A personalized knowledge tracking method based on after-class exercises and students' exercise-solving situations

Abstract

The present invention discloses a personalized knowledge tracking method based on after-class exercises and students' exercise-solving situations. By establishing a knowledge tree, it is possible to judge the knowledge points implied by the words included in the standard answers corresponding to the after-class exercise statements and the students' answers, reflecting the knowledge information that exists in the statements but is not a keyword for its existence. It can more accurately extract the structure and information contained in the knowledge, and thus more accurately extract the features corresponding to the knowledge points. By inputting the knowledge tree into the neural network in two ways, it is judged whether the knowledge points in the knowledge tree are the knowledge points understood by the students. And through the detection of the knowledge points, the nodes in the knowledge tree are extracted to obtain a standard subtree and a different subtree. According to the standard subtree and the different subtree, it is possible to more accurately distinguish the knowledge points that the students have understood and those that they have not understood, reflecting the technical effect of the students' state of mastering the knowledge points of the after-class exercises.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and more particularly, to a personalized knowledge tracking method based on after-class exercises and students' exercise-solving situations. Background Art

[0002] Currently, with the development of educational informatization, high-quality educational resources under big data can be fully utilized to improve teaching quality. In traditional evaluation methods, only the right or wrong answers of students are often considered. However, after obtaining the right or wrong answers, it is still necessary to manually judge the knowledge that students do not understand, and it is impossible to judge the knowledge points hidden under the questions or answers, nor can it judge the knowledge points that students understand based on the similar or different words in the answers. Summary of the Invention

[0003] The purpose of the present invention is to provide a personalized knowledge tracking method based on after-class exercises and students' exercise-solving situations to solve the above problems existing in the prior art.

[0004] An embodiment of the present invention provides a personalized knowledge tracking method based on after-class exercises and students' exercise-solving situations, including:

[0005] Obtaining after-class exercise statements, corresponding standard answers, students' answers, and a knowledge base; the after-class exercise statements represent the content of the questions corresponding to the students' answers; the knowledge base is a database storing multiple knowledge points corresponding to the after-class exercise statements;

[0006] Based on the after-class exercise statements and the knowledge base, constructing a tree structure to obtain a knowledge tree; the knowledge tree represents the knowledge points included in the after-class exercise statements;

[0007] Based on the standard answers and the students' answers, through the knowledge tree, discriminating the knowledge points understood and not understood by the students to obtain a standard subtree and a difference subtree;

[0008] Based on the standard subtree and the difference subtree, respectively finding the knowledge points understood by the students and the knowledge points not understood by the students.

[0009] Optionally, the step of based on the standard answers and the students' answers, through the knowledge tree, discriminating the knowledge points understood and not understood by the students to obtain a standard subtree and a difference subtree includes:

[0010] Based on the standard answers, through a first neural network, extracting the knowledge in the standard answers to obtain standard knowledge labels; the knowledge labels represent all the knowledge points in the knowledge base related to the standard answers;

[0011] Through the first neural network, obtaining students' answer keywords corresponding to the students' answers;

[0012] Among multiple knowledge trees, respectively find the corresponding knowledge trees that contain the keywords of the standard answer and the keywords of the student answer for marking, obtaining multiple standard knowledge trees and multiple answer knowledge trees;

[0013] Mark the nodes in the standard knowledge tree whose corresponding values coincide with the nodes in the answer knowledge tree, obtaining marked nodes;

[0014] Retain the marked nodes and their corresponding root nodes in the standard knowledge tree, obtaining a standard subtree;

[0015] Set the values of the nodes corresponding to the standard subtree in the answer knowledge tree to 0, obtaining a modified answer knowledge tree;

[0016] Delete the nodes in the subtree of the modified answer knowledge tree whose values are all 0, obtaining a difference subtree.

[0017] Optionally, based on the standard answer, through a first neural network, extract the knowledge in the standard answer to obtain a standard knowledge label, including:

[0018] Establish the relationship between the after-class exercise statement and the standard answer, and jointly input them into the first neural network to obtain multiple first knowledge labels;

[0019] Use the standard answer as labeled data to train a second neural network to obtain a trained second neural network;

[0020] Input the after-class exercise statement into the trained second neural network to obtain multiple second knowledge labels;

[0021] Fuse the multiple first knowledge labels and multiple second knowledge labels to obtain a knowledge label.

