A method, device and storage medium for recommending mathematical resources based on a knowledge graph

Abstract

The application provides a kind of method, equipment and storage medium based on knowledge graph mathematical resource recommendation, and the application relates to the field of online learning and mathematical resource personalized recommendation, which comprises: extracting the application mode of mathematical resources;According to the application mode, a multi-layer knowledge graph is generated;Diagnose the cognitive state of the learner, and plan the learning path according to the weak knowledge attribute of the cognitive state of the learner;According to the generated knowledge graph, the mathematical resources corresponding to the key application mode are found, and the online learners are personalized recommended based on the recent development zone and support degree.The technical effect of the application is that the interrelationship of knowledge attributes in mathematical learning resources is highlighted while ensuring good recommendation, and the corresponding resources are recommended through the application mode, making the recommendation more targeted and effectively helping online learners to flexibly use associated knowledge attributes.

Description

Technical Field

[0001] This invention relates to the field of online learning and personalized recommendation of mathematical resources, and in particular to a method, device and storage medium for recommending mathematical resources based on knowledge graphs. Background Technology

[0002] The structural relationships among the knowledge attributes within learning resources are particularly prominent in mathematical resources. Because mathematical resources follow a logical structure and are not simply a collection of knowledge attributes, to help online learners master mathematical knowledge more easily and efficiently, learners must first grasp the key application patterns of the knowledge attributes within mathematical resources. This allows them to move from understanding knowledge to mastering it, and from mastering it to flexibly applying it. Existing personalized learning resource recommendations only scratch the surface, recommending relevant learning resources based on users' weak knowledge attributes, neglecting the structural relationships between the knowledge attributes within the learning resources. They fail to delve into the knowledge application patterns constituted by the knowledge attributes within the learning resources, thus hindering users' mastery and flexible application of these knowledge attributes.

[0003] Knowledge graphs are large-scale semantic networks, and many recommendation-based learning methods already exist. Mapping mathematical resources, knowledge attributes, and the logical relationships between them onto multi-layered knowledge graphs can effectively help online learners construct learning paths. However, not all knowledge attribute application patterns carried by learning resources are valuable. How to generate and filter key knowledge attribute application patterns from a large number of learning resources and quantitatively consider the cognitive load of recommended resources is a pressing problem that needs to be solved. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by using patterns to generate multi-layer knowledge graphs for recommendation of key knowledge attributes of a large number of mathematical resources, and considers the cognitive load problem of recommended resources. It proposes a method, device and storage medium for recommending mathematical resources based on knowledge graphs.

[0005] According to a first aspect of the present invention, a method for recommending mathematical resources based on knowledge graphs is provided, comprising the following steps:

[0006] Patterns for extracting and utilizing mathematical resources;

[0007] Generate multi-layered knowledge graphs based on application patterns;

[0008] Diagnose learners’ cognitive status and plan learning paths based on learners’ weak knowledge attributes.

[0009] Based on the generated knowledge graph, mathematical resources corresponding to key application patterns are found, and personalized recommendations are made to online learners based on their zone of proximal development and support levels.

[0010] Based on the above technical solution, the present invention can also be improved as follows.

[0011] Optionally, the application modes of extracting mathematical resources include:

[0012] The sequence pattern mining algorithm is used to construct a sequence set for the knowledge attribute sequence contained in each mathematical resource in the mathematical resource set, and the frequent itemsets are extracted as the application pattern of the mathematical resource.

[0013] Optionally, generating a multi-layer knowledge graph based on the application pattern includes: embedding the obtained knowledge attribute application pattern, learning resources, and knowledge attributes into the knowledge graph to form a multi-layer knowledge graph.

[0014] Optionally, the diagnosis of the learner's cognitive state includes:

[0015] Estimate the results of online learners' responses to learning resources under error-free conditions, and the probability of the learners' responses after being tested with multiple online exercise resources, and estimate the learners' knowledge mastery through the probabilities.

