An adaptive student after-school exercise question recommendation method and system

Abstract

This invention discloses an adaptive method and system for recommending after-school exercises to students. Based on student answer steps, a relational graph structure is obtained. A learning association discriminant network is constructed based on this relational graph structure. By using the found relational graph structure to represent the relationships between student answer steps, the relationships between the knowledge points to be tested are identified. The learning association discriminant network is designed using the relational graph structure so that when a set of knowledge points to be tested is input, the relationships between these points are used to recommend corresponding exercises.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a self-adaptive student after-school exercise recommendation method and system. BACKGROUND

[0002] At present, because a exercise contains multiple knowledge points, the exercise and the corresponding answer can be deconstructed to examine multiple learning knowledge points of students. However, in the process of answering, in addition to directly selecting the selected questions and judging the judgment questions, the statement questions and the answer questions all need multiple steps to make judgments, and each step corresponds to different knowledge points. When an error occurs in a step, it may be an error in the previous step, so it is necessary to find the association relationship between multiple steps and find the knowledge points. However, how to find the relationship between the steps and design the network to make it more accurate is a problem. SUMMARY

[0003] The present application relates to the technical field of computer, in particular to a self-adaptive student after-school exercise recommendation method and system.

[0004] In a first aspect, the embodiments of the present application provide a self-adaptive student after-school exercise recommendation method, comprising:

[0005] Obtaining p student answer steps; the student answer step represents the step of the student answering an exercise; the student answer step is associated with multiple student mastered knowledge points;

[0006] Discriminating the correctness and error of the p student answer steps to obtain q to-be-detected mastered sets; the to-be-detected mastered set includes multiple to-be-detected mastered knowledge points; the to-be-detected mastered knowledge point represents a knowledge point that the student does not master in the exercise;

[0007] Based on the p student answer steps, an association graph structure is obtained; the association graph structure represents the association relationship of the multiple student answer steps;

[0008] Based on the association graph structure, a learning association discriminant network is constructed; the learning association discriminant network is used to discriminate the association of the to-be-detected mastered knowledge points in the student answer step;

[0009] Through the learning association discriminant network, based on the q to-be-detected mastered sets and the association graph structure, a recommended exercise is obtained; the recommended exercise represents an exercise containing the to-be-detected mastered knowledge points and the association relationship thereof.

[0010] Optionally, the association graph structure based on the p student answer steps comprises:

[0011] In the student answering step, keywords are extracted to obtain an answer keyword set; the answer keyword set includes multiple answer keywords; the answer keywords represent keywords associated with the knowledge points mastered by the students;

[0012] The number of the union set of the two answer keyword sets is calculated to obtain an association degree value; p student answering steps correspond to p*(p-1) / 2 association degree values;

[0013] The student answering steps are taken as points, and the association degree values are taken as edges to construct a graph structure to obtain an association graph structure; the association graph structure represents the association relationship between the two answering steps.

[0014] Optionally, the learning association discriminant network is constructed based on the association graph structure, including:

[0015] The learning association discriminant network includes multiple switch structures, a first neural network, a second neural network, a third neural network, and an association structure.

[0016] The multiple to-be-detected mastery sets are input into the first neural network to extract the features between the to-be-detected mastery knowledge points to obtain knowledge association features.

[0017] The knowledge association features are stored in the switch structure; the switch value of the switch structure is used to control whether the knowledge association features are output.

[0018] The input of the second neural network is the output of the switch structure; the input of the association structure is the output of the second neural network and the output of the switch structure; the input of the third neural network is the output of the association structure; and the third neural network is a recommended exercise.

[0019] Optionally, the to-be-detected mastery sets are fully connected with c neurons in the first neural network; q to-be-detected mastery sets correspond to q*c neurons.

[0020] The input of one switch structure is the knowledge association features output by one first neural network and the corresponding association degree value.

[0021] If the edge in the association graph structure is 0, the switch value of the switch structure is set to 0; if the edge in the association graph structure is greater than 0, the switch value of the switch structure is set to 1.

[0022] Optionally, the output of one switch structure corresponds to m neurons of the second neural network; and p switch structures correspond to m*p neurons in the second neural network.

[0023] If the knowledge association features output by the switch structure with the switch value of 0 are all 0.

[0024] Optionally, the input of the association structure is the first feature vector output by the second neural network and the association degree value output by the switch structure.

