Learning path recommendation method, device, electronic device and storage medium

Abstract

The present invention relates to the field of data mining, and provides a learning path recommendation method, device, electronic device and storage medium, the method comprising: determining the predecessor knowledge concept of the target knowledge concept, and hierarchically dividing the predecessor knowledge concept to obtain candidate knowledge concepts; looping the following steps until the knowledge state at the current moment indicates that the target knowledge concept has been mastered: based on the knowledge state at the current moment, determining the current moment concept combination from the candidate knowledge concepts, recommending exercises based on the current moment concept combination to obtain recommended exercises at the current moment, and updating the current moment knowledge state based on the answer status of the recommended exercises; and recommending learning paths based on the recommended exercises at each moment. The learning path recommendation method, device, electronic device and storage medium provided by the present invention can better reduce the practice burden of learners and promote a deeper understanding of interrelated knowledge concepts.

Description

Technical Field

[0001] The present invention relates to the field of data mining technology, and in particular to a learning path recommendation method, device, electronic device and storage medium. Background Art

[0002] With the development of educational technology and the popularity of online learning platforms, learners have more convenient access to learning resources. However, with the explosive growth of knowledge and learning resources in the information age, it has also brought new pain points for people to acquire new knowledge, that is, how learners can accurately obtain learning resources that suit their needs from a large number of complex learning resources, and learn efficiently in an appropriate learning order.

[0003] In view of this, the technology of recommending learning paths for learners has emerged. Generally, learning path recommendation aims to recommend a sequence of knowledge concepts and corresponding exercises to learners to achieve learning goals efficiently, which is an important part of adaptive learning.

[0004] Related technologies usually adopt a learning path recommendation framework of single concept recommendation + exercise recommendation. This learning path recommendation framework is not flexible enough and is relatively inefficient, which will limit its application in actual scenarios. Summary of the invention

[0005] The present invention provides a learning path recommendation method, device, electronic device and storage medium, which are used to solve the defects of the prior art that the learning path recommendation framework is not flexible enough and is relatively inefficient.

[0006] The present invention provides a learning path recommendation method, comprising: Determine the predecessor knowledge concepts of the target knowledge concept, and hierarchically divide the predecessor knowledge concepts to obtain candidate knowledge concepts; The following steps are executed in a loop until the current knowledge state indicates that the target knowledge concept has been mastered: based on the current knowledge state, determining a current concept combination from the candidate knowledge concepts, performing exercises recommendation based on the current concept combination to obtain current recommended exercises, and updating the current knowledge state based on the answer status of the recommended exercises; Recommend learning paths based on recommended exercises at each moment.

[0007] According to the learning path recommendation method provided by the present invention, the precursor knowledge concepts are hierarchically divided to obtain candidate knowledge concepts, including: Based on the predecessor-successor relationship between the predecessor knowledge concepts, the predecessor knowledge concepts are hierarchically divided to obtain candidate knowledge concepts.

[0008] According to the learning path recommendation method provided by the present invention, the predecessor knowledge concepts are hierarchically divided based on the predecessor-successor relationship between the predecessor knowledge concepts to obtain candidate knowledge concepts, including: Based on the predecessor-successor relationship between the predecessor knowledge concepts, construct a hierarchical knowledge tree with the target knowledge concept as the root node and the predecessor knowledge concept as the leaf node; Dividing the hierarchical knowledge tree into hierarchical levels to obtain nodes included in each level; Based on the nodes included in each level, candidate knowledge concepts corresponding to each level are determined.

[0009] According to the learning path recommendation method provided by the present invention, the step of determining the current concept combination from the candidate knowledge concepts based on the current knowledge state includes: Based on the mastery degree of the concept combination at the previous moment, determining the candidate knowledge concept at the current moment from the candidate knowledge concepts; Based on the current knowledge state and the current candidate knowledge concepts, a current concept combination is determined through a large language model.

[0010] According to the learning path recommendation method provided by the present invention, the step of determining the candidate knowledge concept at the current moment from the candidate knowledge concepts based on the mastery degree of the concept combination at the previous moment includes: Based on the mastery degree of the concept combination at the previous moment, the candidate knowledge concept at the previous level or the candidate knowledge concept at the next level is determined as the candidate knowledge concept at the current moment.

[0011] According to the learning path recommendation method provided by the present invention, the step of recommending exercises based on the current concept combination to obtain the current recommended exercises includes: Based on the concept combination at the current moment and the knowledge state at the current moment, the recommended exercises at the current moment are generated through a large language model.

[0012] According to the learning path recommendation method provided by the present invention, the method of generating the recommended exercises at the current moment through a large language model based on the current concept combination and the current knowledge state includes: Searching in the question bank based on the current concept combination to obtain search questions; Based on the current concept combination, the current knowledge state and the search questions, the current recommended questions are generated through a large language model.