[0016] Optionally, the personalized recommendations for online learners based on the generated knowledge graph, using the zone of proximal development and support level, include:

[0017] Learning path planning is performed on the knowledge attribute layer of the generated knowledge graph to generate dynamic learning paths for knowledge that learners have not mastered. Based on the knowledge attributes that appear in the learning path, key application patterns are searched in the knowledge graph, and learning resources are recommended based on the key application patterns and the zone of proximal development theory.

[0018] Optionally, the knowledge graph includes one or more of the following: "containing attributes", "containing patterns", "prerequisite knowledge", "parent class knowledge", "knowledge patterns", and "subclass patterns". The relationships between "containing attributes" and "containing patterns" are extracted using mathematical semantic extraction networks and matrix automatic generation algorithms, or extracted using expert labeling. The relationships between "prerequisite knowledge" and "subclass knowledge" are obtained based on mathematics textbooks and examination syllabi, through the analysis and annotation of the "chapter and section" two-level structure by domain experts. The relationships between "knowledge patterns" and "subclass patterns" are extracted using an application pattern extraction algorithm based on frequent itemsets.

[0019] Optionally, the process of generating the learning path includes:

[0020] If a knowledge attribute is selected, that knowledge will be used as the target knowledge; if no knowledge attribute is selected, learning resources will be randomly selected, and the learner will answer the questions. After the teacher corrects the answers, the knowledge that the learner has not mastered will be used as the target knowledge based on the diagnosis of the learner's weak cognitive state.

[0021] The set of knowledge attributes to be hidden is selected according to the learner's own learning status or progress, and the set of knowledge attributes to be hidden does not appear in the learning path;

[0022] In the knowledge attribute layer of the knowledge graph, a topological sort is performed with the target knowledge as the endpoint to generate a dynamic learning path. The set of knowledge attributes that are hidden and the set of knowledge attributes that have been mastered are removed from the dynamic path to obtain the current learning path.

[0023] Optionally, the recommended learning resources based on key usage patterns and the zone of proximal development theory include:

[0024] For each knowledge attribute in the learning path, obtain its key application mode corresponding to the application mode layer of the knowledge graph; based on the zone of proximal development theory, recommend learning resources corresponding to the application mode in the learning resource layer to the learner according to the learner's ability and in order of application mode from shortest to longest.

[0025] According to a third aspect of the present invention, an electronic device is provided, including a memory and a processor, the processor being configured to implement a method for recommending new mathematical resources based on a knowledge graph when executing a computer program stored in the memory.

[0026] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a computer management program stored thereon, the computer program being executed by a processor as steps of a method for recommending new mathematical resources based on a knowledge graph.

[0027] The technical effects and advantages of this invention are as follows:

[0028] This invention, based on the application pattern of knowledge attributes extracted from frequent itemsets, can more deeply explore the correlation between knowledge attributes. It integrates the application pattern of knowledge attributes, knowledge attributes, and learning resources into a multi-layered knowledge graph. It proposes a method, device, and storage medium for recommending mathematical resources based on the knowledge graph. The method of this invention can highlight the interrelationship of knowledge attributes in mathematical learning resources while ensuring good recommendations. By recommending corresponding resources through application patterns, the recommendations are more targeted and effectively help online learners to flexibly apply related knowledge attributes.

[0029] This invention combines the zone of proximal development theory and support levels to more accurately help users smoothly advance beyond their current cognitive level, thereby enabling personalized math resource recommendations.

[0030] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0031] Figure 1 This is an overall flowchart of a knowledge graph-based mathematical resource recommendation method provided in an embodiment of the present invention;

[0032] Figure 2 This is a flowchart of the application mode for searching knowledge attributes provided in the embodiments of the present invention;

[0033] Figure 3 This is a schematic diagram of a multi-layer knowledge graph provided in an embodiment of the present invention;

[0034] Figure 4 This is a flowchart of cognitive diagnosis provided in an embodiment of the present invention;

[0035] Figure 5 This is a flowchart of the learning resource recommendation process provided in an embodiment of the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] It should be noted that, in this embodiment, learning resources are natural language text, videos, audio, etc., containing knowledge that learners can study. In this invention, learning resources specifically refer to natural language text containing knowledge, while mathematical resources are exercise resources containing mathematical knowledge. The knowledge attribute of a learning resource refers to the knowledge contained in that learning resource. For example, if a math exercise resource contains knowledge such as trigonometric functions, Green's theorem, and indefinite integrals, then that math exercise can be called a mathematical resource, and the knowledge such as trigonometric functions, Green's theorem, and indefinite integrals can be called knowledge attributes.