[0025] In the association structure, the first feature vector is multiplied by the corresponding association degree value to obtain a second feature vector.

[0026] The second feature vector is input into a third neural network to obtain a recommended exercise; the recommended exercise represents an exercise containing the knowledge not mastered by the user.

[0027] Optionally, if the association degree value in the switch structure is equal to 0, the switch value is set to 0, and if the association degree value is greater than 0, the switch value 1 is set to 1.

[0028] The switch value of 1 indicates that the input knowledge association feature and the association degree value are output; the switch value of 0 indicates that the input knowledge association feature and the association degree value are not output.

[0029] Optionally, the training method of the learning association discriminant network comprises:

[0030] Obtain training data; the training data includes historical data and labeled data; the historical data represents historical knowledge points not mastered; the labeled data represents exercises containing knowledge points in the historical data and the knowledge points have corresponding association relationships;

[0031] Input the historical data into the learning association discriminant network, extract features, and obtain training recommended exercises;

[0032] Calculate the loss of the training recommended exercises and the labeled data, and backward train the switch structure, the first neural network, the second neural network, the third neural network, and the association structure in the learning association discriminant network.

[0033] Optionally, the step of discriminating the correctness and error of the p student answers comprises:

[0034] Obtain a standard answer; the standard answer contains multiple standard steps; the multiple standard steps correspond to multiple student mastered knowledge point sets; one student mastered knowledge point set includes multiple student mastered knowledge points.

[0035] Match the standard steps with the corresponding student answer steps to discriminate the correctness and error of the student answer steps.

[0036] If the student answer step is incorrect, the student mastered knowledge points corresponding to the standard step corresponding to the student answer step are regarded as the detected mastered knowledge points; one student answer step corresponds to multiple detected mastered knowledge points.

[0037] A plurality of to-be-detected mastered knowledge points form a to-be-detected mastered set; q erroneous student answer steps correspond to obtaining q to-be-detected mastered sets.

[0038] In a second aspect, the embodiment of the present application provides a self-adaptive student after-class exercise recommendation system, comprising:

[0039] An acquisition module is configured to acquire p student answer steps; the student answer steps represent steps of student answers to an exercise; and the student answer steps are associated with a plurality of student mastered knowledge points;

[0040] An erroneous step detection module is configured to determine the correctness and the error of the p student answer steps, and obtain q to-be-detected mastered sets; the to-be-detected mastered sets comprise a plurality of to-be-detected mastered knowledge points; and the to-be-detected mastered knowledge points represent knowledge points that are not mastered by the student in the exercise;

[0041] An association module is configured to obtain an association graph structure based on the p student answer steps; and the association graph structure represents the association relationship of the plurality of student answer steps;

[0042] A network construction module is configured to construct a learning association discriminant network based on the association graph structure; and the learning association discriminant network is used to determine the association of the to-be-detected mastered knowledge points in the student answer steps;

[0043] A recommendation module is configured to obtain a recommended exercise based on the q to-be-detected mastered sets and the association graph structure through the learning association discriminant network; and the recommended exercise represents an exercise comprising the to-be-detected mastered knowledge points and the association relationship thereof

[0044] Compared with the prior art, the embodiment of the present application has the following beneficial effects:

[0045] The embodiment of the present application further provides a self-adaptive student after-class exercise recommendation method and system, finds erroneous steps in student answer steps, and then takes the student mastered knowledge points corresponding to the erroneous steps as to-be-detected mastered knowledge points that are not mastered by the student. However, there is an association relationship between the student answer steps, the next student answer step uses the data corresponding to the previous student answer step, so the association graph structure is used to represent the association relationship between the student answer steps, thereby finding the association relationship between the to-be-detected mastered knowledge points. The learning association discriminant network is designed by using the association graph structure, so that the corresponding exercise is recommended by using the association relationship when the to-be-detected mastered set is input. The technical effect of more accurately recommending the exercise of the knowledge points that are not mastered by the student is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 It is a self-adaptive student after-class exercise recommendation method flowchart provided by the embodiment of the present application.

[0047] Figure 2 This is a schematic diagram of the learning association discriminant network in an adaptive student after-school exercise recommendation method provided in an embodiment of the present invention. Detailed Implementation

[0048] The present invention will now be described in detail with reference to the accompanying drawings.