[0013] The present invention also provides a learning path recommendation device, comprising: A hierarchical division unit, used to determine the predecessor knowledge concepts of the target knowledge concept, and hierarchically divide the predecessor knowledge concepts to obtain candidate knowledge concepts; The state updating unit is used to cyclically execute the following steps until the current knowledge state indicates that the target knowledge concept has been mastered: based on the current knowledge state, determine the current concept combination from the candidate knowledge concepts, recommend exercises based on the current concept combination to obtain the current recommended exercises, and update the current knowledge state based on the answer status of the recommended exercises; The path recommendation unit is used to recommend learning paths based on recommended exercises at each moment.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any one of the above-mentioned learning path recommendation methods is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the learning path recommendation method described in any one of the above is implemented.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned learning path recommendation methods.

[0017] The learning path recommendation method, device, electronic device and storage medium provided by the present invention determine the predecessor knowledge concepts of the target knowledge concepts and hierarchically divide the predecessor knowledge concepts to obtain candidate knowledge concepts; then, based on the knowledge state at the current moment, determine the current moment concept combination from each candidate knowledge concept, and recommend exercises based on the current moment concept combination to obtain the recommended exercises at the current moment. Concept combination recommendation and exercise recommendation are performed alternately until the target knowledge concept is mastered. The concept combinations determined from each candidate knowledge concept are interrelated and have internal correlation. Exercise recommendation is performed based on the concept combination, so that a single exercise can learn and solve at least one single concept at the same time. Compared with the related art that has and can only solve one single concept, this can better reduce the learner's practice burden and promote a deeper understanding of interrelated knowledge concepts. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 This is one of the flow charts of the learning path recommendation method provided by the present invention.

[0020] Figure 2 This is the second flowchart of the learning path recommendation method provided by the present invention.

[0021] Figure 3 It is a flowchart of step 110 in the learning path recommendation method provided by the present invention.

[0022] Figure 4 It is a flowchart of step 120 in the learning path recommendation method provided by the present invention.

[0023] Figure 5 This is the third flowchart of the learning path recommendation method provided by the present invention.

[0024] Figure 6 It is a structural schematic diagram of the learning path recommendation device provided by the present invention.

[0025] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] Existing methods usually adopt a learning path recommendation framework of single concept recommendation + exercise recommendation. Specifically, the recommendation process of existing methods has the following two steps: 1) Single concept recommendation: First, by evaluating the learner's current knowledge level, a knowledge concept that matches the learner's knowledge level is selected based on the knowledge structure. This concept is usually a key knowledge concept that the learner needs to master during the learning process or a knowledge concept that is weak in mastering.

[0028] 2) Exercise recommendation: Once a single knowledge concept to be learned is determined, relevant exercises are recommended based on the knowledge concept. These exercises are designed to test learners' understanding and application of the concept and help learners consolidate knowledge through practice.

[0029] Ultimately, by alternating between single concept selection and exercise recommendation until the learner achieves their learning goals, the knowledge concepts selected by the method and the corresponding exercises constitute a complete learning path.

[0030] The learning path recommendation framework of single concept recommendation + exercise recommendation followed by existing methods is not flexible enough and inefficient, which may limit their application in practical scenarios: First, knowledge concepts are interrelated. Learning them in isolation is inefficient, and ignoring these internal associations may hinder learners' deeper understanding of related concepts. For example, the two prerequisite concepts c1 integer addition and c2 integer subtraction are inverse operations, where addition is summation and subtraction is difference. If learners learn these concepts separately, they must solve exercises such as addition exercise 12 + 6 and subtraction exercise 9 − 4 separately, which will increase their practice burden and weaken their understanding of the relationship between concepts, resulting in inefficient learning.

[0031] In view of the above problems, an embodiment of the present invention proposes a learning path recommendation method, in which the predecessor knowledge concepts of the target knowledge concepts are first determined, and the predecessor knowledge concepts are hierarchically divided to obtain candidate knowledge concepts; then, based on the current knowledge state, the current concept combination is determined from the candidate knowledge concepts, and exercises are recommended based on the current concept combination to obtain recommended exercises at the current moment. Concept combination recommendations and exercise recommendations are performed alternately until the target knowledge concept is mastered. The concept combinations determined from the candidate knowledge concepts are interrelated and have internal correlation. Exercise recommendations based on concept combinations can better reduce the learner's practice burden and promote a deeper understanding of interrelated knowledge concepts.

[0032] For example, c1 Integer Addition and c2 Integer Subtraction could form a concept group, allowing a single exercise (e.g., 34 + 7 − 15) to address both concepts.

[0033] The embodiments of the present invention can be applied to scenarios where learning path recommendation is required, such as online learning platforms. The execution subject of the method can be an electronic device such as a terminal device, a computer, a server, a server cluster, or a specially designed learning path recommendation device, or a learning path recommendation device set in the electronic device, and the learning path recommendation device can be implemented by software, hardware, or a combination of the two.