[0038] To address the problems of existing technologies, this patent combines the zone of proximal development theory and support levels to more accurately help users smoothly progress across their current cognitive levels, thereby achieving personalized mathematical resource recommendations. This invention provides a method for recommending mathematical resources based on knowledge graphs, specifically as follows... Figure 1 As shown, the method includes the following steps:

[0039] Step 1. Extract the application patterns of mathematical resources;

[0040] Step 2. Generate a multi-layered knowledge graph;

[0041] Step 3. Diagnose the learner's cognitive state and plan a learning path based on the learner's weak knowledge attributes;

[0042] Step 4. Based on the generated knowledge graph, find the mathematical resources corresponding to the key application patterns and make personalized recommendations to online learners based on their zone of proximal development and support.

[0043] It should be noted that the application mode of extracting knowledge attributes based on frequent itemsets in this invention can more deeply explore the correlation between knowledge attributes. It integrates the application mode of knowledge attributes, knowledge attributes and learning resources into a multi-layer knowledge graph. This method can not only ensure better recommendations, but also highlight the interrelationships of knowledge attributes in mathematical learning resources. By recommending corresponding resources through application modes, the recommendations are more targeted.

[0044] Specifically, in step 1, the extraction and application patterns of mathematical resources include:

[0045] The sequence pattern mining algorithm is used to extract frequent itemsets as the application patterns of mathematical resources from the sequence set consisting of the knowledge attribute sequences contained in each mathematical resource in the mathematical resource set.

[0046] Furthermore, the specific application mode for extracting mathematical resources includes: using sequence pattern mining algorithms to analyze the mathematical resource set. Each mathematical resource L i The sequence set consisting of the knowledge attribute sequences contains frequent itemsets with support greater than the minimum support β, forming a pattern for the use of mathematical resources.

[0047] Among them, the extraction of mathematical resource application models includes the formation of mathematical resource sets. Given a two-dimensional sequence R containing knowledge attributes, a sequence pattern mining algorithm is used to extract a set F of all subsequences R containing a support greater than the minimum support and a length n ranging from 1 to 4. Based on set F, the knowledge attribute q is obtained. j Key application patterns of different lengths

[0048] In this embodiment of the invention, a mathematical resource set Each mathematical resource L i Using knowledge attribute sequences Let N represent this, where N c This indicates the number of knowledge attributes of the current learning resource. This indicates that the knowledge attribute with number j is located at the r-th position of the current resource, forming a two-dimensional sequence R of knowledge attributes; based on knowledge attribute q j Divide the two-dimensional sequence R into subsequences R that contain this attribute. j , Where R k That is, the knowledge attribute sequence N represents the total number of mathematical resources; a knowledge attribute sequence of length n. The support rate is When the support of sequence s is greater than the minimum support, s is a frequent itemset. The sequence pattern mining algorithm can be the Generalized Sequence Pattern Algorithm (GSP), which extracts frequent itemsets of length n from 1 to 4 from a two-dimensional sequence R. The minimum support is set to β, and those skilled in the art can set the value of β according to specific circumstances. This yields key application patterns for lengths n = 2, 3, and 4. and the level of support for each application mode;

[0049] Specifically, the method for extracting frequent itemsets as a pattern for using mathematical resources is as follows:

[0050] When n=1, calculate the support of each knowledge attribute. Knowledge attributes with support greater than β are candidate subsequences represented by the set SET.

[0051] When n=2, the elements in the set SET are paired to form candidate sequences, the support of each candidate sequence is calculated, and the candidate sequences with support greater than β are stored in the frequent itemset set F.

