A resource learning path planning method and device based on deep learning

Abstract

The application discloses a resource learning path planning method and device based on deep learning, which comprises the following steps: acquiring the knowledge level and historical record information of multiple learners, constructing a learner model according to the historical record information and the knowledge level, calculating the learner similarity between any two different learners in the learner model, constructing a learning behavior model corresponding to the learners according to the learner similarity, predicting the cognitive level of the learners according to the learning behavior model and the learner model, constructing a learning path recommendation model according to the cognitive level, planning a target learning path corresponding to the learners according to the learning path recommendation model and a preset knowledge graph, quantifying the learning behavior index of the learning behavior of the learners themselves to construct the learner model, recommending the learning path to the learners, and further completing the accurate learning path planning in combination with the knowledge graph, so that a practical, reasonable and interpretable learning path is provided for the learners, and the learning efficiency and quality of the learners are improved.

Description

Technical Field

[0001] This invention belongs to the field of education, and in particular relates to a resource learning path planning method and apparatus based on deep learning. Background Technology

[0002] The rapid development of information technology is a major driving force behind the development of online learning. Artificial intelligence is profoundly changing the way people learn, and web-based learning has experienced rapid development and widespread application. A learning path refers to an ordered sequence of courses and knowledge points that a learner studies. After setting learning goals, learners need to develop corresponding learning plans to complete learning tasks in a certain order. Because different courses have different characteristics, corresponding teaching strategies are used to assist specific teaching tasks in different courses, helping learners achieve their learning goals. Typically, a learning path can be a sequence of learning multiple related courses for systematically learning a particular area of ​​knowledge, or a sequence of learning multiple related knowledge points for learning a specific course. In the field of education, the relationships between courses and between courses and knowledge points are relatively few, so manual labeling can be used. However, with the popularization of internet education, the number of questions on online education platforms is constantly increasing, but the relationships between questions and courses / knowledge points are sparse. Many questions only have the question text but lack the knowledge points they involve, making it impossible to accurately and effectively assess learners' learning outcomes, thus reducing their learning efficiency. Summary of the Invention

[0003] In view of this, the present invention provides a resource learning path planning method and apparatus based on deep learning that can take into account the different learning abilities and needs of different learners, improve learners' cognitive level and enhance learning efficiency, in order to solve the above-mentioned technical problems. Specifically, the following technical solutions are adopted to achieve this.

[0004] In a first aspect, the present invention provides a resource learning path planning method based on deep learning, comprising:

[0005] Acquire the knowledge level and historical record information of multiple learners, and construct a learner model based on the historical record information and the knowledge level, wherein the knowledge level includes knowledge point ID and mastery level, and the historical record information includes learning sequence, learning resources and assignment data;

[0006] Calculate the learner similarity between any two different learners in the learner model, and construct the learning behavior model corresponding to each learner based on the learner similarity. Each learner is represented as a vector in the learner model. Cosine similarity is used to evaluate the similarity between two vectors; that is, the cosine of the angle between the two learner vectors is taken as the similarity score between the learners. The expression for the similarity score is: Where u a Let u be the vector representation of learner a. b The vector representation of learner b;

[0007] The learner's cognitive level is predicted based on the learning behavior model and the learner model, and a learning path recommendation model is constructed based on the cognitive level. The learning path recommendation model includes constructing two state sets. These represent whether the learner accepts the recommendation, with C indicating acceptance of the recommendation. This indicates that recommendations are not accepted.

[0008] Based on the learning path recommendation model and the preset knowledge graph, the learner's target learning path is planned.

[0009] As a further improvement to the above technical solution, the learning path recommendation model and the preset knowledge graph are used to plan the target learning path corresponding to the learner, including:

[0010] Obtain all knowledge points in the preset knowledge graph as a set of knowledge points to be sorted, and preset an initial knowledge point KP0;

[0011] Export list L from the database corresponding to the preset knowledge graph to determine the relationships between knowledge points. These relationships include inclusion and parallel relationships between knowledge points. The record information in list L is (KP) m ,R,KP n ) represents knowledge point KP m With KP n The relation is R, and the knowledge point relation includes knowledge point importance, difficulty, centrality, topological hierarchy attribute feature data, and knowledge point ranking index;

[0012] The target learning path is generated using a topological sorting algorithm based on a pre-defined knowledge graph and knowledge point attribute feature data.

[0013] As a further improvement to the above technical solution, a topological sorting algorithm is used to generate a target learning path based on a preset knowledge graph and knowledge point attribute feature data, including:

[0014] The target learning path is compared with the preset expert path to obtain the path similarity. The expression for calculating the path similarity is as follows: Where pathep path at These represent the preset expert path and the target learning path, respectively, KP mat KP represents the number of learning paths that match the preset expert path with the target learning path. tot Indicates the total number of learning paths;

[0015] The quality of the target learning path is evaluated using the learning path assessment metric "fitness". The formula for calculating the fitness metric is as follows: Among them, pen adj Pen represents the number of learning paths that violate the principle of adjacency between knowledge points. ord This indicates the number of times the learning path violates the principle of prioritizing knowledge points. The fewer the number of violations, the lower the fitness value, indicating a higher quality path generation.