[0034] Figure 1 is one of the flow charts of the learning path recommendation method provided by the present invention, such as Figure 1 As shown, the method comprises the following steps: Step 110 , determining the predecessor knowledge concepts of the target knowledge concept, and hierarchically dividing the predecessor knowledge concepts to obtain candidate knowledge concepts.

[0035] Specifically, the target knowledge concept can be understood as the learner's learning goal, that is, the specific knowledge point or concept that the learner hopes to master or have a deeper understanding of. The precursor knowledge concept of the target knowledge concept refers to the basic knowledge or concept that must be mastered or understood before learning the target knowledge concept. These precursor knowledge concepts are the basis and prerequisite for learning the target knowledge concept. For example, if the target knowledge concept is two-digit multiplication, the precursor knowledge concepts may include addition, one-digit multiplication, etc. For another example, for the target knowledge concept of "deep learning", its precursor knowledge concepts may include "basics of machine learning", "neural networks", "optimization algorithms", "calculus", "linear algebra", etc.

[0036] The determination of the predecessor knowledge concept can be achieved through the knowledge structure diagram. The knowledge structure diagram shows the predecessor and successor relationship between knowledge concepts. First, all adjacent nodes of the target knowledge concept can be searched in the knowledge structure diagram to determine its direct predecessor. It can also be recursively expanded outward based on the direct predecessor to obtain the predecessor knowledge concept of the target knowledge concept.

[0037] Considering that the concept of precursor knowledge may contain more knowledge points and concepts, it is necessary to further divide the concept of precursor knowledge into different levels. The level division can be divided according to the importance and difficulty of each precursor knowledge concept in the school process to form different levels, such as basic level, advanced level and advanced level. The level division helps learners to arrange their learning plans more systematically and gradually master the required knowledge points and concepts.

[0038] After the hierarchical division, learners can obtain a set of candidate knowledge concepts at each level. These candidate knowledge concepts are the key points that learners need to learn before mastering the target knowledge concepts. Learners can gradually learn these candidate knowledge concepts according to their actual situation and learning plan, and finally master the target knowledge concepts.

[0039] Step 120, looping through the following steps until the current knowledge state indicates that the target knowledge concept has been mastered: based on the current knowledge state, determining a current concept combination from each candidate knowledge concept, performing exercises recommendation based on the current concept combination, obtaining current recommended exercises, and updating the current knowledge state based on the answer status of the recommended exercises; Step 130 , recommending a learning path based on the recommended exercises at each time.

[0040] Specifically, after obtaining each candidate knowledge concept, concept combination recommendation and exercise recommendation are performed alternately until the learner achieves the learning goal, that is, the learner's current knowledge state indicates that the target knowledge concept has been mastered.

[0041] For the current concept combination recommendation, the knowledge points or concept combinations suitable for the learner's current level will be screened out from the candidate knowledge concepts according to the current knowledge status. The current knowledge status refers to the learner's understanding and mastery of the knowledge or concepts learned at a specific point in time. This status can be measured in many ways, such as test scores, learning progress, depth of understanding of knowledge points, etc. The current concept combination is a group of knowledge points or concept combinations suitable for the learner's current level selected from the candidate knowledge concepts. This combination is designed to help learners further consolidate the knowledge they have mastered and challenge new learning content.

[0042] Recommended exercises for the current moment are implemented based on the current concept combination. Recommended exercises are designed to help learners deepen their understanding and mastery of knowledge points through practice.

[0043] The answer status of the recommended exercises here can be the answer status of the learner's historical recommended exercises, including the difficulty of the questions, the correct rate, the error rate, the answer speed, etc. This information is used to evaluate the learner's mastery of the knowledge points and update the current knowledge status accordingly.

[0044] If the current knowledge status indicates that the learner has not mastered the target knowledge concept, the concept combination recommendation, exercise recommendation and knowledge status update are executed in a loop until the current knowledge status indicates that the learner has mastered the target knowledge concept. In this loop, the recommended exercises at each moment can be added to the learning path. The loop ends and the complete learning path is returned until the current knowledge status indicates that the learner has mastered the target knowledge concept.

[0045] The method provided in the embodiment of the present invention has the following advantages over the solution of recommending single knowledge concepts and exercises through the technical solution of combining concepts and recommending exercises: 1) Stronger knowledge relevance: Concept combination considers the joint recommendation of multiple related knowledge points and can reflect the intrinsic connection between knowledge. However, single knowledge concept solutions often recommend a certain knowledge point in isolation and ignore the comprehensive application of knowledge.

[0046] 2) More efficient learning paths: By recommending combinations of related concepts, learners can master multiple related knowledge points in one stage, reducing the steps and time of learning and avoiding learning gaps and inefficiencies caused by frequent switching of knowledge points.

[0047] 3) Adapting to complex learning needs: Problems in many disciplines require the comprehensive application of multiple concepts. The recommendation method of concept combination can better meet these complex scenarios, while single concept retrieval may be difficult to cope with complex problems across domains or knowledge modules.

