Learning path optimization method and system based on machine learning

Abstract

The invention discloses a learning path optimization method and system based on machine learning, and the method comprises the steps: collecting the learning data of a learner, and enabling the learning data to comprise a learning object and the information of the learner; establishing a double-layer knowledge graph according to the knowledge type of the learning object; collecting behavior information data of the learner, evaluating a learning style of the learner user according to the behavior information data of the learner, and analyzing and dynamically calculating a learning ability value of the learner and a difficulty value of learning resources in real time according to the learning behavior information data; according to the ability value of the learner, the difficulty value of the learning resource and the information of the learner, generating an optimal learning path by utilizing a double-layer knowledge graph, monitoring the current learning duration, score and participation degree change of the learner in real time, and circularly feeding back and updating the learning path, so that the learning efficiency is improved. And higher-quality and more personalized learning path recommendation is provided for learners.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a learning path optimization method and system based on machine learning. Background Art

[0002] In recent years, with the rapid development of various online learning platforms, massive amounts of text resources and multimedia teaching content have been continuously integrated into digital learning environments, creating an unprecedented space for learners to independently learn. Researchers in this field have conducted a series of exploratory studies, with recommender system technology being widely used to provide personalized learning paths. Existing research focuses on three main dimensions: Personalized matching based on intelligent recommendation algorithms. These algorithms employ collaborative filtering and content-based recommendation algorithms to analyze learners' historical interaction data, establish a learner-resource association matrix, and achieve personalized recommendations for learning objects. These methods offer high response speeds but suffer from reduced recommendation accuracy in data-sparse scenarios. Furthermore, they use knowledge graphs to construct a structured representation of the knowledge system by constructing a domain knowledge ontology and a semantic network of teaching resources. Based on this, the system automatically generates a learning path sequence that conforms to cognitive laws based on the learner's prior knowledge level, ensuring gradual and systematic knowledge acquisition. Leveraging educational data mining techniques, the system collects and analyzes multidimensional learning behavior data in real time. Using Bayesian knowledge tracing methods, it dynamically updates the learner model, enabling immediate optimization of the recommendation strategy.

[0003] Although existing research has made significant progress in personalized recommendation and learning path generation, there are still several key issues that need to be addressed: First, in terms of dynamic adaptive mechanisms, most studies regard learner characteristics and learning resource difficulty as static parameters, and fail to fully consider the dynamic changes in the learning process, resulting in the matching degree between recommendation results and learners' actual needs decreasing over time; Second, in terms of difficulty adaptability, existing systems overly rely on initial ability assessment and lack a dynamic calibration mechanism based on real-time performance data, making it difficult to achieve accurate matching between learning objects and learners' ability levels; Finally, at the path structuring level, faced with a complex curriculum system with multiple knowledge points intersecting and interdisciplinary connections, current methods mainly focus on single resource recommendations or simple sequence combinations, failing to fully explore the topological dependencies between knowledge units, resulting in the generated learning paths lacking systematicity and completeness. Summary of the Invention

[0004] The technical problems solved by the present invention are: treating learner characteristics and learning resource difficulty as static parameters, failing to fully consider the dynamic change characteristics of the learning process, over-reliance on initial ability assessment, lack of a dynamic calibration mechanism based on real-time performance data, difficulty in achieving accurate matching of learning objects and learner ability levels, and failure to fully explore the topological dependency relationship between knowledge units.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, a learning path optimization method based on machine learning comprises:

[0006] Step S100: Collecting the learner's learning data, the learning data includes the learning object and learner information, the learner information includes the learner ID, learning time, test scores, number of learning times, number of tests, learner participation, ratings of learning resources, and focus on learning resources;

[0007] Step S200: establishing a two-layer knowledge graph according to the knowledge type of the learning object;

[0008] Step S300: collecting the learner's behavior information data, evaluating the learner's learning style based on the learner's behavior information data, and dynamically calculating the learner's learning ability value and the difficulty value of the learning resource in real time based on the learning behavior information data;

[0009] Step S400: Based on the learner's ability value, the difficulty value of the learning resource, and the learner information, a double-layer knowledge graph is used to generate the optimal learning path, monitor the learner's current learning time, score, and participation changes in real time, and provide cyclical feedback and update the learning path.