[0016] As a further improvement to the above technical solution, predicting the learner's cognitive level based on the learning behavior model and the learner model includes:

[0017] The learner's cognitive level is divided into three domains: the positive domain (POS), the boundary domain (BND), and the negative domain (NEG). The learner's cognitive level θ is determined according to Item Response Theory (2PL). The 2PL model uses item characteristic curves to describe the probability of the learner answering a pre-set question correctly. Where f represents the difficulty coefficient of the question, f∈[-2,2], d represents the constant 1.072, θ represents the ability value, i.e. the learner's cognitive level, p(θ) represents the probability of answering the test question correctly, POS field represents learners with high cognitive level, NEG field represents learners with low cognitive level, and BND field represents learners with medium cognitive level.

[0018] Input the test-taking results data from the obtained score matrix and the known question difficulty coefficients into the IRT model, and establish the expression for the maximum likelihood function of the ability parameters as follows: Where p i y represents the probability of a correct answer obtained from the project response theory model function. i Let the labels in the score matrix represent the learners' true responses, thus yielding the expression for the log-maximum likelihood function. The capability parameter θ is obtained by differentiating the log-maximal likelihood function.

[0019] As a further improvement to the above technical solution, the decision information given for each region is A = {α}. P ,α B ,α N} represent learning path recommendation information for learners at different cognitive levels, α P This indicates that you have completed the learning task for the current knowledge point and are now accepting recommendations for learning other knowledge points. N This indicates a low current level of cognitive ability, suggesting a return to reinforce basic knowledge. (α) B This indicates acceptance of the recommended exercises to improve cognitive abilities;

[0020] The expression for the learning utility function matrix is: Where λ PP , λ BP and λ NP When an object belongs to C, the behavior α is taken. P α B and α N The corresponding loss value, λ PN , λ BN and λ NN When the object does not belong to C, take action α. P α B and α N The corresponding loss value, s(λ) PP ), s(λ BP ) and s(λ NP ) indicates the action α taken when the object belongs to C. P α B and α N The corresponding utility value, s(λ) PN ), s(λ BN ) and s(λ NN ) indicates the action to be taken when the object does not belong to C. P α B and α N The corresponding utility value; s(λ) PP )≥s(λ BP )≥s(λ NP ), representing an object In this state, the utility of classifying [x] as a positive domain is greater than the utility of classifying it as a boundary domain, and thus greater than the utility of classifying it as a negative domain; s(λ PN )≤s(λ BN )≤s(λ NN ), representing an object In this state, the utility of classifying [x] as a negative domain is less than the utility of classifying it as a boundary domain, and thus less than the utility of classifying it as a negative domain.

[0021] As a further improvement to the above technical solution, a list L is exported from the database corresponding to the preset knowledge graph to determine the relationships between knowledge points, including:

[0022] A keyword extraction algorithm is used to process text resources to obtain a set of knowledge points and the importance of each knowledge point, expressed as follows: Where Ws(V i ) represents node V i The weights, Ws(V) j ) represents node V after the last iteration. j The weight, ω ji Represents node V j With node V i The similarity between them, where g represents the damping coefficient;

[0023] Based on the properties of knowledge graphs, the centrality of a knowledge point is calculated using the ratio of its in-degree to its out-degree. A higher ratio indicates a higher centrality. The expression for the in-degree and out-degree of a knowledge point is as follows: The expression for calculating centrality is: in Represents node V i The ratio of in-degree to out-degree, Pre(V) i ) represents knowledge point V i The set of first-order precursor knowledge points, Suc(V) i ) represents knowledge node V i The set of first-order successor knowledge points;

[0024] The above attribute values ​​are used to calculate the knowledge point ranking index. Among them, importance Imp kpi Using the above expression Ws(V i The expression for obtaining the weight value of a node is: Imp kpi Diff kpi Cent kpi and Tp kpi These represent the importance, difficulty, centrality, and topological level of a knowledge point's KPI, respectively, ω imp ω diff ω cent and ω tp This indicates the attribute weight value assigned.

[0025] As a further improvement to the above technical solution, a learning path recommendation model is constructed based on the cognitive level, including:

[0026] Preset T = [i1, i2...i m ], where T represents the set of knowledge points and i represents a single knowledge point, then Where L represents the learner path sequence, and ls represents a single learner. This represents the learning path of learner ls1.

[0027] As a further improvement to the above technical solution, relevant knowledge points are tracked through learners' questions and the questions are categorized. The set is predefined as follows: i' = 1, 2...M represents M question texts, j' = 1, 2...N represents N pre-defined knowledge points, and the classification matrix is ​​W = w i′j′ , where w i′j′ This represents the relationship between the i'th question text and the j'th knowledge point.

[0028] The formula for calculating the ratio of the number of nodes completed to the number of nodes where a particular learning objective is located, as recommended to learners, is as follows: Where n represents the total number of nodes in the learning path, and i0 represents the total number of nodes actually recommended. When the path value is smaller, the recommendation result is more satisfactory and in line with expectations; when the path value is larger, that is, closer to 1, the recommendation result is less in line with expectations.