[0048] 4) Reduce repeated learning between knowledge points: By dividing the concepts of precursor knowledge into levels and then recommending concept combinations, learners can avoid repeatedly learning the same knowledge point at different stages, thereby optimizing the learning path design.

[0049] Figure 2 This is the second flow chart of the learning path recommendation method provided by the present invention. Figure 2 As shown, (a) is a knowledge structure diagram, which shows the predecessor and successor relationship between knowledge concepts, where c1 is the predecessor concept of c3, because logically c1 is the learning basis of c3, and c3 is the successor concept of c1. (b) is the process of the single concept recommendation and question retrieval method in the related art. The learning path obtained based on the recommended exercises is: 12+6, 9-4, 5×16, 81÷9, 6+27÷3, that is, at least five retrieval questions are needed to practice and learn c5. (c) is the recommendation process of concept combination and exercise generation provided by the present invention. The learning path obtained based on the recommended exercises is: 34+7-15, 8×9÷3. Under ideal circumstances, only two exercises are needed to effectively learn c5.

[0050] Based on any of the above embodiments, in step 110, the precursor knowledge concepts are hierarchically divided to obtain candidate knowledge concepts, specifically including: Step 111 , based on the predecessor-successor relationship between the predecessor knowledge concepts, hierarchical division is performed on the predecessor knowledge concepts to obtain candidate knowledge concepts.

[0051] Specifically, a predecessor relationship refers to a knowledge point or concept being the premise and foundation of another knowledge point or concept, which can include direct predecessor relationships and indirect predecessor relationships. A direct predecessor relationship refers to a knowledge point or concept being the direct foundation of another knowledge point or concept; an indirect predecessor relationship refers to a knowledge point or concept indirectly providing a foundation for another knowledge point or concept through other knowledge points or concepts.

[0052] A successor relationship means that a certain knowledge point or concept is an extension, extension or application of another knowledge point or concept. A successor relationship indicates that the learning and understanding of a knowledge point or concept needs to be based on another knowledge point or concept. The predecessor-successor relationship between predecessor knowledge concepts can be identified through the constructed knowledge structure diagram.

[0053] Based on the predecessor-successor relationship, the predecessor knowledge concepts are arranged in a logical order and difficulty level to form a hierarchical structure. Each level contains the knowledge points or concepts that need to be mastered at that level, and these knowledge points or concepts are expanded and extended based on the knowledge points or concepts of the previous level.

[0054] After the hierarchical structure is initially constructed, it can be adjusted and optimized according to actual conditions. For example, if it is found that the logical relationship between certain knowledge points or concepts is not clear or there is overlap, it may be necessary to reclassify them or adjust the hierarchical order.

[0055] In addition, considering that different learners may have different learning styles, learning habits and learning foundations, when constructing the hierarchical structure, it is necessary to fully consider the individual differences of learners and provide personalized candidate knowledge concepts for different learners.

[0056] Based on any of the above embodiments, Figure 3 is a flow chart of step 110 in the learning path recommendation method provided by the present invention, such as Figure 3 As shown, based on the predecessor-successor relationship between the predecessor knowledge concepts, the predecessor knowledge concepts are hierarchically divided to obtain candidate knowledge concepts, including: Step 111 - 1 , based on the predecessor-successor relationship between the predecessor knowledge concepts, construct a hierarchical knowledge tree with the target knowledge concept as the root node and the predecessor knowledge concept as the leaf node; Step 111-2, dividing the hierarchical knowledge tree into hierarchical levels to obtain nodes included in each level; Step 111 - 3 , based on the nodes included in each level, determine the candidate knowledge concepts corresponding to each level.

[0057] Specifically, the candidate knowledge concepts can be realized by constructing a knowledge tree. The knowledge structure diagram is recorded as , where V represents the set of knowledge concepts and E represents the relationship between them (i.e., edge). First, a hierarchical knowledge tree is constructed with the target knowledge concept as the root node and the predecessor knowledge concept as the leaf node, and then it is divided hierarchically, treating each parent node and its child nodes as candidate spaces.

[0058] Specifically, the target knowledge concept is recorded as , by searching all its adjacent nodes in the graph G to determine its direct predecessor, and recursively expand outward until the required depth d is reached, and the predecessor knowledge concept of the target knowledge concept is obtained. This traversal forms a network with Hierarchical tree structure with , the formula is as follows: in, Represents a knowledge concept. Represents the predecessor and successor relationship of knowledge concepts, where the depth d of the tree is dynamically determined according to the learner's current learning level.

[0059] Once the hierarchical knowledge tree is generated , it is divided into candidate concept spaces by hierarchical partitioning method for concept combination recommendation. The candidate concept space here refers to candidate knowledge concepts.

[0060] By adopting a hierarchical partitioning strategy, the nodes of each layer correspond to a set of candidate concepts that can be grouped together. For any layer k, the candidate set Defined as the set of child nodes of all nodes in the previous layer: in, represents a concept in the kth layer, and i is the index of the concept in this layer.