[0052] When n = k, the sequences of length k-1 in the frequent itemset F are combined pairwise to form candidate sequences. The combination method is as follows: if the sequence s1 after removing the first element is the same as the sequence s2 after removing the last element, then the first element of s1, the remaining sequence, and the last element of s2 are merged into one sequence. If all its subsequences of length k-1 are in the set F, then this sequence is taken as a candidate sequence. The support of all candidate sequences is calculated, and the candidate sequences with support greater than β are stored in the frequent itemset set F, until k = 4.

[0053] Therefore, for each knowledge attribute q j This knowledge can be acquired in up to three key application patterns of varying lengths. And the support level for each application mode, The corresponding key application pattern has a length of l+1.

[0054] In specific implementation, such as Figure 2 As shown, for a two-dimensional sequence R of knowledge attributes, for each row, first generate a candidate sequence of length 1, that is, generate a sequence set { for each knowledge attribute as a separate candidate sequence}. <q1> ， <q2> ， <q3> ，...， <q8>}, filter candidate sequences with support greater than a threshold to obtain sequence set T1: { <q1> ， <q2> ， <q3> ， <q4> ， <q5> ， <q6>}, combine the sequences in the sequence set pairwise to obtain a candidate sequence set, and then filter the candidate sequences with support greater than a threshold to obtain the sequence set T2: {<q1,q1> ,<q1,q2> ,...,<q2,q1> ...}, combine the sequences in the sequence set pairwise, remove the first and last sequences. If the remaining sequences are the same, combine the "first" and "last" sequences with the remaining sequences to form a new sequence. If every subsequence in the new sequence is in T1 and T2, then this sequence is considered a candidate sequence. Select candidate sequences with support greater than a threshold to obtain the sequence set T3: {<q1,q1,q1> ,<q1,q2,q3> ,...,<q2,q3,q4> Similarly, we obtain the sequence set T4: {<q1,q1,q2,q3> ,<q1,q2,q3,q4> ...}; From this, we can obtain 2-knowledge pattern T2, 3-knowledge pattern T3, 4-knowledge pattern T4; and the application pattern corresponding to knowledge attribute q1. For a sequence containing q1 in a 2-knowledge pattern, apply the pattern... For a sequence containing q1 in a 3-knowledge pattern, apply the pattern... The sequence containing q1 is a 4-knowledge pattern.

[0055] Specifically, in step 2, generating a multi-layer knowledge graph includes: embedding the obtained knowledge attribute application patterns and learning resources and knowledge attributes into the knowledge graph to form a multi-layer knowledge graph.

[0056] Furthermore, specifically, this includes: embedding the knowledge attribute application patterns obtained in step 1, along with learning resources and knowledge attributes, into the knowledge graph to form a three-layer knowledge graph. The three-layer knowledge graph includes a learning resource generation layer, a knowledge attribute layer, and an application pattern layer. The entities in the knowledge graph include two types: mathematical resources and knowledge attributes. The relationships in the knowledge graph are of two types: inter-layer relationships and intra-layer relationships. The entities in the learning resource layer are learning resources, which are recommended objects of mathematical resources.

[0057] In this embodiment of the invention, the nodes of the learning resource layer in the knowledge graph are composed of learning resource sets. Mathematical Resources L i The nodes of the knowledge attribute layer and the application mode layer are composed of knowledge attribute sets. Knowledge attribute q in j Composition, where N Q It represents the total number of knowledge attributes; there are six types of relationships between and within a knowledge graph, including "containing attributes", "containing patterns", "prerequisite knowledge", "parent knowledge", "knowledge patterns" and "child patterns".