[0022] Optionally, the establishing the relationship between the after-class exercise statement and the standard answer, and jointly inputting them into the first neural network to obtain multiple first knowledge labels, includes:

[0023] Through a trained statement detection network, extract the keywords in the after-class exercise statement to obtain first exercise keywords;

[0024] Among multiple knowledge trees, extract the knowledge trees that contain the first exercise keywords to obtain multiple first knowledge trees;

[0025] Arrange the multiple first knowledge trees according to the hierarchical relationship to obtain a knowledge set;

[0026] Multiple first exercise keywords correspond to obtain multiple knowledge sets;

[0027] Input the multiple knowledge sets into the first neural network respectively to obtain multiple output features;

[0028] Input the multiple output features into a keyword fusion network to extract the features of the knowledge corresponding to multiple keywords, and obtain multiple first knowledge labels.

[0029] Optionally, constructing a tree structure based on the after-class exercise sentences and the knowledge base to obtain a knowledge tree, including:

[0030] Replace the after-class exercise sentences according to the corresponding labels of the characters to obtain an after-class exercise matrix;

[0031] Input the after-class exercise input sentence detection network to detect whether there are words in the knowledge base, and obtain multiple surface knowledge keywords;

[0032] Based on the knowledge base, construct the learning relationship between knowledge to obtain multiple knowledge general trees;

[0033] Find the positions of the surface knowledge keywords in the multiple knowledge general trees to obtain key point positions;

[0034] Taking the key point positions as root nodes, extract the subtrees of the root nodes to obtain knowledge subtrees;

[0035] Multiple key point positions correspond to obtain multiple knowledge subtrees.

[0036] Optionally, using the standard answer as the labeled data to train the second neural network to obtain a trained second neural network, including:

[0037] Obtain the training second historical data; the training second historical data contains multiple historical after-class exercise sentences;

[0038] Input the training second historical data into the second neural network to obtain a second predicted answer;

[0039] Calculate the loss between the second predicted answer and the standard answer, and backpropagate to train the second neural network.

[0040] Optionally, constructing the learning relationship between knowledge based on the knowledge base to obtain multiple knowledge general trees, including:

[0041] Obtain the table of contents in the knowledge base; the table of contents is the table of contents recorded in the textbook;

[0042] Taking the table of contents in the knowledge base as a root node, construct sub-nodes in sequence according to the inclusion order of multiple knowledge points under the table of contents to obtain a knowledge general tree;

[0043] Multiple tables of contents correspond to obtain multiple knowledge totals.

[0044] Optionally, use multiple historical after-class exercise sentence matrices as training data;

[0045] Use the knowledge points corresponding to the matrix of after-class exercise statements in the knowledge base as labeled data;

[0046] The two-dimensional convolution kernel performs convolution on the rows of the after-class exercise statement matrix with a stride of 1 to train the statement detection network;

[0047] The size of the two-dimensional convolution kernel is 3*2.

[0048] Optionally, extract the same labels in the first label and the second label as knowledge labels.

[0049] Optionally, based on the standard subtree and the different subtree, find the knowledge points understood by the student and the knowledge points not understood by the student respectively, including:

[0050] Take the knowledge points corresponding to all nodes in the standard subtree as the knowledge points understood by the student;

[0051] Take the knowledge points corresponding to all nodes in the different subtree as the knowledge points not understood by the student.

[0052] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0053] The embodiments of the present invention also provide a personalized knowledge tracking method based on after-class exercises and students' answering situations. The method includes: obtaining after-class exercise statements, corresponding standard answers, students' answers, and a knowledge base; the after-class exercise statements represent the content of the questions corresponding to the students' answers; the knowledge base is a database storing multiple knowledge points corresponding to the after-class exercise statements; based on the after-class exercise statements and the knowledge base, construct a tree structure to obtain a knowledge tree; the knowledge tree represents the knowledge points included in the after-class exercise statements; based on the standard answers and the students' answers, through the knowledge tree, determine the knowledge points understood and not understood by the students to obtain a standard subtree and a different subtree; based on the standard subtree and the different subtree, find the knowledge points understood by the student and the knowledge points not understood by the student respectively.