[0061] The nodes contained in each level can be determined as candidate knowledge concepts corresponding to each level. In this way, a candidate space is formed at each level. , each candidate knowledge concept package Both can be used for subsequent concept package recommendations.

[0062] The method provided by the embodiment of the present invention obtains candidate knowledge concepts by constructing a hierarchical knowledge tree with the target knowledge concept as the root node and the predecessor knowledge concept as the leaf node, and adopts a strategy of hierarchically dividing the hierarchical knowledge tree, thereby improving the efficiency and accuracy of subsequent concept combinations.

[0063] Based on any of the above embodiments, Figure 4 is a flow chart of step 120 in the learning path recommendation method provided by the present invention, such as Figure 4 As shown, in step 120, based on the knowledge state at the current moment, the concept combination at the current moment is determined from each candidate knowledge concept, specifically including: Step 121, based on the mastery level of the concept combination at the previous moment, determine the candidate knowledge concept at the current moment from each candidate knowledge concept; Step 122, based on the current knowledge state and the current candidate knowledge concepts, determine the current concept combination through a large language model.

[0064] Specifically, for the determination of concept combinations, it is necessary to consider the learner's current knowledge state and the candidate knowledge concepts at the current moment. Here, the current knowledge state can use the knowledge tracking model to model the learner's historical answer log to obtain the learner's current knowledge state, which is expressed as: In the formula, represents the learner's knowledge state at the current time t, KT represents the knowledge tracking model, Represents the learner's historical answer log, e represents the question, s represents the answer result, if s=1 means the answer is correct, s=0 means the answer is wrong.

[0065] The candidate knowledge concepts at the current moment are determined from the candidate knowledge concepts based on the mastery of the concept combination at the previous moment. If the mastery of the concept combination at the previous moment is higher and more proficient, the candidate knowledge concepts at the current moment can be a higher level of the concept combination at the previous moment, and continue to learn towards the target knowledge concepts step by step; if the mastery of the concept combination at the previous moment is not ideal and does not achieve the expected effect, the candidate knowledge concepts at the current moment can be a lower level of the concept combination at the previous moment to further consolidate the basic knowledge.

[0066] Here, the mastery level of the concept combination at the previous moment can be understood as the knowledge state at the previous moment, which can also be obtained through the knowledge tracking model.

[0067] After obtaining the current knowledge state and the current candidate knowledge concepts, the current concept combination can be determined through the large language model. The current knowledge state and the current candidate knowledge concepts are input into the large language model, which combines the current knowledge state to select the most suitable concept combination for the learner from the current candidate knowledge concepts and outputs the current concept combination. It can be expressed as: in is the concept combination at time t, is the candidate concept at time t, is the knowledge state at time t. LLM can be any large language model that supports input and output, such as the Spark model.

[0068] The method provided by the embodiment of the present invention determines the concept combination at the current moment through a large language model based on the current knowledge state and the candidate knowledge concepts at the current moment, fully considers the current knowledge state of the learner, and further improves the rationality and applicability of the concept combination recommendation.

[0069] Based on any of the above embodiments, based on the mastery level of the concept combination at the previous moment, the candidate knowledge concept at the current moment is determined from each candidate knowledge concept, that is, step 121 specifically includes: Based on the mastery degree of the concept combination at the previous moment, the candidate knowledge concept of the previous level or the candidate knowledge concept of the next level at the level where the candidate knowledge concept at the previous moment is located is determined as the candidate knowledge concept at the current moment.

[0070] Specifically, the determination of the candidate knowledge concepts at the current moment can be determined by the mastery of the concept combination at the previous moment. A mastery threshold is set in advance. If the mastery of the concept combination at the previous moment is greater than the mastery threshold, the candidate knowledge concept at the previous level (i.e., a higher level) of the level where the candidate knowledge concept at the previous moment is located is used as the candidate knowledge concept at the current moment; if the mastery of the concept combination at the previous moment is less than or equal to the mastery threshold, the candidate knowledge concept at the next level (i.e., a lower level) of the level where the candidate knowledge concept at the previous moment is located is used as the candidate knowledge concept at the current moment. It can be expressed as: in, represents the candidate knowledge concept at the current time t, assuming that the concept combination at time t-1 is in the candidate knowledge concept at the kth layer In, Represents a higher level of knowledge tree hierarchical division, Represents a lower level of the knowledge tree hierarchical division. Indicates the student's concept combination at time t-1 The threshold represents the preset mastery level threshold.

[0071] Based on any of the above embodiments, in step 120, exercises are recommended based on the current concept combination to obtain the current recommended exercises, specifically including: Step 123, based on the concept combination at the current moment and the knowledge state at the current moment, generate the recommended exercises at the current moment through a large language model.

[0072] Specifically, after obtaining the concept combination at the current moment, it is also necessary to recommend exercise questions to students based on their knowledge status at the current moment t.