[0010] Preferably, step S100 specifically includes:

[0011] Step S101, collecting raw data of learners, wherein the raw data includes learning objects and uncleaned learner information data;

[0012] Step S102, performing noise screening on the uncleaned learner information data, deleting the uncleaned learner information data whose learning time is less than a threshold, and obtaining learner information;

[0013] Step S103: Match the learner information with the learning object one-to-one to generate learning data:

[0014] The learner information includes learner ID, learning time, test scores, learning times, test times, learner participation, learning resource ratings, and focus on learning resources.

[0015] Preferably, step S200 specifically includes:

[0016] Step S201, classifying the knowledge characteristics of the learning object according to the knowledge type of the learning object, wherein the knowledge characteristics include: knowledge type, difficulty level, dependency relationship and expression form;

[0017] Step S202, classifying the knowledge features to obtain the classified learning objects and corresponding knowledge features;

[0018] Step S203: Using the knowledge features as nodes, the knowledge features are connected in series one-to-one to form relationship edges, thereby obtaining a knowledge feature relationship network diagram; the learning objects are mapped to the corresponding knowledge features to form knowledge relationship edges with weights, thereby generating a learning resource relationship network diagram;

[0019] Step S204: Use the knowledge feature relationship network diagram as the first layer relationship, use the learning resource relationship network diagram as the second layer relationship, and establish a double-layer knowledge graph.

[0020] Preferably, step S300 specifically includes:

[0021] Step S301: collect the learner's behavior information data and obtain the student's learning behavior sequence, which is {x1, x2, ..., x n};

[0022] Step S302: Based on the four-dimensional evaluation system of the Felder-Silverman model, a quantitative learning style evaluation is performed on the learning behavior information data to obtain the student's learning style characteristics;

[0023] Step S303: Real-time analysis and dynamic calculation are performed on the learning data and learning behavior information data to obtain the learner's learning ability value and the difficulty value of the learning resource. The difficulty value of the learning resource is represented by L0.

[0024] Preferably, step S400 specifically includes:

[0025] Search all learning paths and find the optimal path;

[0026] The searching for all learning paths includes: using a depth-first search based on the two-layer knowledge graph and the starting topic and target topic input by the learner to find all paths from the "starting topic" to the "target topic" in the two-layer knowledge graph;

[0027] The optimal path search includes: based on each knowledge point on the path, referring to the current learner's learning ability value, the difficulty value of the learning resources, and the learning style characteristics, generating the optimal or suboptimal path for the learner in a multi-objective optimization manner, providing interactive options, continuously monitoring the learner's new time, score, and participation changes, and dynamically updating the learner's ability, L0, and participation trends.

[0028] Preferably, the dual-layer knowledge graph includes a first relationship layer and a second relationship layer;

[0029] The first relationship layer includes a course concept layer, which represents the knowledge features to which the learning objects belong, with knowledge features as nodes and dependency edges formed between knowledge features;

[0030] The second relationship layer includes a learning object layer, which represents learning objects related to the data features in the upper layer, and associates specific learning resources under the learning object nodes.

[0031] Preferably, the four-dimensional evaluation system includes:

[0032] According to the collected learner behavior sequence {x1, x2, ..., x n}, quantify the probability value of learning behavior, the mode quantization threshold is divided into three levels, the threshold levels are L, M and H, and the learning behavior pattern is preliminarily judged based on the probability value of the learning behavior and the threshold level of the learning behavior. The learning behavior pattern is divided into active or contemplative type, perceptual or intuitive type, visual or verbal type, sequential type or comprehensive type. The style adaptability 1 is further verified based on the quantitative expression, and the quantitative expression is:

[0033]

[0034] Among them, maxS i To maximize the fitness score S of the learning behavior pattern i i , x i is the i-th learning behavior, x i (L), x i (M), x i (H) is the learner's behavioral performance data at the low, medium, and high difficulty levels. The numerator in the formula is the calculation of the Gaussian probability density, where x is the currently observed behavior value, μ is the target value of the ideal behavior, and σ is the standard deviation of the behavior fluctuation. The denominator in the quantitative expression calculates the normalization term, where f(x) i is the probability density function of the ith behavior mode, L and H are the lower and upper limits of the reasonable range of behavior values.

