efficiency.

[0153] By selecting questions from the question bank that target specific knowledge points and have the highest overall scores. These questions form the set of test questions for that session.

[0154] (5.2) Analysis of test results and update of mastery

[0155] After students complete the questions, their knowledge status is immediately analyzed and updated. The historical mastery is combined with the current test results to recalculate the probability of mastering the knowledge points.

[0156]

[0157] in, It is knowledge point k i Updated mastery probability, This is the historical weighting coefficient, typically ranging from 0.3 to 0.5; It is knowledge point k i Probability of mastery before the update It is to check the number of correct answers. This is the number of questions tested.

[0158] (5.3) Knowledge state tag update

[0159] Based on the updated mastery probability The color status markers for knowledge points are updated using a unified and clear standard:

[0160] if This knowledge point k i Marked in green, indicating mastery; if : Marked in yellow, indicating partial mastery; if Marked in red, indicating that it is not yet mastered.

[0161] The knowledge state matrix is ​​updated synchronously according to the method in step S1.

[0162] (5.4) Dynamic learning path and strategy adjustment

[0163] Based on the latest color-coding and mastery probabilities, the learning plan is automatically adjusted dynamically, with strategy optimization based on the results of individual knowledge point tests and the overall learning trend.

[0164] The immediate optimization process based on the results of individual knowledge point verification is as follows:

[0165] If knowledge point k i If the knowledge point is mastered well (marked in green), it is determined that the learning objective has been achieved, and the knowledge point is removed from the learning path. This knowledge point will not appear in the learning tasks of the next day.

[0166] If knowledge point k iIf you have partially mastered the knowledge point (marked in yellow), it will remain in your learning path, but its priority will be lowered. Your personalized learning path will be updated as a result of the reduced priority of the knowledge point, and the knowledge point will reappear in the next day's learning tasks or the learning tasks after the next day, depending on its position in the updated personalized learning path.

[0167] If knowledge point k i If the knowledge point is not mastered (marked in red), initiate the reinforcement learning process: First, identify the specific weak areas, such as analyzing incorrect answers to determine the specific reasons, such as misunderstanding of concepts, calculation errors, or improper application. Next, review the prerequisite knowledge points to determine if the difficulty in learning the current knowledge point is due to a weak grasp of those prerequisites. If so, retain the knowledge point in the learning path but lower its priority, and automatically reinsert related prerequisite knowledge points into the personalized learning path; if not, simply retain the knowledge point in the learning path but lower its priority, without considering its related prerequisite knowledge points. Finally, reinforce the learning materials for this knowledge point (k). i For subsequent learning, we recommend more detailed and foundational materials, such as increasing the level of detail or the number of examples.

[0168] The adjustment process based on overall learning trends is as follows:

[0169] If a student's performance on multiple consecutive knowledge points (e.g., three) is deemed unsatisfactory, the current learning strategy may not be suitable for that student. The system will then automatically increase the detail and number of examples in subsequent recommended learning materials to provide more comprehensive explanations and practice. Furthermore, it will insert general guidance materials on learning techniques and thinking methods into the learning path. For example, when automatically increasing the detail and number of examples in subsequent recommended learning materials, the following rules can be used to increase the detail and number of examples for the knowledge points to be learned in the current personalized learning path: , ( ).

[0170] In summary, after each test, if all knowledge points k of the learning task for that day... i If you have mastered it well (marked in green), then continue to generate the next day's learning task according to the daily learning task generation method in step (4.3);

[0171] If the daily learning task detects that a knowledge point is partially mastered (marked in yellow) or not mastered (marked in red), then... i First, update the personalized learning path according to the above optimization strategy. Then, based on the order of the knowledge points to be learned in the updated personalized learning path, generate the next day's learning task that conforms to the above optimization strategy according to the method in step S4.

[0172] In this embodiment, a method for updating a personalized learning path is as follows: According to the above optimization strategy, the priority of knowledge points marked in yellow and red in the daily learning task is reduced (based on the previous priority calculation), while the priority of the remaining knowledge points in the current personalized learning path remains unchanged. An initial learning queue Q is obtained based on the priority of all knowledge points to be learned, sorted by priority score. Then, the method of step S3 is run again to generate a new personalized learning path. It should be noted that the knowledge points marked in red in the daily test need to have their prerequisite knowledge points added back to the new personalized learning path.