[0049] Example 1

[0050] like Figure 1 As shown in the figure, this embodiment of the invention provides an adaptive method for recommending after-school exercises for students, the method comprising:

[0051] S101: Obtain p student answer steps; each student answer step represents the steps a student takes to answer a question; each student answer step is associated with multiple knowledge points mastered by students.

[0052] In the student answering step, the associated knowledge points refer to the knowledge points that students need to master in the corresponding exercises.

[0053] The knowledge points that students master are set manually by entering questions into the database.

[0054] Where p is a positive integer.

[0055] S102: Determine the correctness and error of the answer steps of p students to obtain q sets of knowledge to be tested; the sets of knowledge to be tested include multiple knowledge points to be tested; the knowledge points to be tested represent the knowledge points that students have not mastered in the exercises.

[0056] S103: Based on p student answer steps, obtain the association graph structure; the association graph structure represents the association relationship of multiple student answer steps.

[0057] S104: Based on the aforementioned association graph structure, a learning association discriminant network is constructed; the learning association discriminant network is used to discriminate the association of the knowledge points to be mastered in the student's answer steps.

[0058] The structural diagram of the learning association discriminant network is shown below. Figure 2 As shown.

[0059] S105: By learning the association discriminant network, based on the q sets of knowledge to be tested and the association graph structure, recommended exercises are obtained; the recommended exercises represent exercises that contain the knowledge points to be tested and their association relationships.

[0060] Optionally, the process of obtaining the association graph structure based on multiple student response steps includes:

[0061] In the student answering step, keywords are extracted to obtain an answer keyword set; the answer keyword set includes multiple answer keywords; the answer keywords represent keywords related to the knowledge points mastered by the student.

[0062] In this embodiment, the exercise is "Which dynasties are included in the Northern and Southern Dynasties?" Three student answer steps were obtained: "The Northern and Southern Dynasties is a collective term for the Southern and Northern Dynasties," "The Southern Dynasties are Song, Qi, Liang, and Chen," and "The Northern Dynasties are Northern Wei, Eastern Wei, Western Wei, Northern Qi, and Northern Zhou." The student answer step "The Northern and Southern Dynasties is a collective term for the Southern and Northern Dynasties" corresponds to the set of keywords "Northern and Southern Dynasties," "Southern Dynasties," and "Northern Dynasties," and the corresponding knowledge point mastered by the student is "the meaning of the Northern and Southern Dynasties."

[0063] Calculate the number of the union of the keyword sets of each pair of answers to obtain the relevance value; n*(n-1) / 2 relevance values ​​are obtained for each intermediate step of the n answers.

[0064] For example, if the keyword set of the first answer is "Northern and Southern Dynasties", "Southern Dynasties", and "Northern Dynasties", the keyword set of the second answer is "Southern Dynasties", "Song Dynasty", "Qi Dynasty", "Liang Dynasty", and "Chen Dynasty", and the keyword set of the third answer is "Northern Dynasties", "Northern Wei Dynasty", "Eastern Wei Dynasty", "Western Wei Dynasty", "Northern Qi Dynasty", and "Northern Zhou Dynasty", then the union of the keyword sets of the first and second answers is "Southern Dynasties", with a correlation value of 1. The union of the keyword sets of the first and third answers is "Northern Dynasties", with a correlation value of 1. The union of the keyword sets of the second and third answers is empty, with a correlation value of 0.

[0065] By treating the student's answer steps as nodes and the correlation values ​​as edges, a graph structure is constructed to obtain a correlation graph structure; the correlation graph structure represents the correlation relationship between the intermediate steps of each pair of answers.

[0066] Optionally, the step of constructing a learning association discrimination network based on the association graph structure includes:

[0067] The learning association discriminant network includes multiple switch structures, a first neural network, a second neural network, a third neural network, and an association structure.

[0068] The first neural network is a deep neural network (DNN), the second neural network is a deep neural network (DNN), and the third neural network is a deep neural network (DNN) that performs classification using softmax.

[0069] Multiple sets of knowledge to be tested are input into the first neural network, and the features between each pair of knowledge points to be tested are extracted to obtain knowledge association features.

[0070] The knowledge association features are stored in a switch structure; the switch structure uses a switch value to control whether the knowledge association features are output.