[0035] Preferably, the real-time analysis and dynamic calculation include:

[0036] The learner's learning ability value is A ij , which is used to characterize the ability performance of "the i-th learner learning the j-th L0". The calculation expression of the learning ability value is:

[0037]

[0038] Among them, S ij For the jth learning object LO of the i-th learner ij The score obtained on max为 Its maximum possible value, Iij Its total duration, T max is the expected learning time, R ij is the number of repeated learning, E ij It represents the measure of “learner engagement” or “attention” in the period, and the value is normalized to [0, 1]. α, β, γ, and δ are weight values, and they satisfy α+β+γ+δ=1, α≥0, β≥0, γ≥0, and δ≥0.

[0039] The difficulty value D of the learning resource j is the difficulty of the jth learning object for the learners in the current period. The calculation expression of the difficulty value of the learning resource is:

[0040]

[0041] Among them, D j is the new difficulty value of L0 in the current iteration or current period, and D is the difficulty value of the previous iteration (or initialization). LO ij The average score and average learning time of all learners on For all learners ij The average participation of , μ, λ, θ, φ are weight coefficients, and they satisfy μ≥0, λ≥0, θ≥0, φ≥0.

[0042] Preferably, the depth-first search includes: traversing and searching for a path in the double-layer knowledge graph according to the starting topic (starting knowledge point) and target topic (target knowledge point) input by the learner, finding the target node through logical limitations, and adding the path to the target node to the path result set. The logical limitations include: terminating the search if the path length exceeds a set threshold or the current node has been visited.

[0043] Secondly, the learning path optimization system based on machine learning is implemented, including data acquisition module, knowledge graph construction module, learner modeling module and path optimization and recommendation module;

[0044] The data collection module is used to collect learner behavior data and resource interaction logs in real time;

[0045] The knowledge graph construction module is used to divide the knowledge features of the learning object according to the knowledge type of the learning object, classify the knowledge features, and construct a two-layer graph network;

[0046] The learner modeling module is used to collect the learner's behavior information data, evaluate the learner's learning style based on the learner's behavior information data, and analyze and dynamically calculate the learner's learning ability value and the difficulty value of the learning resources in real time based on the learning behavior information data;

[0047] The path optimization and recommendation module is used to generate the optimal learning path based on the learner's ability value, the difficulty value of the learning resources, and the learner information, using a double-layer knowledge graph, monitor the learner's current learning time, score, and participation changes in real time, and provide cyclical feedback and update the learning path.

[0048] The beneficial effects of the present invention are as follows: by collecting and updating learners' learning time, test scores, participation and resource scores in real time, a three-dimensional learner portrait can be constructed to ensure the accuracy of the assessment of learning ability value and learning resource difficulty value. Based on the collected learning information data and learner portrait, a two-layer knowledge graph is established, and the learner's learning style is analyzed. A multi-objective optimization strategy is adopted to provide learners with comprehensive and intelligent path planning and real-time interactive path selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram of the basic flow of a learning path optimization method and system based on machine learning provided in one embodiment of the present invention.

[0050] Figure 2 A schematic diagram of module generation content of a learning path optimization method and system based on machine learning provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0052] Example 1, with reference to Figure 1 , as one embodiment of the present invention, provides a learning path optimization method based on machine learning, comprising:

[0053] Step S100: Collecting the learner's learning data, which includes the learning object and learner information. The learner information includes the learner ID, learning time, test scores, number of learning times, number of tests, learner participation, ratings of learning resources, and focus on learning resources.