[0029] As a further improvement to the above technical solution, acquiring the knowledge level and historical record information of multiple learners, and constructing a learner model based on the historical record information and the knowledge level, includes:

[0030] Based on the obtained learner behavior data, determine the cluster value k for similar learners, and calculate the expression for the distance between the coordinates of all learners and the k seed points as follows: Where Y represents the learner model;

[0031] Compare the distances between each learner's coordinates and each seed point. If the distance between learner coordinates G and seed point Kn is the smallest, then learner coordinates G belongs to learner cluster Kn.

[0032] Move the seed point to the center of its cluster, minimizing the sum of the coordinate distances between the seed point Kn and all learners belonging to that learner cluster. Repeat this process until the seed point stops moving.

[0033] Secondly, the present invention also provides a resource learning path planning device based on deep learning, comprising:

[0034] The data acquisition module is used to acquire the knowledge level and historical record information of multiple learners, and to construct a learner model based on the historical record information and the knowledge level. The knowledge level includes knowledge point ID and mastery level, and the historical record information includes learning sequence, learning resources and assignment data.

[0035] The first construction module is used to calculate the learner similarity between any two different learners in the learner model, and to construct the learning behavior model corresponding to each learner based on the learner similarity. Each learner is represented as a vector in the learner model, and the similarity between two vectors is evaluated using cosine similarity, i.e., the cosine of the angle between the two learner vectors is used as the similarity score between the learners. The expression for the similarity score is: Where u a Let u be the vector representation of learner a. b The vector representation of learner b;

[0036] The second construction module is used to predict the learner's cognitive level based on the learning behavior model and the learner model, and to construct a learning path recommendation model based on the cognitive level. The learning path recommendation model includes constructing two state sets. These represent whether the learner accepts the recommendation, with C indicating acceptance of the recommendation. This indicates that recommendations are not accepted.

[0037] The path planning module is used to plan the target learning path corresponding to the learner based on the learning path recommendation model and the preset knowledge graph.

[0038] This invention provides a resource learning path planning method and apparatus based on deep learning. It acquires the knowledge level and historical record information of multiple learners, constructs a learner model based on the historical record information and the knowledge level, calculates the learner similarity between any two different learners in the learner model, constructs a learning behavior model corresponding to each learner based on the learner similarity, predicts the learner's cognitive level based on the learning behavior model and the learner model, constructs a learning path recommendation model based on the cognitive level, plans the learner's target learning path based on the learning path recommendation model and a preset knowledge graph, quantifies the learner's own learning behavior into learning behavior indicators to construct the learner model, and predicts the learner's cognitive level based on the learner behavior model to accurately improve the learner's cognitive level, provides personalized learning path recommendations to learners, and further completes precise learning path planning by combining with the knowledge graph, providing learners with a practical, reasonable, and explainable learning path, thereby improving learners' learning efficiency and quality. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 A flowchart of the resource learning path planning method based on deep learning provided by the present invention;

[0041] Figure 2 The structural block diagram of the resource learning path planning device based on deep learning provided by the present invention is shown. Detailed Implementation

[0042] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0043] See Figure 1 This invention provides a resource learning path planning method based on deep learning, comprising:

[0044] S1: Obtain the knowledge level and historical record information of multiple learners, and construct a learner model based on the historical record information and the knowledge level, wherein the knowledge level includes knowledge point ID and mastery level, and the historical record information includes learning sequence, learning resources and assignment data;

[0045] S2: Calculate the learner similarity between any two different learners in the learner model, and construct the learning behavior model corresponding to each learner based on the learner similarity. Each learner is represented as a vector in the learner model. Cosine similarity is used to evaluate the similarity between two vectors; that is, the cosine of the angle between the two learner vectors is taken as the similarity score between the learners. The expression for the similarity score is: Where u a Let u be the vector representation of learner a. b The vector representation of learner b;

[0046] S3: Predict the learner's cognitive level based on the learning behavior model and the learner model, and construct a learning path recommendation model based on the cognitive level, wherein the learning path recommendation model includes constructing two state sets. These represent whether the learner accepts the recommendation, with C indicating acceptance of the recommendation. This indicates that recommendations are not accepted.

[0047] S4: Based on the learning path recommendation model and the preset knowledge graph, plan the target learning path corresponding to the learner.