[0058] A schematic diagram of a multi-layered knowledge graph is shown below. Figure 3 As shown in Table 1, the relationship is defined as follows: Learning resources are natural language text, videos, and audio containing knowledge available for learners. In this invention, learning resources specifically refer to natural language text containing knowledge, and knowledge attributes are the knowledge contained in the learning resource. For example, a math problem resource may contain trigonometric functions, Green's formula, and indefinite integrals. The application pattern is represented as the application pattern of extracting frequent itemsets as mathematical resources. The relationships between "containment attributes" and "containment patterns" are extracted using the results of a mathematical semantic extraction network and an automatic Q-matrix generation algorithm, or through expert labeling. The relationships between "prerequisite knowledge" and "subclass knowledge" are obtained based on math textbooks and exam outlines, through the analysis and annotation of the "chapter / section" two-level structure by domain experts. The relationships between "knowledge patterns" and "subclass patterns" are extracted by the application pattern extraction algorithm based on frequent itemsets in step 1, and the set of all relationships is represented as...

[0059] Table 1: Inter-layer and Intra-layer Relationships in a Multi-layer Knowledge Graph

[0060]

[0061] Specifically, in step 3, diagnosing the learner's cognitive state and planning a learning path based on the learner's weak knowledge attributes include:

[0062] Estimate the results of online learners' responses to learning resources under error-free conditions, and the probability of the learners' responses after being tested with multiple online exercise resources, and estimate the learners' knowledge mastery through the probabilities.

[0063] Furthermore, diagnosing the learner's cognitive state specifically includes: estimating the online learner's u under error-free conditions. k Answering Learning Resources L i The result η ki The probability P(X) of a learner's answer after taking I online exercise resources with errors. k1 X k1 , ..., X k1 |α k ), and through it, estimate learner u k Knowledge mastery;

[0064] Diagnosing learners' cognitive state, including estimating the error-free state of online learners. k Answering Learning Resources L i The result η ki The learner's answer to the test after completing one online exercise resource, with errors. The probability of occurrence P(X) k1 X k1 , ..., X k1 |α k ), and through it, estimate learner u k The knowledge mastery of α k For learners u k The knowledge mastery vector.

[0065] Specifically, such as Figure 4 As shown, for online learner u k Conduct cognitive diagnosis to obtain learners' responses to learning resources L i The learner's knowledge acquisition status. Cognitive diagnosis uses the DINA model, with learner u k Mastery of each knowledge attribute Where α kj =1 indicates learner u k Knowledge attribute q has been mastered j α kj =0 indicates learner u k Knowledge attribute q not mastered j Learning Resources L i The assessment of knowledge attributes is represented by the vector Q. i express, Where q ij >0 indicates learning resource L i The knowledge attribute q was examined. j q ij =0 indicates learning resource L i Knowledge attribute q was not examined. j According to α k and Q i To obtain learner u in error-free condition k Is the answer correct? i The result η ki :

[0066]

[0067] There is a guess for the parameter g. i and error parameter s i In cases where there are errors, the learner correctly answers L. i The probability is

[0068] Among them, X ki =1 indicates learner u k Correct answer L i Therefore, learner u k After taking one online learning resource test, the results were graded and obtained. The probability of this occurring in cognitive diagnosis is...

[0069] The EM algorithm and maximum likelihood estimation method can be used to obtain the answer. In the case of maximum expected value, parameter α k The value of u is obtained to determine the learner's value. k The extent of their knowledge acquisition.

[0070] In the specific implementation process, mathematical resources are divided into four categories: multiple choice questions, true / false questions, fill-in-the-blank questions, and problem-solving questions. For different types of learning resources, assuming that learners guess randomly with equal probability, the method for estimating the lower limit of the guessing parameters is as follows:

[0071] Multiple-choice questions with n options, guessing parameters for single-choice questions. m-option multiple choice question: guess parameters Combinations; guessing parameters in multiple-choice questions.

[0072] True or False, guess the parameter

[0073] Fill in the blanks and answer the problem-solving questions, guessing the parameter g. i =0.