[0073] In the related art, relevant exercises are usually retrieved from the question bank. Since the number of questions in the question bank is limited, it may be difficult to find suitable exercises that meet the different needs of learners among the available choices. If students want to learn exercises on a specific combination of knowledge points, such as students want to learn addition and subtraction at the same time, there may be no questions in the limited question bank that test these two knowledge concepts at the same time.

[0074] The traditional question retrieval strategy may be limited by the number of questions in the question bank, so the embodiment of the present invention adopts a question generation strategy to break this limitation. Based on the current concept combination and the current knowledge state, the powerful generation ability of the large language model is used to generate the current recommended exercises. On the one hand, it is more flexible, and on the other hand, it also increases the diversity and novelty of the questions.

[0075] The current concept combination and the current knowledge state can be input into the large language model to generate the current recommended exercises. The formula is: In the formula, represents the recommended exercises at the current time t, is the concept combination at time t, is the knowledge state at time t.

[0076] The method provided by the embodiment of the present invention generates recommended exercises at the current moment through a large language model, which makes exercise generation more flexible and increases the diversity and novelty of the questions.

[0077] Based on any of the above embodiments, based on the current concept combination and the current knowledge state, a large language model is used to generate the current recommended exercises, specifically including: Search the question bank based on the current concept combination to obtain search questions; Based on the current concept combination, current knowledge status and retrieval exercises, a large language model is used to generate recommended exercises for the current moment.

[0078] Specifically, in addition to using large language models to freely generate recommended exercises, it is also possible to combine question retrieval to generate recommended exercises, which can ensure that the newly generated questions are as consistent as possible with the questions in the existing question bank to a certain extent.

[0079] First, we searched the question bank and examined the concept combination. Questions based on the knowledge concepts in the question bank are retrieved based on the current concept combination to obtain retrieval exercises. Although these questions may only test Some knowledge points: in, Indicates the questions in the question bank The set of knowledge concepts examined, Represents the set of retrieved topics.

[0080] Then we can use the generative power of the large language model to combine concepts at the current moment. , the student’s knowledge status at the current time t And the set of retrieved topics To generate the current recommended exercises for students to practice , expressed as: The method provided by the embodiment of the present invention combines the two methods of question bank retrieval and large language model generation to generate recommended exercises at the current moment, thereby improving flexibility while ensuring the consistency of exercises.

[0081] Based on any of the above embodiments, a learning path recommendation method is provided, including: S1, determine the predecessor knowledge concepts of the target knowledge concept, and divide the predecessor knowledge concepts into levels to obtain candidate knowledge concepts. Specifically include: Based on the predecessor-successor relationship between predecessor knowledge concepts, a hierarchical knowledge tree is constructed with the target knowledge concept as the root node and the predecessor knowledge concept as the leaf node; the hierarchical knowledge tree is divided into levels to obtain the nodes contained in each level; based on the nodes contained in each level, the candidate knowledge concepts corresponding to each level are determined.

[0082] S2, looping through the following steps until the current knowledge state indicates that the target knowledge concept has been mastered: S21, based on the current knowledge state, determine the current concept combination from each candidate knowledge concept. Specifically, it includes: based on the mastery level of the concept combination at the previous moment, determine the previous level candidate knowledge concept of the level where the previous candidate knowledge concept is located, or the next level candidate knowledge concept as the current candidate knowledge concept; based on the current knowledge state and the current candidate knowledge concept, determine the current concept combination through a large language model.

[0083] S22, based on the current moment concept combination, exercises are recommended to obtain the current moment recommended exercises. Specifically, it includes: based on the current moment concept combination and the current moment knowledge state, the current moment recommended exercises are generated through a large language model; or, based on the current moment concept combination, searching in the question bank to obtain the search exercises; based on the current moment concept combination, the current moment knowledge state and the search exercises, the current moment recommended exercises are generated through a large language model.

[0084] S23, updating the current knowledge status based on the answer status of the recommended exercises.

[0085] S3, recommends learning paths based on the recommended exercises at each moment.

[0086] Based on any of the above embodiments, Figure 5 This is the third flow chart of the learning path recommendation method provided by the present invention. Figure 5As shown, the concept combination module and the question generation module are combined, and the concept combination and question generation are repeated until the students master the learning objectives. Among them, the concept combination module is responsible for recommending the most appropriate set of related knowledge concepts to the learners at each learning step of the students. Next, the exercise generation module will generate personalized exercises that are consistent with the recommended concept package and the learners' current knowledge level. The process is shown as follows: Step 1: Initialize the learner simulator and initialize the learning path to an empty set .

[0087] Step 2: Enter the loop until the learning goal is reached: Step 2.1: Evolving the learner’s knowledge state ; Step 2.2: Obtain candidate knowledge concepts ; Step 2.3: Recommend the most appropriate concept bundle ; Step 2.4: Recommend generated exercise ; Step 2.5: Practice Add to learning path middle; Step 2.6: Get exercise feedback score ; Step 2.7: Update learning records , add the exercise and score to your learning history.