[0048] In this embodiment, planning the target learning path for the learner based on the learning path recommendation model and the preset knowledge graph includes: obtaining a set of knowledge points to be sorted from all knowledge points in the preset knowledge graph, and setting an initial knowledge point KP0; exporting a list L from the database corresponding to the preset knowledge graph to determine the relationships between knowledge points, wherein the relationships between knowledge points include the inclusion relationship and the parallel relationship between knowledge points, and the record information of list L is (KP0). m ,R,KP n ) represents knowledge point KP m With KP n The relationship is R, and the knowledge point relationships include knowledge point importance, difficulty, centrality, topological hierarchy attribute feature data, and knowledge point ranking indicators. A target learning path is generated using a topological ranking algorithm based on the preset knowledge graph and knowledge point attribute feature data. This generation of the target learning path using the topological ranking algorithm includes: comparing the target learning path with preset expert paths to obtain path similarity; the expression for calculating path similarity is... Where path ep path at These represent the preset expert path and the target learning path, respectively, KP mat KP represents the number of learning paths that match the preset expert path with the target learning path. tot This represents the total number of learning paths; the quality of the target learning paths is evaluated using the learning path evaluation metric "fitness". The calculation formula for the fitness metric is as follows: Among them, pen adj Pen represents the number of learning paths that violate the principle of adjacency between knowledge points. ord This indicates the number of times a learning path violates the principle of prioritizing knowledge points. The fewer violations, the lower the fitness score, indicating higher path generation quality. The process involves acquiring the knowledge level and historical records of multiple learners, and constructing a learner model based on these records and knowledge levels. This includes determining the cluster value k for similar learners based on acquired learner behavior data, and calculating the expression for the distance between all learner coordinates and k seed points. Where Y represents the learner model; compare the distances between each learner coordinate and each seed point. If the distance between learner coordinate G and seed point Kn is the smallest, then learner coordinate G belongs to learner cluster Kn; move the seed point so that each sub-point moves to the center of its cluster, so that the sum of the distances between seed point Kn and the coordinates of each learner belonging to the learner cluster is the smallest. Repeat the above process until the seed point no longer moves.

[0049] It should be noted that basic information can include name, student ID, username, password, gender, class, etc. The mastery of individual knowledge points is calculated based on the learner's ability to master that knowledge point after completion. First, a mastery value for each individual knowledge point is obtained through calculation. The calculation expression uses a single-parameter model based on the relevant project theory. Where P ij (θ k Let θ represent learner k's mastery of knowledge point i. k Q is the probability that the j-th question on the i-th knowledge point can be answered correctly. ij (θ k Let θ represent learner k's mastery of knowledge point i. k When the i-th knowledge point is answered incorrectly in the j-th question, θ is the probability. k Through P ij (θ k Perform maximum likelihood estimation when LL(μ1,μ2...μ) n θ at its maximum k The value is used for estimation, and its calculation expression is: Where μ j μ takes the value of 0 or 1. j =1 indicates that learner k answered exercise j under knowledge point i correctly, μ j =0 indicates that learner k answered exercise j under knowledge point i incorrectly, and n represents the number of exercises testing that knowledge point. Then, the Newton-Raphson iteration method is used to iterate through θ. k Different valuations, up to LL(μ1,μ2...μ n To reach the maximum, obtain θ. k For the specific value, the iterative expression for θ is: Where θ t+1 With θ t h is the mastery ability value of a certain knowledge point obtained in the (t+1)th and tth iterations. t This indicates a correction factor for the ability to master knowledge points. It is the first derivative of the log-likelihood function lnLL with respect to θ. The second derivative of the log-likelihood function lnLL with respect to θ. When h... t When θ is sufficiently small or the number of iterations is sufficiently large, the value of θ represents the learner's mastery of the knowledge point. k The value is set to a preset range of [-m, m], where m represents the learner k's best mastery of knowledge point j, 0 represents average mastery, and -m represents the worst mastery. The expression for the learner's inability to master knowledge point i is... γ i This represents the weighted score for each knowledge point; when it exceeds a threshold c... iAt that time, it is considered that the learner has mastered the knowledge point i; otherwise, they have not mastered it.

[0050] It should be understood that constructing a learner model is essential for recommending suitable learning paths to learners. A learning path refers to the sequence of learning various knowledge points in order to master a target knowledge point. Based on the learner's mastery of each knowledge point, the learning objective, and the current learning status, a series of knowledge points are arranged in an orderly manner under the guidance of appropriate learning strategies, ultimately forming a learning path. The learning path reflects the sequential order of learning activities and should be consistent with the order of relationships between knowledge points, conforming to their inherent predecessor and successor relationships. A learning path is a sequence formed by a learning activity, and the knowledge points in the sequence must be coherent, meaning there should be no missing knowledge nodes in the path. Each knowledge point forms a knowledge graph through complex relationships. Knowledge points are connected by directed graph segments, and countless interconnected directed segments form the knowledge graph. The learning path mainly involves the orderly arrangement of knowledge point objects, and the path to the target object is not unique. From knowledge graphs to automatic learning path planning, a solid foundation is laid for the generation of personalized learning paths, the recommendation of learning resources, and even the construction of personalized learning models. The learning path planning is highly feasible and of good quality. In the process of online learning, it can replace the path formulated by experts and provide online learners with a general, reasonable and explainable learning path, thereby improving the learning efficiency and quality of online learners.

[0051] Optionally, predicting the learner's cognitive level based on the learning behavior model and the learner model includes:

[0052] The learner's cognitive level is divided into three domains: the positive domain (POS), the boundary domain (BND), and the negative domain (NEG). The learner's cognitive level θ is determined according to Item Response Theory (2PL). The 2PL model uses item characteristic curves to describe the probability of the learner answering a pre-set question correctly. Where f represents the difficulty coefficient of the question, f∈[-2,2], d represents the constant 1.072, θ represents the ability value, i.e. the learner's cognitive level, p(θ) represents the probability of answering the test question correctly, POS field represents learners with high cognitive level, NEG field represents learners with low cognitive level, and BND field represents learners with medium cognitive level.