[0054] Step S200: Establish a two-layer knowledge graph based on the knowledge type of the learning object;

[0055] Step S300: Collecting learner behavior information data, evaluating the learner's learning style based on the learner behavior information data, and dynamically calculating the learner's learning ability value and the difficulty value of the learning resource in real time based on the learning behavior information data;

[0056] Step S400: Based on the learner's ability value, the difficulty value of the learning resource, and the learner information, the optimal learning path is generated using a double-layer knowledge graph, and the learner's current learning time, score, and participation changes are monitored in real time, with cyclic feedback and updates to the learning path.

[0057] In this embodiment, by collecting and updating the learner's learning data in real time, analyzing the learner's learning behavior data, and establishing connections between the data, the learner's portrait can be analyzed in three dimensions. Based on the collected learning information data and the learner's portrait, a two-layer knowledge graph is established, and the learner's learning style is analyzed to ensure the accuracy of the assessment of the learning ability value and the learning resource difficulty value. A multi-objective optimization strategy is adopted to provide learners with comprehensive and intelligent path planning and real-time interactive path selection, thereby achieving accurate matching of learning objects and learners' ability levels.

[0058] Step S100 specifically includes:

[0059] Step S101: collecting raw data of learners, including learning objects and uncleaned learner information data;

[0060] Step S102, performing noise screening on the uncleaned learner information data, deleting the uncleaned learner information data with a learning time shorter than a threshold, and obtaining learner information;

[0061] Step S103: Match learner information with learning objects one-to-one to generate learning data:

[0062] Learner information includes learner ID, learning time, test scores, number of learning attempts, number of tests, learner engagement, ratings given to learning resources, and focus on learning resources.

[0063] In this embodiment, the learning time threshold is set to 3, in minutes. The data is denoised by the collected learner information data, and the data with a learning time of less than 3 minutes is deleted to obtain valid learner information. The learner number is used to correspond the learner information and the learning object data one by one, so as to obtain complete learning data, improve the accuracy of learning data analysis, and provide data support for the subsequent establishment of a two-layer knowledge graph.

[0064] Step S200 specifically includes:

[0065] Step S201, classifying the knowledge characteristics of the learning object according to the knowledge type of the learning object, wherein the knowledge characteristics include: knowledge type, difficulty level, dependency relationship and expression form;

[0066] Step S202: Classify the knowledge features according to their content to obtain the classified learning objects and their corresponding knowledge features;

[0067] Step S203: Using knowledge features as nodes, the knowledge features are connected one-to-one to form weighted relationship edges, thereby obtaining a knowledge feature relationship network diagram. Learning objects are mapped to corresponding knowledge features to form knowledge relationship edges, and a learning resource relationship network diagram is generated.

[0068] Step S204: Use the knowledge feature relationship network diagram as the first relationship layer, use the learning resource relationship network diagram as the second relationship layer, and establish a double-layer knowledge graph.

[0069] In this embodiment, the weights of knowledge relationship edges are labeled by experts, and the weight range is 0 to 1. The first layer and the second layer are established by using knowledge features to establish a relationship, and a two-layer knowledge graph is established. Through this method, the logical structuring of knowledge points and the precise matching of resources can be achieved, providing a reliable underlying architecture for subsequent path optimization.

[0070] Step S300 specifically includes:

[0071] Step S301: collect learners' behavior information data and obtain students' learning behavior sequence. The learning behavior sequence is {x1, x2, ..., x n};

[0072] Step S302: Based on the four-dimensional evaluation system of the Felder-Silverman model, a quantitative learning style evaluation is performed on the learning behavior information data to obtain the student's learning style characteristics;

[0073] Step S303: Perform real-time analysis and dynamic calculation on the learning data and learning behavior information data to obtain the learner's learning ability value and the difficulty value of the learning resource. The difficulty value of the learning resource is represented by L0.

[0074] In this embodiment, learning behavior sequences such as the number of times knowledge points are clicked, the duration of video viewing, and the frequency of text viewing are recorded. Compared with the method of only counting the "learning duration", this behavioral data can more truly reflect the learning status; using the Felder-Silverman four-dimensional assessment optimization, the learning style can be scientifically quantified; in addition, using the item response theory, the learner's ability value is analyzed in real time, and timely feedback on the learner's ability value is provided, thereby achieving the effect of dynamic resource utilization.