[0054] In the present invention, according to the established knowledge tree, judge the knowledge points implied by the words included in the standard answers and the students' answers corresponding to the after-class exercise statements, which reflects the knowledge information that exists in the statements but is not a keyword for its existence. It can more accurately extract the structure and included information of the knowledge, and thus more accurately extract the features corresponding to the knowledge points. By inputting the knowledge tree into the neural network in two ways, judge whether the knowledge points in the knowledge tree are the knowledge points understood by the student, and through the detection of the knowledge points, extract the nodes in the knowledge tree to obtain a standard subtree and a different subtree. According to the standard subtree and the different subtree, it can more accurately judge the knowledge points that the student has understood and not understood, reflecting the technical effect of the student's state of mastering the knowledge points of the after-class exercises. Description of the Drawings

[0055] Figure 1 It is a flowchart of a personalized knowledge tracking method based on after-class exercises and students' exercise-solving situations provided by an embodiment of the present invention. Detailed implementation manners

[0056] The present invention will be described in detail below with reference to the accompanying drawings.

[0057] Embodiment 1

[0058] As Figure 1 shown, an embodiment of the present invention provides a personalized knowledge tracking method based on after-class exercises and students' exercise-solving situations. The method includes:

[0059] S101: Obtain after-class exercise statements, corresponding standard answers, students' answers, and a knowledge base; the after-class exercise statements represent the content of the questions corresponding to the students' answers; the knowledge base is a database storing multiple knowledge points corresponding to the after-class exercise statements.

[0060] Among them, in this embodiment, the knowledge points in the knowledge base represent the corresponding keywords in the textbook, which are represented by labels, and the keywords are obtained by crawling. For example, if the after-class exercise statement is "According to the characteristics of data access, data structures are divided into which two structures", the classification of data structures corresponds to keywords including linear structure and non-linear structure.

[0061] S102: Based on the after-class exercise statements and the knowledge base, construct a tree structure to obtain a knowledge tree; the knowledge tree represents the knowledge points included in the after-class exercise statements.

[0062] S103: Based on the standard answers and the students' answers, through the knowledge tree, identify the knowledge points that the students understand and do not understand, and obtain a standard subtree and a difference subtree;

[0063] S104: Based on the standard subtree and the difference subtree, respectively find the knowledge points that the students understand and the knowledge points that the students do not understand.

[0064] Optionally, the step of based on the standard answers and the students' answers, through the knowledge tree, identifying the knowledge points that the students understand and do not understand, and obtaining a standard subtree and a difference subtree includes:

[0065] Based on the standard answers, through a first neural network, extract the knowledge in the standard answers to obtain standard knowledge labels; the knowledge labels represent all the knowledge points in the knowledge base related to the standard answers.

[0066] Among multiple knowledge trees, respectively find the corresponding knowledge trees containing the keywords of the standard answers and the keywords of the students' answers for marking, and obtain multiple standard knowledge trees and multiple answer knowledge trees.

[0067] Mark the nodes in the standard knowledge tree whose corresponding values coincide with those of the nodes in the answer knowledge tree to obtain marked nodes.

[0068] Retain the marked nodes and their corresponding root nodes in the standard knowledge tree to obtain a standard subtree.

[0069] Among them, the knowledge points corresponding to each node in the standard subtree represent the knowledge points that the student has learned and understood.

[0070] Find the part outside the standard subtree in the student's answer to obtain a difference subtree.

[0071] Among them, the knowledge points corresponding to each node in the difference subtree represent the knowledge points that the student has not learned and understood.

[0072] Optionally, based on the standard answer, extracting the knowledge in the standard answer through a first neural network to obtain a standard knowledge label includes:

[0073] Establish the relationship between the after-class exercise statement and the standard answer, and input them into the first neural network together to obtain multiple first knowledge labels.