[0053] Input the test-taking results data from the obtained score matrix and the known question difficulty coefficients into the IRT model, and establish the expression for the maximum likelihood function of the ability parameters as follows: Where p i y represents the probability of a correct answer obtained from the project response theory model function. iLet the labels in the score matrix represent the learners' true responses, thus yielding the expression for the log-maximum likelihood function. The capability parameter θ is obtained by differentiating the log-maximal likelihood function.

[0054] In this embodiment, the decision information given for each region is A = {α} P ,β B ,α N} represent learning path recommendation information for learners at different cognitive levels, α P This indicates that you have completed the learning task for the current knowledge point and are now accepting recommendations for learning other knowledge points. N This indicates a low current level of cognitive ability, suggesting a return to reinforce basic knowledge. (α) B This indicates acceptance of recommended exercises to improve cognitive level; the expression for the learning utility function matrix is: Where λ PP , λ BP and λ NP When an object belongs to C, the behavior α is taken. P α B and α N The corresponding loss value, λ PN , λ BN and λ NN When the object does not belong to C, take action α. P α B and α N The corresponding loss value, s(λ) PP ), s(λ BP ) and s(λ NP ) indicates the action α taken when the object belongs to C. P α B and α N The corresponding utility value, s(λ) PN ), s(λ BN ) and s(λ NN ) indicates the action to be taken when the object does not belong to C. P α B and α N The corresponding utility value; s(λ) PP )≥s(λ BP )≥s(λ NP ), representing an object In this state, the utility of classifying [x] as a positive domain is greater than the utility of classifying it as a boundary domain, and thus greater than the utility of classifying it as a negative domain; s(λ PN )≤s(λ BN )≤s(λ NN ), representing an object In this state, the utility of classifying [x] as a negative domain is less than the utility of classifying it as a boundary domain, and thus less than the utility of classifying it as a negative domain.

[0055] It should be noted that cognitive level refers to the learner's domain knowledge level, reflecting the learner's learning performance information. The cognitive level model consists of two parts: the target curriculum and the initial level, and its expression is Cog=(Les,<t1,l1> ,<t2,l2> ... <t m ,l m >), where Cog represents the learner's cognitive level, Les represents the learner's target course (i.e., the course the learner wants to learn), and t m Let l represent the m-th learning unit. m Indicates the level of knowledge mastery. <t m ,l m The value > represents a learner's cognitive level regarding the knowledge of the m-th learning unit, primarily obtained through pre-learning testing. Item Response Theory (BND) is used to establish a functional relationship between learners' cognitive levels and item parameters. This relationship can be described using a function of item characteristics, essentially a regression curve of the learner's correct answer probability on an item against latent trait scores. Appropriate models can be selected to estimate parameters based on different response data. Although learners' cognitive levels vary, each learner aims to achieve a higher level of cognition through a certain period of learning, building upon their current level. Learners assigned to the BND domain require a delayed decision-making approach, i.e., reprocessing the BND domain using a three-branch decision boundary domain processing model based on learning utility to obtain personalized learning paths. These personalized learning paths recommend different paths based on the learner's behavioral habits and individual cognitive levels. The personalized learning paths must align with the learner's individual learning habits and enable them to achieve a higher cognitive level through the recommended paths, thereby improving the accuracy of learning path planning.

[0056] Optionally, a list L is exported from the database corresponding to the preset knowledge graph to determine the relationships between knowledge points, including:

[0057] A keyword extraction algorithm is used to process text resources to obtain a set of knowledge points and the importance of each knowledge point, expressed as follows: Where Ws(V i ) represents node V i The weights, Ws(V) j ) represents node V after the last iteration. j The weight, ω ji Represents node V j With node V i The similarity between them, where g represents the damping coefficient;

[0058] Based on the properties of knowledge graphs, the centrality of a knowledge point is calculated using the ratio of its in-degree to its out-degree. A higher ratio indicates a higher centrality. The expression for the in-degree and out-degree of a knowledge point is as follows: The expression for calculating centrality is: in Represents node V i The ratio of in-degree to out-degree, Pre(V) i ) represents knowledge point V i The set of first-order precursor knowledge points, Suc(V) i ) represents knowledge node V i The set of first-order successor knowledge points;

[0059] The above attribute values ​​are used to calculate the knowledge point ranking index. Among them, importance Imp kpi Using the above expression Ws(V i The expression for obtaining the weight value of a node is: Imp kpi Diff kpi Cent kpi and Tp kpi These represent the importance, difficulty, centrality, and topological level of a knowledge point's KPI, respectively, ω imp ω diff ω cent and ω tp This indicates the attribute weight value assigned.