[0173] In this embodiment, the method for generating the next day's learning task according to step S4 is as follows: Based on the new personalized learning path, the latest knowledge state matrix, and the learning ability vector, the available data set for the knowledge points to be learned is re-matched using the personalized matching model in step (4.2), and the above-mentioned optimization strategy rules are introduced in addition: knowledge points marked in red in the daily test need to be recommended with more detailed and basic information than the matching results; for cases where the test results of multiple consecutive knowledge points in the daily test are not ideal, all knowledge points need to be recommended with more detailed and basic information than the matching results. Based on the finally matched information, the next day's learning task is generated according to the daily learning task generation method in step (4.3).

[0174] For example, after completing the daily task containing knowledge points A (function definition), B (function properties), and C (linear functions), the test results show that knowledge point A is correct 4 / 4, with mastery increasing from 0.6 to 0.8; knowledge point B is correct 2 / 4, with mastery decreasing from 0.4 to 0.3; and knowledge point C is correct 3 / 4, with mastery increasing from 0.5 to 0.65. After the color update, knowledge point A is green, knowledge point B is red, and knowledge point C is yellow. The priority of the knowledge points marked yellow and red in the daily learning task is reduced (based on the previous priority calculation), while the priority of the remaining knowledge points in the current personalized learning path remains unchanged. An initial learning queue Q=[knowledge point B, knowledge point C, knowledge point E] is obtained based on the priority of all knowledge points to be learned, sorted by priority score. Then, step S3 is run again to generate a new personalized learning path [knowledge point B, knowledge point D, knowledge point C, knowledge point E], where knowledge point D is a newly added dependent knowledge point. New recommended materials are then generated for each knowledge point.

[0175] It should be noted that the learning strategy recommendation method based on knowledge graphs and dynamic evaluation in the above embodiments can essentially be executed by a computer program. Therefore, similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer electronic device corresponding to the method provided in the above embodiments, which includes a memory and a processor;

[0176] The memory is used to store computer programs;

[0177] The processor is configured to implement the learning strategy recommendation method based on knowledge graph and dynamic evaluation in the above embodiments when executing the computer program.

[0178] From a hardware perspective, such as Figure 6 The diagram shown is a hardware structure diagram provided in this embodiment. In addition to the processor, memory, network interface and non-volatile memory shown in the diagram, any device with data processing capabilities in the embodiment may also include other hardware depending on the actual function of the device with data processing capabilities, which will not be described in detail here.

[0179] When the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium.

[0180] Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer-readable storage medium corresponding to the method provided in the above embodiments. The storage medium stores a computer program, which, when executed by a processor, can realize the learning strategy recommendation method based on knowledge graph and dynamic evaluation in the above embodiments.