[0074] Use the standard answer as labeled data to train a second neural network to obtain a trained second neural network;

[0075] Input the after-class exercise statement into the trained second neural network to obtain multiple second knowledge labels;

[0076] Fuse the multiple first knowledge labels and the multiple second knowledge labels to obtain a knowledge label.

[0077] Optionally, the establishing the relationship between the after-class exercise statement and the standard answer, and inputting them into the first neural network together to obtain multiple first knowledge labels includes:

[0078] Extract the keywords in the after-class exercise statement to obtain the first exercise keywords.

[0079] Among them, the statement detection network is a Convolutional Neural Networks (CNN).

[0080] In multiple knowledge trees, extract the knowledge trees containing the first exercise keywords to obtain multiple first knowledge trees.

[0081] Among them, the knowledge tree containing the first exercise keywords in the knowledge tree is used as the first knowledge tree.

[0082] Among them, because one after-class question corresponds to one standard answer, the relationship between the question and the answer is constructed by matching the knowledge tree labels.

[0083] Arrange the multiple first knowledge trees according to the hierarchical relationship to obtain a knowledge set.

[0084] Among them, the smaller the level of the first exercise keyword in the first knowledge tree, the more forward it is arranged. For example, if the level is 0, then the corresponding first knowledge tree is used as the first element in the knowledge set. If the levels are the same, the one with the smallest level number in the first knowledge tree is arranged forward. If the level numbers are the same, the one with the smaller node is arranged forward, and then they are arranged according to the detection order.

[0085] Multiple first exercise keywords correspond to obtain multiple knowledge sets;

[0086] Input the multiple knowledge sets into the first neural network respectively to obtain multiple output features.

[0087] Among them, the first neural network is a fully connected neural network (Fully Connected Neural Network, FCNN).

[0088] Input the multiple output features into the keyword fusion network, extract the features of the knowledge corresponding to the multiple keywords, and obtain multiple first knowledge labels.

[0089] Among them, the keyword fusion network is a fully connected neural network (Fully Connected Neural Network, FCNN).

[0090] Among them, the first knowledge label represents the label of a node corresponding to the after-class exercise statement in the knowledge tree.

[0091] Optionally, constructing a tree structure based on the after-class exercise statement and the knowledge base to obtain a knowledge tree includes:

[0092] Replace the after-class exercise statement according to the corresponding labels of the characters to obtain an after-class exercise matrix;

[0093] Among them, the number of rows of the after-class exercise matrix is fixed, representing the number of characters in the after-class exercise statement. The number of rows of the after-class exercise matrix does not exceed the length of the after-class exercise statement. The number of columns of the after-class exercise matrix is equal to a fixed value, representing the number of digits of the label. In this embodiment, the number of columns of the after-class exercise matrix is 3.

[0094] Among them, the after-class exercise sentences and corresponding labels are stored in the database, such as "Which two structures are data structures divided into according to the characteristics of data access" is stored in the database. The label corresponding to "number" is "001", the label corresponding to "data" is "002", the label corresponding to "result" is "003", the label corresponding to "structure" is "004", the label corresponding to "root" is "005", the label corresponding to "visit", the label corresponding to "question", the label corresponding to "007", the label corresponding to "of", the label corresponding to "008", the label corresponding to "special" is "009", the label corresponding to "point" is "010", the label corresponding to "point" is "011", the label corresponding to "for", the label corresponding to "012", the label corresponding to "which", the label corresponding to "013", the label corresponding to "two", and the label corresponding to "kind" is "015". The corresponding matrix of some after-class exercises is:

[0095] 0 0 1 0 0 2 0 0 3 0 0 4 0 0 5 0 0 2 0 0 1 0 0 2 0 0 6 0 0 7 0 0 8 0 0 9 0 1 0 0 1 1 0 1 2 0 1 3 0 1 4 0 1 5 0 0 3 0 0 4

[0096] The after-class exercises are input into a sentence detection network to detect whether there are words in the knowledge base, and a plurality of surface knowledge keywords are obtained.