[0060] In this embodiment, a learning path recommendation model is constructed based on the cognitive level, including: a preset T = [i 1, i2...i m ], where T represents the set of knowledge points and i represents a single knowledge point, then Where L represents the learner path sequence, and ls represents a single learner. This represents the learning path of learner ls1. The system tracks relevant knowledge points through learner questions and categorizes the questions. Preset i' = 1, 2...M represents the M question texts in the set, j' = 1, 2...N represents the N pre-defined knowledge points, and the classification matrix is ​​W = w i′j′ , where w i′j′ This represents the relationship between the i'th question text and the j'th knowledge point. The formula for calculating the ratio of the number of nodes completed to the number of nodes where a particular learning objective is located, as recommended to learners, is as follows: Where n represents the total number of nodes in the learning path, and i represents the total number of nodes actually recommended. When the path value is smaller, the recommendation result is more satisfactory and in line with expectations; when the path value is larger, that is, closer to 1, the recommendation result is less in line with expectations.

[0061] It's important to note that learners often confuse the target knowledge points involved in a question, unsure which knowledge point they should actually learn. To quickly trace relevant knowledge points through learners' questions, it's necessary to categorize the questions. For example, first collect text corpora, categorize and store them according to instructions, clean and segment the data, calculate the term frequency-document frequency (TF-IDF) as a weighting factor, and introduce this weighting factor into the conditional probability of the Naive Bayes function. Then, determine the ranking factor from the weighting factor and integrate it into the Bayes function to increase the importance of keywords, thereby reducing the impact of the feature independence assumption of Naive Bayes and improving the system's data processing efficiency to some extent.

[0062] See Figure 2 The present invention also provides a resource learning path planning device based on deep learning, comprising:

[0063] The data acquisition module is used to acquire the knowledge level and historical record information of multiple learners, and to construct a learner model based on the historical record information and the knowledge level. The knowledge level includes knowledge point ID and mastery level, and the historical record information includes learning sequence, learning resources and assignment data.

[0064] The first construction module is used to calculate the learner similarity between any two different learners in the learner model, and to construct the learning behavior model corresponding to each learner based on the learner similarity. Each learner is represented as a vector in the learner model, and the similarity between two vectors is evaluated using cosine similarity, i.e., the cosine of the angle between the two learner vectors is used as the similarity score between the learners. The expression for the similarity score is: Where u a Let u be the vector representation of learner a. b The vector representation of learner b;

[0065] The second construction module is used to predict the learner's cognitive level based on the learning behavior model and the learner model, and to construct a learning path recommendation model based on the cognitive level. The learning path recommendation model includes constructing two state sets. These represent whether the learner accepts the recommendation, with C indicating acceptance of the recommendation. This indicates that recommendations are not accepted.

[0066] The path planning module is used to plan the target learning path corresponding to the learner based on the learning path recommendation model and the preset knowledge graph.

[0067] In this embodiment, a feature-weighted Naive Bayes classifier is used to obtain target knowledge points. The main process of knowledge point acquisition includes: Feature processing: Jieba word segmentation is performed on the text set of each knowledge point category. Each word in each question text under that category is read sequentially. The number of texts h in which the word appears in that category is counted. The frequency r of the word in that category is calculated. The TF-IDF value of the feature word is obtained by multiplying h and r. The feature word and its weight are recorded in the candidate set of the corresponding category. After completion, the next feature word is processed sequentially. Vector matrix formation: The candidate set of text feature values ​​under each knowledge point category is listed in descending order of weight. Feature terms are selected from each question text to form a vector matrix representing the knowledge point. The set of features of the question text is used to construct a dictionary of key words extracted from each question text under all knowledge point categories. The words are arranged according to the order of the dictionary positions. The text involved is represented by a Boolean model, resulting in two matrices: one is the TF-IDF weight factor space vector matrix of all knowledge point categories, and the other is the vector matrix representing the question text under each knowledge point category. To construct the classifier: obtain the above two vector matrices, calculate the TF and TF-IDF under the corresponding knowledge point categories, obtain the Naive Bayes conditional probability vector of each feature item under the relevant knowledge point categories, determine the rank factor vector from the word frequency-document TF-IDF vector under each knowledge point category, and then effectively calculate the posterior probability under each knowledge point category.

[0068] It should be noted that by acquiring the knowledge level and historical records of multiple learners, and constructing learner models based on the historical records and knowledge levels, the similarity between any two different learners in the learner models is calculated. Based on the learner similarity, a learning behavior model corresponding to the learner is constructed. Based on the learning behavior model and the learner model, the learner's cognitive level is predicted, and a learning path recommendation model is constructed based on the cognitive level. Based on the learning path recommendation model and a preset knowledge graph, a target learning path corresponding to the learner is planned. The learner's own learning behavior is quantified into learning behavior indicators to construct the learner model, and the learner's cognitive level is predicted based on the learner behavior model to accurately improve the learner's cognitive level. Personalized learning path recommendations are made to learners, and then combined with the knowledge graph to further complete the accurate learning path planning, providing learners with a practical, reasonable, and explainable learning path, thereby improving learners' learning efficiency and quality.

[0069] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limitations; therefore, other examples of exemplary embodiments may have different values.