[0074] Error parameters are estimated using a method based on the maximum inverse file frequency, s i =tanh(γθ) i ),in

[0075] θ i The rarity of the current resource is represented by N, where N represents the set of mathematical resources. The total number, This represents the number of learning resources containing the j-th knowledge attribute, where q represents the knowledge attribute. j The inverse logarithm of the proportion of the number of learning resources covered to the total number of learning resources N is taken as the inverse file frequency, and the maximum inverse file frequency is used as the rarity of the current resource; γ is a parameter controlling the slope of the function. In other words, the mathematical resource set contains knowledge attribute q. j The fewer mathematical resources available, the lower the knowledge attribute q. j The rarer the knowledge attribute, the rarer the current mathematical resource, the rarer it is, and this is used to measure the error parameter.

[0076] Given learners I learning resources, collect their responses. Based on the examination of knowledge attributes by I learning resources, a Q matrix is ​​formed: Q = [q ij ] i×j , where q ij >0 indicates learning resource L i The knowledge attribute q was examined. j q ij =0 indicates learning resource L i Knowledge attribute q was not examined. j In the event of errors, all combinations U of knowledge attributes are used as cognitive states to answer the question, where U = 2. j This indicates that each knowledge attribute has two possibilities: it exists or it does not exist in the learner's cognitive state. The full probability likelihood of all results is calculated.

[0077]

[0078] Where, α u γ represents the cognitive state represented by the u-th combination of knowledge attributes. u =1 indicates that the student has mastered the cognitive state of the u-th knowledge attribute combination, θ represents the parameter, which refers to the prior probability of each cognitive state, μ u Let represent the probability that the student masters the u-th combination of knowledge attributes. Using the learner's cognitive state as a latent variable, the EM algorithm is employed to calculate the expected function of the total probability likelihood.

[0079]

[0080] Maximize the expectation function to update the parameter θ, and iterate to obtain the optimal solution for θ and the maximum value of the Q function, at which point α... u For learners u k The cognitive state refers to the learner's mastery of the knowledge attributes in a given set of learning resources.

[0081] Specifically, in step 4, the mathematical resources corresponding to key application patterns are searched based on the generated knowledge graph, and personalized recommendations are made to online learners based on their zone of proximal development and support levels, including:

[0082] Learning path planning is performed on the knowledge attribute layer of the generated knowledge graph to generate dynamic learning paths for knowledge that learners have not mastered. Based on the knowledge attributes that appear in the learning path, key application patterns are searched in the knowledge graph, and learning resources are recommended based on the key application patterns and the zone of proximal development theory.

[0083] Specifically, personalized recommendations for online learners based on their zone of proximal development and support level include: planning learning paths for the knowledge attribute layers of the knowledge graph generated in step 2, and generating dynamic learning paths for knowledge that learners have not yet mastered. Where L P Summarize the points for the path; based on the q appearing in the learning path jk Find key application patterns Based on key application patterns Recommended learning resources

[0084] Personalized recommendations for online learners include: matching the multi-layered knowledge graph generated in step 2 with the target knowledge q selected by the online learner. t Planning and Generating Learning Paths Where L P Summarize the points for the path; if the online learner does not select the target path, use the knowledge they have not mastered in the cognitive diagnosis results of step 3 as the target knowledge; shield the knowledge already mastered in the learning path, obtain the corresponding application mode according to the knowledge attributes in the learning path, and use the learner's knowledge ability vector α as the "current development level" to gradually increase the length of the application mode, helping learners smoothly cross cognitive levels.

[0085] In this embodiment of the invention, the learning path generation process is as follows:

[0086] If the online learner selects knowledge attribute q t If the learner selects a specific knowledge resource, that knowledge will be used as the target knowledge. If the learner does not select any knowledge, learning resources will be randomly selected. After the learner answers and the teacher corrects the answers, a cognitive diagnosis will be performed on the learner according to step 3, and the knowledge not mastered in the diagnosis results will be used as the target knowledge q. t ;

[0087] Learners can choose to hide the set of knowledge attributes according to their learning status or progress. The knowledge attributes in the learning path will not appear in the learning path;

[0088] In the knowledge attribute layer Q of the knowledge graph, the target knowledge q t Perform topological sorting on the endpoint to generate a dynamic learning path. Remove the set of knowledge attributes hidden in dynamic paths Based on the learner's existing set of knowledge attributes α, obtain the current learning path. q jt This represents the knowledge attribute at position t in the learning path.