[0075] Step S400 specifically includes:

[0076] Search all learning paths and find the optimal path;

[0077] Searching all learning paths involves: using a depth-first search based on the two-layer knowledge graph and the starting and target topics input by the learner, to find all paths from the "starting topic" to the "target topic" in the two-layer knowledge graph;

[0078] The optimal path search includes: based on each knowledge point on the path, referring to the current learner's learning ability value, the difficulty value of the learning resources, and the learning style characteristics, the optimal or suboptimal path is generated for the learner in a multi-objective optimization manner, and interactive options are provided. The learner's new time, score, and participation changes are continuously monitored, and the learner's ability, L0, and participation trends are dynamically updated.

[0079] In this embodiment, the learner inputs a linear equation as the starting topic and a quadratic function as the target topic. Through depth-first search, the double-layer knowledge graph is traversed to search for all possible paths. Then, multi-objective optimization is used to dynamically combine the learner's learning ability and resource difficulty to achieve path optimization.

[0080] The two-layer knowledge graph includes the first relationship layer and the second relationship layer;

[0081] The first relationship layer includes the course concept layer, which represents the knowledge features to which the learning objects belong, with knowledge features as nodes and dependency edges between knowledge features;

[0082] The second relationship layer includes a learning object layer, which represents learning objects related to the upper layer data features and associates specific learning resources under the learning object nodes.

[0083] In this embodiment, the weights of knowledge relationship edges are labeled by experts, and the weight range is 0 to 1. In the first layer, mathematical knowledge is used to represent the knowledge feature relationship as follows: integer operation-dependency intensity 0.9-fractional operation. The knowledge features are connected one-to-one to form a knowledge feature relationship network diagram. In the second layer, the learning object is set as [video A, interactive exercise B], and its setting relationship is integer operation-[video A, interactive exercise B]. The integer operation of the knowledge features of the first and second layers is used as exercise to establish a two-layer knowledge graph. Through this method, the logical structuring of knowledge points and the precise matching of resources can be achieved, providing a reliable underlying architecture for subsequent path optimization.

[0084] The four-dimensional assessment system includes:

[0085] According to the collected learner behavior sequence {x1, x2, ..., x n}, quantify the probability value of learning behavior, the pattern quantification threshold is divided into three levels, threshold levels L, M and H. Based on the probability value of learning behavior and the threshold level of learning behavior, the learning behavior pattern is preliminarily judged. The learning behavior pattern is divided into active or contemplative type, perceptual or intuitive type, visual or verbal type, sequential type or comprehensive type. The style adaptability is further verified based on the quantitative expression 1. The quantitative expression is:

[0086]

[0087] Among them, maxS i To maximize the fitness score S of the learning behavior pattern i i , x i is the i-th learning behavior, x i (L), x i (M), x i (H) is the learner's behavioral performance data at the low, medium, and high difficulty levels. The numerator in the formula is the calculation of the Gaussian probability density, where x is the currently observed behavior value, μ is the target value of the ideal behavior, and σ is the standard deviation of the behavior fluctuation. The denominator in the quantitative expression calculates the normalization term, where f(x) i is the probability density function of the ith behavior mode, L and H are the lower and upper limits of the reasonable range of behavior values.

[0088] In this embodiment, L, M, and H represent low, medium, and high difficulty levels, respectively. i It can represent the learning behavior of the number of times a knowledge point is clicked. The quantitative expression is the number of times a knowledge point is clicked ÷ the total number of times × 100%. This value describes the probability density value of the number of times a learner clicks a knowledge point, providing a new quantitative reference for learning behavior for the learner's learning style.

[0089] Real-time analysis and dynamic calculation include:

[0090] The learner's learning ability value is A ij , which is used to characterize the ability performance of "the i-th learner learning the j-th L0". The calculation expression of the learning ability value is:

[0091]

[0092] Among them, S ij For the jth learning object LO of the i-th learner ij The score obtained on max is its maximum possible value, T ij Its total duration, T max is the expected learning time, R ij is the number of repeated learning, E ijIt represents the measure of “learner engagement” or “attention” in the period, and the value is normalized to [0, 1]. α, β, γ, and δ are weight values, and they satisfy α+β+γ+δ=1, α≥0, β≥0, γ≥0, and δ≥0.