[0071] The input to the second neural network is the output of the switch structure; the input to the correlation structure is the output of the second neural network and the output of the switch structure; the input to the third neural network is the output of the correlation structure; and the third neural network is the recommended exercises.

[0072] Optionally, each pair of master sets to be detected is fully connected to c neurons in the first neural network; q master sets to be detected correspond to q*c neurons.

[0073] Wherein, c is a positive integer, and the number of c is fixed, which is 20 in this embodiment.

[0074] The input to a switch structure is the knowledge association feature and the corresponding association degree value output by a first neural network.

[0075] The degree of association is recorded in the edges of the association graph structure, and the values ​​of the edges of the association graph structure are stored in the switch structure.

[0076] If the edge in the association graph structure is 0, the switch value of the switch structure is set to 0; if the edge in the association graph structure is greater than 0, the switch value of the switch structure is set to 1.

[0077] Optionally, the output of one switch structure corresponds to m neurons in the second neural network; p switch structures correspond to m*p neurons in the second neural network.

[0078] In this embodiment, m is a positive integer. The knowledge association features of the output of one switch structure are fully connected with m neurons in the second neural network, and the output is the first feature vector.

[0079] If the switch value is 0, all knowledge association features output by the switch structure will be 0.

[0080] In this embodiment, m is 128.

[0081] In this embodiment, since the value of c is 20, the knowledge association feature is a vector with 20 elements. Therefore, when the switch value is 0, the output of the switch structure is [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0].

[0082] Optionally, the input to the correlation structure is the first feature vector output by the second neural network and the correlation degree value output by the switch structure;

[0083] In the association structure, the first feature vector is multiplied by the corresponding association degree value to obtain the second feature vector.

[0084] In this case, the switch structure is fully connected to the second neural network, so the number of elements in the first feature vector is 128. The correlation value corresponding to each element in the first feature vector is multiplied to obtain the second feature vector, which also has 128 elements.

[0085] The second feature vector is input into the third neural network to obtain recommended exercises; the recommended exercises represent exercises that contain knowledge that the user has not yet mastered.

[0086] In this embodiment, the third neural network has 128 neurons in the input layer and 10 neurons in the output layer, which means that 10 exercises that match the relevant knowledge points to be tested are recommended to the user.

[0087] Optionally, if the correlation value in the switch structure is equal to 0, the switch value is set to 0; if the correlation value is greater than 0, the switch value is set to 1.

[0088] A switch value of 1 indicates that the input knowledge association features and association degree values ​​will be output; a switch value of 0 indicates that the input knowledge association features and association degree values ​​will not be output.

[0089] Optionally, the training method for the learning association discrimination network includes:

[0090] Acquire training data; the training data includes historical data and labeled data; the historical data represents previously unmastered knowledge points; the labeled data represents exercises that contain knowledge points from the historical data and that have corresponding relationships between the knowledge points.

[0091] The exercises corresponding to the labeled data are recommended exercises stored in the database.

[0092] Historical data is input into the learning association discriminant network to extract features and obtain training recommendations.

[0093] The loss is calculated using recommended training exercises and labeled data. The network is then back-trained to learn the switch structure, first neural network, second neural network, third neural network, and association structure in the association discrimination network.

[0094] In this embodiment, the cross-entropy loss function is used to calculate the loss.

[0095] Optionally, the process of determining the correctness and error of the answers of p students yields q sets of knowledge to be tested, including:

[0096] Obtain the standard answer; the standard answer contains multiple standard steps; multiple standard steps correspond to multiple sets of knowledge points mastered by students; one set of knowledge points mastered by a student includes multiple knowledge points mastered by students.

[0097] The standard answer is the standard answer to the exercise corresponding to the student's answer steps.

[0098] Where m is a positive integer.

[0099] The standard steps are matched with the corresponding student response steps to determine whether the student's response steps are correct or incorrect.

[0100] In this embodiment, because the student answer steps are text-based, keyword matching is used to determine if there are any identical keywords in the standard steps and student answer steps. If so, multiple student answer steps are judged as correct; otherwise, they are judged as incorrect. If the student answer steps are mathematical formulas, the numerical values ​​and symbols are checked sequentially. If all are the same, multiple student answer steps are judged as correct; otherwise, they are judged as incorrect.

[0101] If a student answers the steps incorrectly, the knowledge points that the student has mastered corresponding to the standard steps will be used as the knowledge points to be tested; one student's answer steps will correspond to multiple knowledge points to be tested.