[0089] In this embodiment, the learning resources are obtained based on knowledge attributes as follows:

[0090] For each knowledge attribute q in the learning path jt To obtain the key application patterns corresponding to its application pattern layer in the knowledge graph. According to the zone of proximal development theory, learning resources corresponding to the learner's ability α are recommended to the learner in order of application mode from front to back.

[0091] Specifically, based on the key application pattern recommended for the zone of proximal development, it is as follows: in knowledge attribute q jt Key application patterns of three lengths In this process, all application patterns are ranked according to the principle of prioritizing shorter patterns, forming multi-width zones of proximal development. If multiple application patterns of the same length exist, they are ranked according to the principle of prioritizing support, with more frequently occurring key patterns ranked higher. Based on the zone of proximal development theory, knowledge attribute q... jt Using the learner's knowledge and ability α as the basic width of the "zone of proximal development" as the "current developmental level," the width of the zone of proximal development is gradually increased, which means gradually increasing the length of the recommended application patterns to form a sequence of recommended application patterns. In the learning resource layer, find the corresponding learning resources based on the usage pattern. Learning resources are recommended to learners in the order of their application patterns.

[0092] Examples of specific implementation processes are as follows: Figure 5 As shown, the target knowledge is q. 12 Block knowledge Let {q1} be the current knowledge and ability vector of the online learner. <q4,q 11 >;In the knowledge attribute layer Q, with target knowledge q 12 The learning path is obtained by performing topological sorting on the endpoint. <q1,q4,q6,q8,q9,q 11 q 12 > The current dynamic learning path is obtained after removing the masked knowledge and learner knowledge and ability vectors. <q6,q8,q9,q 12 In the application pattern layer, the application patterns corresponding to the knowledge attributes in the learning path are found and sorted according to the zone of proximal development and support, resulting in an application pattern sequence. q6 corresponds to the pattern sequence. q8 corresponds to the use of pattern sequences q9, q 11 Similarly, at the learning resource layer, the corresponding learning resources are then found based on the application mode. Recommended to learners in turn. Corresponding learning resources Corresponding learning resources The same applies to the rest; cognitive diagnosis is performed based on the learner's responses, and the learner's knowledge and ability vector is updated.

[0093] In summary, addressing the current issue of personalized mathematics resource recommendation focusing primarily on knowledge attributes, this invention proposes a novel mathematics resource recommendation method based on knowledge attribute extraction using frequent itemsets. This method delves deeper into the relationships between knowledge attributes, integrating these application patterns, knowledge attributes, and learning resources into a multi-layered knowledge graph. This knowledge graph-based approach not only ensures high-quality recommendations but also highlights the interrelationships among knowledge attributes within mathematics learning resources. By recommending relevant resources through application patterns, the method achieves greater targeting and effectively assists online learners in flexibly applying related knowledge attributes.

[0094] The above description is merely a selection of embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions that fall within the scope of the present invention are within the scope of protection of the present invention.

[0095] According to a second aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the processor is configured to execute a computer program stored in the memory to implement a method for recommending new mathematical resources based on a knowledge graph. The method includes the following steps:

[0096] Patterns for extracting and utilizing mathematical resources;

[0097] Generate multi-layered knowledge graphs based on application patterns;

[0098] Diagnose learners’ cognitive status and plan learning paths based on learners’ weak knowledge attributes.

[0099] Based on the generated knowledge graph, mathematical resources corresponding to key application patterns are found, and personalized recommendations are made to online learners based on their zone of proximal development and support levels.

[0100] According to a third aspect of the present invention, a computer-readable storage medium is provided, on which a computer management program is stored, the computer program being executed by a processor as steps of a method for recommending new mathematical resources based on a knowledge graph. The method includes the following steps:

[0101] Patterns for extracting and utilizing mathematical resources;

[0102] Generate multi-layered knowledge graphs based on application patterns;

[0103] Diagnose learners’ cognitive state and plan learning paths based on the learners’ weak knowledge attributes in their cognitive state;

[0104] Based on the generated knowledge graph, mathematical resources corresponding to key application patterns are found, and personalized recommendations are made to online learners based on their zone of proximal development and support levels.