[0093] Difficulty value D of learning resources j is the difficulty of the jth learning object for the learners in the current period. The calculation expression of the difficulty value of the learning resource is:

[0094]

[0095] Among them, D j is the new difficulty value of L0 in the current iteration or current period, and D is the difficulty value of the previous iteration (or initialization). LO ij The average score and average learning time of all learners on For all learners ij The average participation of , μ, λ, θ, φ are weight coefficients, and they satisfy μ≥0, λ≥0, θ≥0, φ≥0.

[0096] In this embodiment, μ, λ, θ, and φ are set to 0.4, 0.3, 0.2, and 0.1 respectively, and the learner's ability value A ij It is used to characterize the ability of the i-th learner to learn the j-th learning object. The specific calculation method is to make a weighted calculation of the learner's learning score, learning time, number of repeated attempts and participation; the difficulty value of the learning resource D j It is used to describe the difficulty of the j-th learning object for the learners in the current period. The calculation method is to make a weighted calculation of the difficulty value of the previous iteration (or initialization), the learners' average learning time, average learning score and participation. ij The value reflects the dynamic change of learners from "extremely unengaged" to "highly engaged". ij When the value is very low, it means that the learner is distracted or low-investment. In subsequent recommendations, the difficulty level can be appropriately reduced, or more interactive resources can be recommended to increase investment. The higher the value, the stronger the participation or attention.

[0097] Depth-first search includes: traversing and searching for paths in the double-layer knowledge graph based on the starting topic (starting knowledge point) and target topic (target knowledge point) input by the learner, finding the target node through logical restrictions, and adding the path to the target node to the path result set. The logical restrictions include: if the path length exceeds the set threshold or the current node has been visited, the search is terminated.

[0098] In this embodiment, the input starting knowledge point is a linear equation, the input target knowledge point is a quadratic function image, and the depth threshold is set to 10. The following is the traversal path:

[0099] Path 1: Linear equation → Steps to solve the equation → Function concept → Quadratic function graph;

[0100] Path 2: Linear equation → Algebraic transformation → Quadratic function graph;

[0101] Path 3: Linear equation → Steps to solve the equation → Function concept → Algebraic transformation → ... (Terminated when the maximum depth is exceeded)

[0102] The final output is ["linear equation", "steps to solve the equation", "function concept", "quadratic function image"], ["linear equation", "algebraic expression transformation", "quadratic function image"].

[0103] Example 2, reference Figure 2 This is another embodiment of the present invention. Different from the first embodiment, this embodiment provides a learning path optimization system based on machine learning. In order to verify and illustrate the technical effects adopted in this method, this embodiment adopts traditional technical solutions and the method of the present invention for comparative testing, and compares the test results by means of scientific demonstration to verify the real effect of this method.

[0104] A learning path optimization system based on machine learning, including a data acquisition module, a knowledge graph construction module, a learner modeling module, and a path optimization and recommendation module;

[0105] Data collection module, used to collect learner behavior data and resource interaction logs in real time;

[0106] The knowledge graph construction module is used to divide the knowledge features of the learning object according to the knowledge type of the learning object, classify the knowledge features, and construct a two-layer graph network;

[0107] The learner modeling module is used to collect learner behavior information data, evaluate the learner user's learning style based on the learner behavior information data, and analyze and dynamically calculate the learner's learning ability value and the difficulty value of learning resources in real time based on the learning behavior information data;

[0108] The path optimization and recommendation module is used to generate the optimal learning path based on the learner's ability value, the difficulty value of the learning resources, and the learner information, using a double-layer knowledge graph, to monitor the learner's current learning time, score, and participation changes in real time, and to provide cyclical feedback and update the learning path.

[0109] In this embodiment, by combing the adaptive learning system, the entire system implementation process is divided into five modules, which more clearly demonstrates the knowledge graph and real-time learning analysis to provide learners with an effective learning path and provide effective reference for real-time personnel.