[0097] Wherein, the sentence detection network is a convolutional neural network (CNN).

[0098] In the knowledge base, multiple knowledge trees are established according to the teaching content;

[0099] Find the positions of surface knowledge keywords in multiple knowledge trees and obtain the key point positions.

[0100] Among them, each knowledge tree is searched by hierarchical traversal.

[0101] Taking the key point position as the root node, the subtree of the root node is extracted as the knowledge tree.

[0102] Multiple knowledge trees are obtained corresponding to multiple key point positions.

[0103] Optionally, the adopting the standard answer as the labeled data to train the second neural network to obtain the trained second neural network includes:

[0104] Acquire training second historical data; the training second historical data includes a plurality of historical after-class exercise sentences;

[0105] The second historical data for training is input into a second neural network to obtain a second prediction answer.

[0106] Among them, the second neural network is a fully connected neural network (Fully Connected Neural Network, FCNN).

[0107] Calculate the loss between the second predicted answer and the standard answer, and backpropagate to train the second neural network.

[0108] Optionally, based on the knowledge base, construct the learning relationships between knowledge to obtain multiple knowledge general trees, including:

[0109] Obtain a table of contents in the knowledge base; the table of contents is the one recorded in the textbook;

[0110] Use the table of contents in the knowledge base as a root node, and construct child nodes in sequence according to the inclusion order of multiple knowledge points under the table of contents to obtain a knowledge general tree;

[0111] Multiple tables of contents correspond to obtaining multiple knowledge general trees.

[0112] Optionally, use multiple historical after-class exercise statement matrices as training data;

[0113] Use the knowledge points corresponding to the after-class exercise statement matrix in the knowledge base as labeled data;

[0114] The two-dimensional convolution kernel performs convolution on the rows of the after-class exercise statement matrix with a stride of 1 to train the statement detection network;

[0115] The size of the two-dimensional convolution kernel is 3*2.

[0116] Optionally, extract the same labels in the first label and the second label as knowledge labels.

[0117] Optionally, based on the standard subtree and the difference subtree, respectively find the knowledge points understood by the student and the knowledge points not understood by the student, including:

[0118] Use the knowledge points corresponding to all nodes in the standard subtree as the knowledge points understood by the student;

[0119] Use the knowledge points corresponding to all nodes in the difference subtree as the knowledge points not understood by the student.