[0105] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0106] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention. < / q5> < / q4> < / q3> < / q2> < / q1> < / q3> < / q2> < / q1>

Claims

1. A method for recommending mathematical resources based on knowledge graphs, characterized in that, Includes the following steps: Extracting the application patterns of mathematical resources includes: constructing a sequence set from the knowledge attribute sequences contained in each mathematical resource in the mathematical resource set, and extracting frequent itemsets as the application patterns of the mathematical resources; the application patterns take the knowledge attribute sequences corresponding to the mathematical resources as input, and use a sequence pattern mining algorithm to extract frequent knowledge attribute sequences with support not less than a preset threshold. Generate multi-layered knowledge graphs based on application patterns, including: embedding the obtained knowledge attribute application patterns, learning resources, and knowledge attributes into the knowledge graph to form multi-layered knowledge graphs; Diagnosing learners' cognitive state includes: estimating the learner's answer to learning resources under error-free conditions, and the probability of the learner's answer after multiple online exercise resource tests, and estimating the learner's knowledge mastery through the probability; planning learning paths based on the learner's weak knowledge attributes in their cognitive state; Based on the generated knowledge graph, mathematical resources corresponding to key application patterns are located, and personalized recommendations are made to online learners based on their zone of proximal development and support levels. This includes: planning learning paths for the knowledge attribute layers of the generated knowledge graph, generating dynamic learning paths for knowledge that learners have not yet mastered, and for each knowledge attribute in the learning path, obtaining the key application patterns corresponding to its application pattern layer in the knowledge graph; and based on the zone of proximal development theory, recommending learning resources corresponding to the application patterns in the learning resource layer to learners according to their abilities and in ascending order of application pattern length.

2. The method for recommending mathematical resources based on knowledge graphs according to claim 1, characterized in that, The multi-layered knowledge graph includes one or more of the following: "containing attributes", "containing patterns", "prerequisite knowledge", "parent knowledge", "knowledge patterns", and "subclass patterns". The relationships of "containing attributes" and "containing patterns" are extracted using mathematical semantic extraction networks and matrix automatic generation algorithms, or extracted using expert labeling. The relationships of "prerequisite knowledge" and "subclass knowledge" are obtained based on mathematics textbooks and examination syllabi, through the analysis and annotation of the "chapter and section" two-level structure by domain experts. The relationships of "knowledge patterns" and "subclass patterns" are extracted using an application pattern extraction algorithm based on frequent itemsets.

3. The method for recommending mathematical resources based on knowledge graphs according to claim 1, characterized in that, Planning learning paths based on learners' weak knowledge attributes in their cognitive state includes: If the learner selects a target knowledge attribute, that target knowledge attribute will be used as the endpoint of the learning path; if no target knowledge attribute is selected, learning resources will be randomly selected, the learner will answer the questions, and the teacher will correct them. Based on the learner's weak cognitive state diagnosis, the knowledge not mastered in the diagnosis results will be used as the target knowledge. The set of knowledge attributes to be hidden is selected according to the learner's own learning status or progress, and the set of knowledge attributes to be hidden does not appear in the learning path; In the knowledge attribute layer of the knowledge graph, a topological sort is performed with the target knowledge as the endpoint to generate a dynamic learning path. The set of knowledge attributes that are hidden and the set of knowledge attributes that have been mastered are removed from the dynamic path to obtain the current learning path.

4. An electronic device, characterized in that, The system includes a memory and a processor, wherein the processor is used to execute a computer program stored in the memory to implement a method for recommending mathematical resources based on a knowledge graph as described in any one of claims 1 to 3.

5. A computer-readable storage medium, characterized in that, It stores computer management programs, which, when executed by a processor, implement a method for recommending mathematical resources based on a knowledge graph as described in any one of claims 1 to 3.

Images