[0070] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0071] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A resource learning path planning method based on deep learning, characterized in that, Includes the following steps: The process involves acquiring the knowledge level and historical records of multiple learners, and constructing a learner model based on these records. The knowledge level includes knowledge point IDs and mastery levels, while the historical records include learning sequences, learning resources, and assignment data. The construction of the learner model includes: determining the cluster value k for similar learners based on the acquired learner behavior data; and calculating the distance between the coordinates of all learners and k seed points, expressed as: in, Let Y represent the coordinate distance between learner i and learner j, and let Y represent the learner model. This represents the coordinates of learner i in the k-th behavioral dimension. Let G represent the coordinates of learner j in the k-th behavioral dimension, and n represent the total number of behavioral dimensions. Compare the distances of each learner's coordinates to each seed point. If the distance between learner coordinates G and seed points is greater than or equal to the distance between learner coordinates G and seed points G, then... If the distance is minimized, then the learner coordinates G belong to the learner cluster. Moving the seed point causes all child points to move to the center of their cluster, even if the seed point... The process is repeated until the sum of the coordinate distances to the learner belonging to the learner cluster is minimized. Calculate the learner similarity between any two different learners in the learner model, and construct the learning behavior model corresponding to each learner based on the learner similarity. Each learner is represented as a vector in the learner model. Cosine similarity is used to evaluate the similarity between two vectors; that is, the cosine of the angle between the two learner vectors is taken as the similarity score between the learners. The expression for the similarity score is: in, Assign a similarity score between learner a and learner b. Let be the cosine of the vector representations of learner a and learner b. Represents the vector representation of learner a. The vector representation of learner b. This represents the magnitude of the learner's vector a. The vector representation of learner b; The learner's cognitive level is predicted based on the learning behavior model and the learner model, and a learning path recommendation model is constructed based on the cognitive level. The prediction of the cognitive level includes: dividing the learner's cognitive level into three domains: the positive domain (POS), the boundary domain (BND), and the negative domain (NEG); and using the Item Response Theory (2PL) model, describing the probability of the learner answering a preset question correctly using an item characteristic curve, the expression of which is: in, Let f represent the probability of answering a test question correctly, and let f represent the difficulty coefficient of the question. d represents the constant 1.

072. The ability value represents the learner's cognitive level. The POS field represents learners with high cognitive levels, the NEG field represents learners with low cognitive levels, and the BND field represents learners with medium cognitive levels. The test-taking results data from the obtained score matrix and the known question difficulty coefficients are input into the IRT model to establish the maximum likelihood function of the ability parameters, whose expression is: Where L is the maximum likelihood function value, This represents the probability of a correct answer obtained from the project response theory model function. Let represent the learner's actual answer status in the score matrix, m represent the total number of questions, and i represent the question index; taking the logarithm of the above maximum likelihood function yields the log-maximum likelihood function, whose expression is: The symbols have the same meaning as above; the capability parameter is obtained by differentiating the log-maximum likelihood function. The learning path recommendation model includes constructing two state sets. These represent whether the learner accepts the recommendation, with C indicating acceptance of the recommendation. This indicates that the recommendation is not accepted; the learning path recommendation model further includes a learning utility function matrix, which satisfies the following constraints. and ,in , , These represent the utility values ​​corresponding to positive-domain, boundary-domain, and negative-domain behaviors when the object belongs to C, respectively. , , These represent the utility values ​​corresponding to positive domain, boundary domain, and negative domain behaviors when the object does not belong to C, respectively. Based on the learning path recommendation model and the preset knowledge graph, the learner's target learning path is planned, including: obtaining a set of knowledge points to be sorted from all knowledge points in the preset knowledge graph, and setting an initial knowledge point. ; Export list L from the database corresponding to the preset knowledge graph to determine the relationships between knowledge points. These relationships include inclusion and parallel relationships between knowledge points. The record information in list L is as follows: Representing knowledge points and The relation is R, and the knowledge point relation includes knowledge point importance, difficulty, centrality, topological hierarchy attribute feature data, and knowledge point ranking index; the target learning path is generated by using a topological ranking algorithm based on the preset knowledge graph and knowledge point attribute feature data. The target learning path is compared with the preset expert path to obtain the path similarity. The expression for calculating the path similarity is as follows: in, To predetermine the path similarity between the expert path and the target learning path, , These represent the preset expert path and the target learning path, respectively. This indicates the number of learning paths that match the preset expert path and the target learning path. This represents the total number of learning paths; the quality of the target learning paths is evaluated using the learning path evaluation metric "fitness," and the formula for calculating the fitness metric is as follows: Where fitness is the learning path quality evaluation index value. This indicates the number of learning paths that violate the principle of adjacency between knowledge points. This indicates the number of times a learning path violates the principle of prioritizing knowledge points. This represents the i-th knowledge point in the target learning path. This represents the (i+1)th knowledge point in the target learning path. q represents the j-th knowledge point in the target learning path, i represents the total number of knowledge points in the target learning path, and j represents the index of the knowledge point. The fewer the number of rules violated in the learning path, the lower the fitness value, indicating that the path generation quality is higher.