Claims

1. A personalized knowledge tracking method based on after-class exercises and students' performance in doing the exercises, characterized in that: include: Get the after-class exercise sentences, corresponding standard answers, student answers and knowledge base; The after-class exercise sentence indicates the content of the question corresponding to the student's answer; The knowledge base is a database storing multiple knowledge points corresponding to after-class exercise sentences; Based on the after-class exercise sentences and the knowledge base, a tree structure is constructed to obtain a knowledge tree; The knowledge tree represents the knowledge points contained in the after-class exercise sentences; The method of constructing a tree structure based on the after-class exercise sentences and the knowledge base to obtain a knowledge tree includes: Replace the after-class exercise sentences according to the labels corresponding to the characters to obtain an after-class exercise matrix; Input the after-class exercise matrix into a sentence detection network to detect whether there are words in the knowledge base, and obtain multiple surface knowledge keywords; Based on the knowledge base, construct learning relationships between knowledge to obtain multiple knowledge trees; Find the positions of surface knowledge keywords in multiple knowledge trees and obtain the key point positions; Taking the key point position as the root node, extract the subtree of the root node to obtain the knowledge tree; Multiple knowledge trees are obtained corresponding to multiple key point positions; Based on the knowledge base, the learning relationship between knowledge is constructed to obtain multiple knowledge trees, including: Obtaining a directory in a knowledge base; the directory is a directory recorded in a textbook; The directory in the knowledge base is taken as a root node, and the multiple knowledge points under the directory are sequentially constructed into child nodes in the order of inclusion to obtain a total knowledge tree; Multiple directories correspond to multiple knowledge trees; Based on the standard answer and the student's answer, the knowledge points understood and not understood by the student are judged through the knowledge tree to obtain a standard subtree and a difference subtree; Based on the standard subtree and the difference subtree, respectively find the knowledge points that the students understand and the knowledge points that the students do not understand; Based on the standard answer and the student answer, the knowledge points understood and not understood by the student are judged through the knowledge tree to obtain the standard subtree and the difference subtree, including: Based on the standard answer, extract the knowledge in the standard answer through the first neural network to obtain a standard knowledge label; the standard knowledge label represents all knowledge points related to the standard answer in the knowledge base; Through the first neural network, the student answer corresponds to the student answer keyword; in multiple knowledge trees, the corresponding knowledge trees containing the standard answer keyword and the student answer keyword are found and marked respectively, to obtain multiple standard knowledge trees and multiple answer knowledge trees; Marking nodes in the standard knowledge tree whose values ​​coincide with those of nodes in the answer knowledge tree to obtain marked nodes; The marked nodes and the corresponding root nodes in the standard knowledge tree are retained to obtain a standard subtree; The value corresponding to the node corresponding to the standard subtree in the answer knowledge tree is set to 0 to obtain a modified answer knowledge tree; Delete the nodes whose values ​​in the subtree of the modified answer knowledge tree are all 0, to obtain the difference subtree; The step of extracting knowledge from the standard answer through a first neural network based on the standard answer to obtain a standard knowledge label includes: Establishing the relationship between the after-class exercise sentences and the standard answers, inputting them into the first neural network together, and obtaining a plurality of first knowledge labels; The standard answer is used as the labeled data to train the second neural network to obtain a trained second neural network; Inputting the after-class exercise sentences into the trained second neural network to obtain a plurality of second knowledge labels; Merging the plurality of first knowledge labels with the plurality of second knowledge labels to obtain a standard knowledge label; The relationship between the after-class exercise sentences and the standard answers is established, and both are input into the first neural network to obtain a plurality of first knowledge labels, including: Through the trained sentence detection network, the keywords in the sentences of the homework exercises are extracted to obtain the keywords of the first exercise; Extracting the knowledge tree containing the first exercise keyword from the plurality of knowledge trees to obtain a plurality of first knowledge trees; Arranging the plurality of first knowledge trees according to a hierarchical relationship to obtain a knowledge set; Obtain multiple knowledge sets corresponding to multiple first exercise keywords; input the multiple knowledge sets into the first neural network respectively to obtain multiple output features; The multiple output features are input into a keyword fusion network, features of knowledge corresponding to the multiple keywords are extracted, and multiple first knowledge labels are obtained.

2. A personalized knowledge tracking method based on after-class exercises and students' performance in solving problems according to claim 1, characterized in that: The method of using the standard answer as the labeled data to train the second neural network to obtain the trained second neural network includes: Acquire training second historical data; the training second historical data includes a plurality of historical after-class exercise sentences; Inputting the training second historical data into a second neural network to obtain a second prediction answer; The loss is calculated by comparing the second predicted answer with the standard answer, and the second neural network is trained by backpropagation.

3. A personalized knowledge tracking method based on after-class exercises and students' performance in solving problems according to claim 1, characterized in that: Using multiple historical after-class exercise sentence matrices as training data; The knowledge points corresponding to the after-class exercise sentence matrix in the knowledge base are used as annotation data; The two-dimensional convolution kernel is convolved on the rows of the after-class exercise sentence matrix with a step size of 1 to train the sentence detection network; the size of the two-dimensional convolution kernel is 3*2.

4. A personalized knowledge tracking method based on after-class exercises and students' performance in solving problems according to claim 1, characterized in that: The same label between the first knowledge label and the second knowledge label is extracted as the standard knowledge label.

5. The personalized knowledge tracking method based on after-class exercises and students' performance in solving the exercises according to claim 1 is characterized in that: The method of finding the knowledge points that the students understand and the knowledge points that the students do not understand based on the standard subtree and the difference subtree includes: Taking the knowledge points corresponding to all nodes in the standard subtree as the students' understanding knowledge points; The knowledge points corresponding to all the nodes in the difference subtree are regarded as the knowledge points that the students do not understand.

Images