[0181] It is understood that the computer-readable storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device of any data processing device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units and external storage devices of any data processing device. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0182] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A method for recommending a learning strategy based on a knowledge graph and dynamic evaluation, characterized in that, The method comprises the following steps: Step 1, periodically or based on trigger conditions, evaluating a learning ability vector and a comprehensive knowledge state matrix of a student, the learning ability vector comprising a learning speed factor, a knowledge retention factor and a knowledge application ability factor; Step 2, determining a target knowledge point set based on a preset learning goal, calculating a priority score of each knowledge point in the set in combination with the learning ability vector and the comprehensive knowledge state matrix, and generating an initial learning queue according to the priority score; Step 3, performing a prerequisite knowledge point integrity check and topological sorting on the initial learning queue, generating a basic learning path, and performing individualized adjustment on the basic learning path based on the learning ability vector to obtain an individualized learning path; Step 4, for each knowledge point in the individualized learning path, matching adaptive learning materials conforming to the learning ability vector from a learning database, and generating a daily learning task based on a daily learning time budget; Step 5, after completing the daily learning task, pushing adaptive test questions, updating the comprehensive knowledge state matrix according to the test results, and performing an immediate adjustment strategy based on a single knowledge point test result and an overall adjustment strategy based on a learning trend; generating a next-day learning task according to the adjusted strategy; Step 6, repeating Step 5 until the preset learning goal in Step 2 is completed. 2.The knowledge graph and dynamic evaluation based learning strategy recommendation method of claim 1, wherein, The comprehensive knowledge state matrix is a state vector constructed for each knowledge point, comprising a mastery probability of the knowledge point, a betweenness centrality in a knowledge graph, a degree centrality, a time interval since the last learning, and an intention intensity. 3.The knowledge graph and dynamic evaluation based learning strategy recommendation method of claim 2, characterized in that, The priority score is calculated by the following formula: ； wherein, is the priority score of knowledge point k i , is the mastery probability of knowledge point k i , is the betweenness centrality of knowledge point k i , is the time decay coefficient, is the time interval since the last time the student learned knowledge point k i , is the intention amplification coefficient, is the intention intensity of the student towards knowledge point k i , is the ability adaptation function, indicating that when the learning speed factor is below the lower speed threshold, the priority of knowledge points with high betweenness centrality is enhanced. 4.The knowledge graph and dynamic evaluation based learning strategy recommendation method of claim 1, wherein, The prerequisite knowledge point integrity check and topological sorting on the initial learning queue in Step 3 comprise: traversing each knowledge point in the initial learning queue, if the mastery probability of a prerequisite knowledge point of the knowledge point is lower than a mastery threshold and the prerequisite knowledge point is not included in the queue, dynamically inserting the prerequisite knowledge point into the initial learning queue according to the priority score of the prerequisite knowledge point and a prerequisite relationship to obtain an expanded learning queue; initializing an empty ordered path list, and performing the following steps in a loop until the expanded learning queue is empty: taking the first knowledge point from the head of the expanded learning queue; judging whether all prerequisite knowledge points of the knowledge point are ready, the readiness being that the prerequisite knowledge point already exists in the ordered path list or its mastery probability reaches the mastery threshold; if all prerequisite knowledge points are ready, removing the knowledge point from the expanded learning queue and adding it to the end of the ordered path list; if there is a prerequisite knowledge point that is not ready, moving the knowledge point from the head of the expanded learning queue to the tail of the queue; the final ordered path list is taken as the basic learning path. 5.The knowledge graph and dynamic evaluation based learning strategy recommendation method of claim 1, wherein, The process of individualized adjustment on the basic learning path based on the learning ability vector comprises: if the learning speed factor is higher than an upper speed threshold, arranging parallel learning for associated knowledge points in the knowledge graph without direct prerequisite dependency relationship; if the learning speed factor is lower than a lower speed threshold, adopting a strict serial learning strategy and splitting a comprehensive knowledge point into sub-knowledge points; If the knowledge retention factor is lower than the retention threshold, review knowledge points are inserted into the base learning path at preset intervals. 6.The knowledge graph and dynamic evaluation based learning strategy recommendation method of claim 1, wherein, The learning materials in the learning database are characterized by feature vectors, including the level of detail, the level of abstraction, the number of examples, and the level of interaction; and the adaptive learning materials that match the learning ability vector are matched from the learning database according to a preset ability-feature matching rule. 7.The knowledge graph and dynamic evaluation based learning strategy recommendation method of claim 1, wherein, The instant adjustment strategy based on the test result of a single knowledge point includes: If the knowledge point is well mastered after the test, it is removed from the personalized learning path; If the knowledge point is partially mastered, its priority is lowered and the personalized learning path is updated so that it appears in subsequent learning tasks; If the knowledge point is not mastered, a reinforcement learning process is started, including identifying weak links, judging and backtracking the prerequisite knowledge points not mastered, and recommending more detailed or more basic learning materials for the knowledge point. 8.The knowledge graph and dynamic evaluation based learning strategy recommendation method of claim 1, wherein, The overall adjustment strategy based on learning trends includes: If it is detected that the test effect of a student on a plurality of knowledge points is lower than expected, the level of detail and the number of examples of the learning materials recommended for all the knowledge points to be learned in the subsequent learning tasks are automatically increased, and general guidance materials are inserted.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and when the computer program is executed by the processor, the learning strategy recommendation method based on the knowledge graph and dynamic evaluation as claimed in any one of claims 1 to 8 is implemented.

10. A computer electronic device, comprising: comprising a memory and a processor; The memory is configured to store a computer program; The processor is configured to implement the learning strategy recommendation method based on the knowledge graph and dynamic evaluation as claimed in any one of claims 1 to 8 when the computer program is executed.

Images