[0102] Multiple knowledge points to be tested are combined into one set of knowledge to be tested; q incorrect student answers correspond to q sets of knowledge to be tested.

[0103] Wherein, q is a positive integer.

[0104] Example 2

[0105] Based on the above-described adaptive student homework recommendation method, this invention also provides an adaptive student homework recommendation system, the system comprising:

[0106] The acquisition module is used to acquire p student answer steps; each student answer step represents the steps a student takes to answer a question; each student answer step is associated with multiple knowledge points mastered by the students.

[0107] The error step detection module is used to determine the correctness and error of the answer steps of p students, and obtain q sets of mastery to be tested; the sets of mastery to be tested include multiple knowledge points to be tested; the knowledge points to be tested represent the knowledge points that students have not mastered in the exercises.

[0108] The association module is used to obtain an association graph structure based on p student answer steps; the association graph structure represents the association relationship of multiple student answer steps.

[0109] The network construction module is used to construct a learning association discriminant network based on the association graph structure; the learning association discriminant network is used to determine the association of the knowledge points to be mastered in the student's answer steps.

[0110] The recommendation module is used to obtain recommended exercises by learning the association discrimination network and based on the q sets of knowledge to be tested and the association graph structure; the recommended exercises represent exercises that contain the knowledge points to be tested and their association relationships.

[0111] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0112] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

Claims

1. An adaptive method for recommending after-school exercises for students, characterized in that, include: Obtain p student answer steps; each student answer step represents the steps a student takes to answer a question; each student answer step is associated with multiple knowledge points mastered by the students. Determine the correctness and error of the steps in the answers of p students to obtain q sets of knowledge to be tested; each set of knowledge to be tested includes multiple knowledge points to be tested; each knowledge point to be tested represents a knowledge point that the student has not mastered in the exercises. Based on p student answer steps, an association graph structure is obtained; the association graph structure represents the association relationship among multiple student answer steps; Based on the aforementioned association graph structure, a learning association discrimination network is constructed; The learning association discriminant network is used to determine the association between the knowledge points to be tested in the student's answer steps; By learning the association discriminant network, based on q objects to be detected, the set and association graph structure are mastered, and recommended exercises are obtained; The recommended exercises refer to exercises that contain the knowledge points to be tested and their relationships. The process of obtaining the association graph structure based on p student responses includes: In the student answering process, keywords are extracted to obtain a set of answer keywords; the set of answer keywords includes multiple answer keywords; the answer keywords represent keywords related to the knowledge points mastered by the student. Calculate the number of unions of the keyword sets of each pair of answers to obtain the relevance value; p students' answer steps correspond to p*(p-1) / 2 relevance values; By treating student answer steps as nodes and correlation values ​​as edges, a graph structure is constructed to obtain a correlation graph structure; the correlation graph structure represents the correlation relationship between intermediate steps of each pair of answers. The construction of a learning association discrimination network based on the association graph structure includes: The learning association discrimination network includes multiple switch structures, a first neural network, a second neural network, a third neural network, and an association structure; Multiple sets of knowledge to be tested are input into the first neural network, and the features between each pair of knowledge points to be tested are extracted to obtain knowledge association features. The knowledge association features are stored in a switch structure; the switch structure uses a switch value to control whether the knowledge association features are output. The input to the second neural network is the output of the switch structure; the input to the correlation structure is the output of the second neural network and the output of the switch structure; the input to the third neural network is the output of the correlation structure; and the third neural network is the recommended exercises.

2. The adaptive student after-school exercise recommendation method according to claim 1, characterized in that, Each pair of master sets to be detected is fully connected to c neurons in the first neural network; q master sets to be detected correspond to q*c neurons. The input to a switch structure is the knowledge-related features and corresponding correlation values ​​output by a first neural network; If the edge in the association graph structure is 0, the switch value of the switch structure is set to 0; if the edge in the association graph structure is greater than 0, the switch value of the switch structure is set to 1.

3. The adaptive student after-school exercise recommendation method according to claim 1, characterized in that, The output of one switch structure corresponds to m neurons in the second neural network; p switch structures correspond to m*p neurons in the second neural network. If the switch value is 0, all knowledge association features output by the switch structure will be 0.