[0088] Step 3: After the loop is complete, return to the complete learning path .

[0089] Based on any of the above embodiments, combined with Figure 2 (c) in the figure further describes the above steps, assuming that students start learning addition and subtraction.

[0090] 1) The initial learning path is empty.

[0091] 2) Determine that the student has not yet completed the learning goal (the goal is to learn equation, i.e. C7).

[0092] 3) Evolving the current knowledge status of students, it is found that the knowledge status of addition and subtraction is poor (addition mastery is 0.3, subtraction mastery is 0.2). Here, the highest mastery is 1, the lowest is 0, and above 0.8 can be considered as good mastery.

[0093] 4) Obtain candidate knowledge concepts: addition, subtraction, multiplication (multiplication is the root node, addition and subtraction are leaf nodes, which are the candidate knowledge concepts of the current layer).

[0094] 5) Recommended concept combinations: addition, subtraction.

[0095] 6) Generate personalized questions that test addition and subtraction at the same time: 34+7-15.

[0096] 7) Add 34+7-15 to the learning path.

[0097] 8) The student answers 34+7-15 and receives feedback on the answer. Assume the student's answer is correct.

[0098] 9) Update learning records.

[0099] 10) Determine if the student has not yet completed their learning objectives.

[0100] ...and so on, eventually returning to the complete learning path.

[0101] The learning path recommendation device provided by the present invention is described below. The learning path recommendation device described below and the learning path recommendation method described above can be referred to each other.

[0102] Based on any of the above embodiments, Figure 6 is a structural diagram of the learning path recommendation device provided by the present invention, such as Figure 6 As shown, the device comprises: A hierarchical division unit 610 is used to determine the predecessor knowledge concepts of the target knowledge concept and hierarchically divide the predecessor knowledge concepts to obtain candidate knowledge concepts; The state updating unit 620 is used to loop through the following steps until the current knowledge state indicates that the target knowledge concept has been mastered: based on the current knowledge state, determining a current concept combination from the candidate knowledge concepts, performing exercise recommendations based on the current concept combination to obtain current recommended exercises, and updating the current knowledge state based on the answer status of the recommended exercises; The path recommendation unit 630 is used to recommend a learning path based on the recommended exercises at each moment.

[0103] The device provided by the embodiment of the present invention determines that the concept combinations determined from each candidate knowledge concept are interrelated and have internal correlation, and exercises are recommended based on the concept combinations, so that a single exercise can simultaneously learn and solve at least one single concept, which can better reduce the learner's practice burden and promote a deeper understanding of interrelated knowledge concepts.

[0104] Based on any of the above embodiments, the hierarchical division unit is specifically used for: Based on the predecessor-successor relationship between the predecessor knowledge concepts, the predecessor knowledge concepts are hierarchically divided to obtain candidate knowledge concepts.

[0105] Based on any of the above embodiments, the hierarchical division unit is specifically used for: Based on the predecessor-successor relationship between the predecessor knowledge concepts, construct a hierarchical knowledge tree with the target knowledge concept as the root node and the predecessor knowledge concept as the leaf node; Dividing the hierarchical knowledge tree into hierarchical levels to obtain nodes included in each level; Based on the nodes included in each level, candidate knowledge concepts corresponding to each level are determined.

[0106] Based on any of the above embodiments, the state updating unit is specifically used for: Based on the mastery degree of the concept combination at the previous moment, determining the candidate knowledge concept at the current moment from the candidate knowledge concepts; Based on the current knowledge state and the current candidate knowledge concepts, a current concept combination is determined through a large language model.

[0107] Based on any of the above embodiments, the state updating unit is specifically used for: Based on the mastery degree of the concept combination at the previous moment, the candidate knowledge concept at the previous level or the candidate knowledge concept at the next level is determined as the candidate knowledge concept at the current moment.

[0108] Based on any of the above embodiments, the state updating unit is specifically used for: Based on the concept combination at the current moment and the knowledge state at the current moment, the recommended exercises at the current moment are generated through a large language model.

[0109] Based on any of the above embodiments, the state updating unit is specifically used for: Searching in the question bank based on the current concept combination to obtain search questions; Based on the current concept combination, the current knowledge state and the search questions, the current recommended questions are generated through a large language model.

[0110] Figure 7 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 7As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730 and a communication bus 740, wherein the processor 710, the communication interface 720 and the memory 730 communicate with each other through the communication bus 740. The processor 710 may call the logic instructions in the memory 730 to execute the learning path recommendation method, which includes: determining the predecessor knowledge concept of the target knowledge concept, and hierarchically dividing the predecessor knowledge concept to obtain each candidate knowledge concept; looping the following steps until the current moment knowledge state indicates that the target knowledge concept has been mastered: based on the current moment knowledge state, determining the current moment concept combination from the candidate knowledge concepts, recommending exercises based on the current moment concept combination, obtaining the current moment recommended exercises, and updating the current moment knowledge state based on the answer status of the recommended exercises; and recommending learning paths based on the recommended exercises at each moment.