[0110] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a fully hardware embodiment, a fully software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium may be implemented by any type of volatile or non-volatile storage device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0111] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A learning path optimization method based on machine learning, characterized in that: include: Step S100: Collecting the learner's learning data, the learning data includes the learning object and learner information, the learner information includes the learner ID, learning time, test scores, number of learning times, number of tests, learner participation, ratings of learning resources, and focus on learning resources; Step S200: establishing a two-layer knowledge graph according to the knowledge type of the learning object; Step S300: collecting the learner's behavior information data, evaluating the learner's learning style based on the learner's behavior information data, and dynamically calculating the learner's learning ability value and the difficulty value of the learning resource in real time based on the learning behavior information; Step S400: Based on the learner's ability value, the difficulty value of the learning resource, and the learner information, a double-layer knowledge graph is used to generate the optimal learning path, monitor the learner's current learning time, score, and participation changes in real time, and provide cyclical feedback and update the learning path.

2. The learning path optimization method based on machine learning according to claim 1, characterized in that: Step S100 specifically includes: Step S101, collecting raw data of learners, wherein the raw data includes learning objects and uncleaned learner information data; Step S102, performing noise screening on the uncleaned learner information data, deleting the uncleaned learner information data whose learning time is less than a threshold, and obtaining learner information; Step S103: Match the learner information with the learning object one-to-one to generate learning data: The learner information includes learner ID, learning time, test scores, learning times, test times, learner participation, learning resource ratings, and focus on learning resources.

3. The learning path optimization method based on machine learning according to claim 2, characterized in that: Step S200 specifically includes: Dividing the knowledge features of the learning objects, classifying the knowledge features to obtain the classified learning objects and their corresponding knowledge features, using the knowledge features as nodes, and connecting the knowledge features one-to-one to form relationship edges, mapping the learning objects to the corresponding knowledge features, and generating a graph network representing the relationship between knowledge structure and content; Step S201, classifying the knowledge characteristics of the learning object according to the knowledge type of the learning object, wherein the knowledge characteristics include: knowledge type, difficulty level, dependency relationship and expression form; Step S202, classifying the knowledge features to obtain the classified learning objects and corresponding knowledge features; Step S203: Using the knowledge features as nodes, the knowledge features are connected in series one-to-one to form relationship edges, thereby obtaining a knowledge feature relationship network diagram; the learning objects are mapped to the corresponding knowledge features to form knowledge relationship edges with weights, thereby generating a learning resource relationship network diagram; Step S204: Use the knowledge feature relationship network diagram as the first layer relationship, use the learning resource relationship network diagram as the second layer relationship, and establish a double-layer knowledge graph.

4. The learning path optimization method based on machine learning according to claim 3, characterized in that: Step S300 specifically includes: Step S301: collect the learner's behavior information data and obtain the student's learning behavior sequence, which is {x1, x2, ..., x n }; Step S302: Based on the four-dimensional evaluation system of the Felder-Silverman model, a quantitative learning style evaluation is performed on the learning behavior information data to obtain the student's learning style characteristics; Step S303: Real-time analysis and dynamic calculation are performed on the learning data and learning behavior information data to obtain the learner's learning ability value and the difficulty value of the learning resource. The difficulty value of the learning resource is represented by LO.

5. The learning path optimization method based on machine learning according to claim 4, characterized in that: Step S400 specifically includes: searching all learning paths and finding the optimal path; The searching for all learning paths includes: using a depth-first search based on the two-layer knowledge graph and the starting topic and target topic input by the learner to find all paths from the "starting topic" to the "target topic" in the two-layer knowledge graph; The optimal path search includes: based on each knowledge point on the path, referring to the current learner's learning ability value, the difficulty value of the learning resources, and the learning style characteristics, generating the optimal or suboptimal path for the learner in a multi-objective optimization manner, providing interactive options, continuously monitoring the learner's new time, score, and participation changes, and dynamically updating the learner's ability, L0, and participation trends.