2. The resource learning path planning method based on deep learning according to claim 1, characterized in that, Exporting list L from the database corresponding to the preset knowledge graph to determine the relationships between knowledge points also includes: A keyword extraction algorithm is used to process text resources to obtain a set of knowledge points and the importance of each knowledge point, expressed as follows: in, Represents a node The weight, Represents the node after the last iteration. The weight, Represents a node With nodes Similarity between them Represents a node With nodes Similarity between them Indicates pointing to a node The set of nodes, Represents a node The set of nodes to be pointed to, where g represents the damping coefficient; Based on the properties of knowledge graphs, the centrality of a knowledge point is calculated using the ratio of its in-degree to its out-degree. A higher ratio indicates a higher centrality. The expression for the in-degree and out-degree of a knowledge point is as follows: The expression for calculating centrality is: in, Represents a node The ratio of in-degree to out-degree. Representing knowledge points The set of first-order precursor knowledge points Representing knowledge nodes The set of first-order successor knowledge points, The cardinality of the set of predecessor knowledge points. The cardinality of the set of subsequent knowledge points; The above attribute values ​​are used to calculate the knowledge point ranking index. Importance Using the above expression The expression for the node's weight value and the knowledge point ranking index is as follows: in, Representing knowledge points Ranking metrics , , and These represent the importance, difficulty, centrality, and topological level of a knowledge point's KPI, respectively. , , and This indicates the attribute weight value assigned.

3. The resource learning path planning method based on deep learning according to claim 1, characterized in that, The learning path recommendation model constructed based on the aforementioned cognitive level also includes: Preset Where T represents the set of knowledge points, i represents a single knowledge point, and m represents the total number of knowledge points, then Where L represents the learner path sequence, Represents a single learner. Indicate learner Learning path These represent the 1st to the nth learners, This represents the 1st to mth knowledge points on the learning path of the nth learner, where n represents the total number of learners and m represents the total number of knowledge points on the learning path.

4. The resource learning path planning method based on deep learning according to claim 3, characterized in that, Also includes: By tracking relevant knowledge points through learners' questions and categorizing the questions, a pre-set... This represents the M question texts in the set. This represents N pre-defined knowledge points, and the classification matrix is... ,in Indicates the first The question text and the first The relationship between these knowledge points is expressed as follows: The formula for calculating the ratio of the number of nodes completed to the number of nodes where a particular learning objective is located, as recommended to learners, is as follows: Where path represents the learning path achievement rate, and n represents the total number of nodes in the learning path. This represents the total number of nodes actually recommended. The smaller the path value, the more satisfactory and expected the recommendation results are; the larger the path value, i.e., the closer it is to 1, the less satisfactory the recommendation results are.

5. A resource learning path planning apparatus for implementing the deep learning-based resource learning path planning method according to any one of claims 1-4, characterized in that, include: The data acquisition module is used to acquire the knowledge level and historical record information of multiple learners, and to construct a learner model based on the historical record information and the knowledge level. The knowledge level includes knowledge point IDs and mastery levels, and the historical record information includes learning sequences, learning resources, and assignment data. The construction of the learner model includes: determining the cluster value k of similar learners based on the acquired learner behavior data, and calculating the distance between the coordinates of all learners and k seed points, expressed as: Compare the distances of each learner's coordinates to each seed point, and assign the learner's coordinates to the learner cluster corresponding to the seed point with the smallest distance; iteratively move the seed point to the center of its cluster until the seed point stops moving; The first construction module is used to calculate the learner similarity between any two different learners in the learner model, and to construct the learning behavior model corresponding to each learner based on the learner similarity. Each learner is represented as a vector in the learner model, and the similarity between two vectors is evaluated using cosine similarity, i.e., the cosine of the angle between the two learner vectors is used as the similarity score between the learners. The expression for the similarity score is: ,in Represents the vector representation of learner a. The vector representation of learner b; The second construction module is used to predict the learner's cognitive level based on the learning behavior model and the learner model, and to construct a learning path recommendation model based on the cognitive level. The prediction of the cognitive level includes dividing the learner's cognitive level into three domains: a positive domain (POS), a boundary domain (BND), and a negative domain (NEG), and using the Item Response Theory (2PL) model. Combining maximum likelihood function Cognitive ability parameters Estimation is performed; the learning path recommendation model includes constructing two state sets. These represent whether the learner accepts the recommendation, with C indicating acceptance of the recommendation. This indicates that the recommendation is not accepted; the learning path recommendation model further includes satisfying constraints. and The learning utility function matrix; The path planning module is used to plan the target learning path for the learner based on the learning path recommendation model and the preset knowledge graph. The path planning module obtains all knowledge points in the preset knowledge graph as a set of knowledge points to be sorted and presets an initial knowledge point. A list L is exported from the database corresponding to the preset knowledge graph to determine the relationships between knowledge points. Based on the preset knowledge graph and the attribute feature data of the knowledge points, a topological sorting algorithm is used to generate a target learning path. The target learning path is then compared with the preset expert path to obtain the path similarity. Learning path evaluation metrics To verify the quality of the target learning path.

Images