4. The adaptive student after-school exercise recommendation method according to claim 1, characterized in that, The input to the correlation structure is the first feature vector output by the second neural network and the correlation degree value output by the switch structure; In the association structure, the first feature vector is multiplied by the corresponding association degree value to obtain the second feature vector; The second feature vector is input into the third neural network to obtain recommended exercises; the recommended exercises represent exercises that contain knowledge that the user has not yet mastered.

5. The adaptive student after-school exercise recommendation method according to claim 1, characterized in that, If the correlation value in the switch structure is equal to 0, set the switch value to 0; if the correlation value is greater than 0, set the switch value 1 to 1. A switch value of 1 indicates that the input knowledge association features and association degree values ​​will be output; a switch value of 0 indicates that the input knowledge association features and association degree values ​​will not be output.

6. The adaptive student after-school exercise recommendation method according to claim 1, characterized in that, The training method for the learning association discrimination network includes: Acquire training data; the training data includes historical data and labeled data; the historical data represents previously unmastered knowledge points; the labeled data represents exercises that contain knowledge points from the historical data and that have corresponding relationships between the knowledge points; Historical data is input into the learning association discriminant network to extract features and obtain training recommendations. The loss is calculated using recommended training exercises and labeled data. The network is then back-trained to learn the switch structure, first neural network, second neural network, third neural network, and association structure in the association discrimination network.

7. The adaptive method for recommending student after-school exercises according to claim 1, characterized in that, The process of determining the correctness and error of p students' answers yields q sets of knowledge to be tested, including: Obtain the standard answer; the standard answer contains multiple standard steps; multiple standard steps correspond to multiple sets of knowledge points mastered by students; one set of knowledge points mastered by a student includes multiple knowledge points mastered by students. The standard steps are matched with the corresponding student response steps to determine whether the student's response steps are correct or incorrect. If a student answers the steps incorrectly, the knowledge points that the student has mastered corresponding to the standard steps will be used as the knowledge points to be tested; one student's answer steps will correspond to multiple knowledge points to be tested. Multiple knowledge points to be tested are combined into one set of knowledge to be tested; q incorrect student answers correspond to q sets of knowledge to be tested.

8. An adaptive student after-school exercise recommendation system, characterized in that, include: The acquisition module is used to acquire p student answer steps; each student answer step represents the steps a student takes to answer a question; each student answer step is associated with multiple knowledge points mastered by the student. An error step detection module is used to determine the correctness and error of the answer steps of p students, and obtain q sets of mastery to be tested; the sets of mastery to be tested include multiple knowledge points to be tested; the knowledge points to be tested represent the knowledge points that students have not mastered in the exercises; The association module is used to obtain an association graph structure based on p student answer steps; the association graph structure represents the association relationship of multiple student answer steps; The network construction module is used to construct a learning association discrimination network based on the association graph structure; The learning association discriminant network is used to determine the association between the knowledge points to be tested in the student's answer steps; The recommendation module is used to learn the association discrimination network, master the set and association graph structure based on q test objects, and obtain recommended exercises; The recommended exercises refer to exercises that contain the knowledge points to be tested and their relationships. The process of obtaining the association graph structure based on p student responses includes: In the student answering process, keywords are extracted to obtain a set of answer keywords; the set of answer keywords includes multiple answer keywords; the answer keywords represent keywords related to the knowledge points mastered by the student. Calculate the number of unions of the keyword sets of each pair of answers to obtain the relevance value; p students' answer steps correspond to p*(p-1) / 2 relevance values; By treating student answer steps as nodes and correlation values ​​as edges, a graph structure is constructed to obtain a correlation graph structure; the correlation graph structure represents the correlation relationship between intermediate steps of each pair of answers. The construction of a learning association discrimination network based on the association graph structure includes: The learning association discrimination network includes multiple switch structures, a first neural network, a second neural network, a third neural network, and an association structure; Multiple sets of knowledge to be tested are input into the first neural network, and the features between each pair of knowledge points to be tested are extracted to obtain knowledge association features. The knowledge association features are stored in a switch structure; the switch structure uses a switch value to control whether the knowledge association features are output. The input to the second neural network is the output of the switch structure; the input to the correlation structure is the output of the second neural network and the output of the switch structure; the input to the third neural network is the output of the correlation structure; and the third neural network is the recommended exercises.

Images