[0111] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0112] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the learning path recommendation method provided by the above-mentioned methods, which includes: determining the predecessor knowledge concept of the target knowledge concept, and hierarchically dividing the predecessor knowledge concept to obtain each candidate knowledge concept; looping the following steps until the current knowledge state indicates that the target knowledge concept has been mastered: based on the current knowledge state, determining the current concept combination from the candidate knowledge concepts, recommending exercises based on the current concept combination to obtain recommended exercises at the current moment, and updating the current knowledge state based on the answer status of the recommended exercises; and recommending learning paths based on the recommended exercises at each moment.

[0113] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the learning path recommendation method provided by the above-mentioned methods, the method comprising: determining the predecessor knowledge concept of the target knowledge concept, and hierarchically dividing the predecessor knowledge concept to obtain candidate knowledge concepts; looping the following steps until the knowledge state at the current moment indicates that the target knowledge concept has been mastered: based on the knowledge state at the current moment, determining the current moment concept combination from the candidate knowledge concepts, recommending exercises based on the current moment concept combination to obtain recommended exercises at the current moment, and updating the current moment knowledge state based on the answer status of the recommended exercises; and recommending learning paths based on the recommended exercises at each moment.

[0114] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0115] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A learning path recommendation method, characterized in that: include: Determine the predecessor knowledge concepts of the target knowledge concept, and hierarchically divide the predecessor knowledge concepts to obtain candidate knowledge concepts; The following steps are executed in a loop until the current knowledge state indicates that the target knowledge concept has been mastered: based on the current knowledge state, determining a current concept combination from the candidate knowledge concepts, performing exercises recommendation based on the current concept combination to obtain current recommended exercises, and updating the current knowledge state based on the answer status of the recommended exercises; Recommend learning paths based on recommended exercises at each moment.

2. The learning path recommendation method according to claim 1, characterized in that: The step of hierarchically dividing the precursor knowledge concepts to obtain candidate knowledge concepts includes: Based on the predecessor-successor relationship between the predecessor knowledge concepts, the predecessor knowledge concepts are hierarchically divided to obtain candidate knowledge concepts.

3. The learning path recommendation method according to claim 2, characterized in that: The step of hierarchically dividing the predecessor knowledge concepts based on the predecessor-successor relationship between the predecessor knowledge concepts to obtain candidate knowledge concepts includes: Based on the predecessor-successor relationship between the predecessor knowledge concepts, construct a hierarchical knowledge tree with the target knowledge concept as the root node and the predecessor knowledge concept as the leaf node; Dividing the hierarchical knowledge tree into hierarchical levels to obtain nodes included in each level; Based on the nodes included in each level, candidate knowledge concepts corresponding to each level are determined.

4. The learning path recommendation method according to any one of claims 1 to 3, characterized in that: The determining of the current moment concept combination from the candidate knowledge concepts based on the current moment knowledge state includes: Based on the mastery degree of the concept combination at the previous moment, determining the candidate knowledge concept at the current moment from the candidate knowledge concepts; Based on the current knowledge state and the current candidate knowledge concepts, a current concept combination is determined through a large language model.

5. The learning path recommendation method according to claim 4, characterized in that: The step of determining the candidate knowledge concepts at the current moment from the candidate knowledge concepts based on the mastery degree of the concept combination at the previous moment includes: Based on the mastery degree of the concept combination at the previous moment, the candidate knowledge concept at the previous level or the candidate knowledge concept at the next level is determined as the candidate knowledge concept at the current moment.

6. The learning path recommendation method according to any one of claims 1 to 3, characterized in that: The step of performing exercise recommendation based on the current concept combination to obtain the current recommended exercise includes: Based on the concept combination at the current moment and the knowledge state at the current moment, the recommended exercises at the current moment are generated through a large language model.

7. The learning path recommendation method according to claim 6, characterized in that: The generating the recommended exercises at the current moment through a large language model based on the current concept combination and the current knowledge state includes: Searching in the question bank based on the current concept combination to obtain search questions; Based on the current concept combination, the current knowledge state and the search questions, the current recommended questions are generated through a large language model.

8. A learning path recommendation device, characterized in that: include: A hierarchical division unit, used to determine the predecessor knowledge concepts of the target knowledge concept, and hierarchically divide the predecessor knowledge concepts to obtain candidate knowledge concepts; The state updating unit is used to cyclically execute the following steps until the current knowledge state indicates that the target knowledge concept has been mastered: based on the current knowledge state, determine the current concept combination from the candidate knowledge concepts, recommend exercises based on the current concept combination to obtain the current recommended exercises, and update the current knowledge state based on the answer status of the recommended exercises; The path recommendation unit is used to recommend learning paths based on recommended exercises at each moment.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the learning path recommendation method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the learning path recommendation method according to any one of claims 1 to 7 is implemented.

Images