6. The learning path optimization method based on machine learning according to claim 5, characterized in that: The dual-layer knowledge graph includes a first relationship layer and a second relationship layer; The first relationship layer includes a course concept layer, which represents the knowledge features to which the learning objects belong, with knowledge features as nodes and dependency edges formed between knowledge features; The second relationship layer includes a learning object layer, which represents learning objects related to the upper layer data features and associates specific learning resources under the learning object nodes.

7. The learning path optimization method based on machine learning according to claim 1, characterized in that: The four-dimensional evaluation system includes: according to the collected learner behavior sequence {x1, x2, ..., x n }, quantify the probability value of learning behavior, the mode quantization threshold is divided into three levels, the threshold levels are L, M and H, and the learning behavior pattern is preliminarily judged based on the probability value of the learning behavior and the threshold level of the learning behavior. The learning behavior pattern is divided into active or contemplative type, perceptual or intuitive type, visual or verbal type, sequential type or comprehensive type. The style adaptability 1 is further verified based on the quantitative expression, and the quantitative expression is: Among them, maxS i To maximize the fitness score S of the learning behavior pattern i i , x i is the i-th learning behavior, x i (L), x i (M), x i (H) is the behavioral performance data of learners at three difficulty levels: low, medium, and high. The numerator in the formula is the calculation of Gaussian probability density, where x is the currently observed behavior value, μ is the target value of the ideal behavior, σ is the standard deviation of the behavior fluctuation, and the denominator in the quantitative expression calculates the normalization term, where f(x) i is the probability density function of the ith behavior mode, L and H are the lower and upper limits of the reasonable range of behavior values.

8. The learning path optimization method based on machine learning according to claim 1, wherein: The real-time analysis and dynamic calculation include: the learner's learning ability value is A ij , which is used to characterize the ability performance of "the i-th learner learning the j-th LO". The calculation expression of the learning ability value is: Among them, S ij For the jth learning object LO of the i-th learner ij The score obtained on max is its maximum possible value, T ij Its total duration, T max is the expected learning time, R ij is the number of repeated learning, E ij represents the measure of "learner engagement" or "attention" during the period. The value is normalized to [0, 1], with higher values indicating stronger engagement or attention. α, β, γ, and δ are weights, satisfying α+β+γ+δ=1, α≥0, β≥0, γ≥0, and δ≥0. The difficulty value D of the learning resource j is the difficulty of the jth learning object for the learners in the current period. The calculation expression of the difficulty value of the learning resource is: Among them, D j is the new difficulty value of L0 in the current iteration or current period, and D is the difficulty value of the previous iteration (or initialization). LO ij The average score and average learning time of all learners on E ij For all learners ij The average participation of , μ, λ, θ, φ are weight coefficients, and they satisfy μ≥0, λ≥0, θ≥0, φ≥0.

9. The learning path optimization method based on machine learning according to claim 1, wherein: The depth-first search includes: traversing and searching for a path in the two-layer knowledge graph based on the starting topic and target topic input by the learner, finding the target node through logical limitations, and adding the path to the target node to the path result set. The logical limitations include: terminating the search if the path length exceeds a set threshold or the current node has been visited.

10. A learning path optimization system based on machine learning, which is based on the learning path optimization system based on machine learning according to any one of claims 1 to 9, characterized in that: It includes data collection module, knowledge graph construction module, learner modeling module and path optimization and recommendation module; The data collection module is used to collect learner behavior data and resource interaction logs in real time; The knowledge graph construction module is used to divide the knowledge features of the learning object according to the knowledge type of the learning object, classify the knowledge features, and construct a two-layer graph network; The learner modeling module is used to collect the learner's behavior information data, evaluate the learner's learning style based on the learner's behavior information data, and analyze and dynamically calculate the learner's learning ability value and the difficulty value of the learning resources in real time based on the learning behavior information data; The path optimization and recommendation module is used to generate the optimal learning path based on the learner's ability value, the difficulty value of the learning resources, and the learner information, using a double-layer knowledge graph, monitor the learner's current learning time, score, and participation changes in real time, and provide cyclical feedback and update the learning path.

Images