Deep knowledge tracking method based on forgetting behavior in multi-knowledge-point topic
By constructing an in-depth knowledge tracking method for forgetful behavior in multi-knowledge points, the problem that existing systems cannot effectively deal with multi-knowledge points and forgetful behavior is solved, and more accurate knowledge status prediction is achieved.
Patent Information
- Application Number
- CN202411767561.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-04
- Publication Date
- 2025-03-28
AI Technical Summary
The existing in-depth knowledge tracking system cannot effectively handle the situation where multiple knowledge points are contained in one exercise data, resulting in low prediction accuracy and neglecting the impact of forgetful behavior in the learning process on the knowledge state.
A deep knowledge tracking method based on forgetful behavior in multi-knowledge points questions is proposed. By obtaining answering data, building a training set, setting input modules, status update modules and prediction modules, building a knowledge tracking model, and training to update model parameters, and finally predicting answering based on the optimal knowledge model.
By considering multiple knowledge points and forgetful behaviors, the in-depth knowledge tracking system can more accurately predict students' current knowledge status and improve prediction accuracy.
Smart Images
Figure CN119721206A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data mining, and in particular to a deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions. Background Art
[0002] With the popularity of online education, intelligent tutoring systems have developed rapidly. Online education platforms have collected a large amount of student online learning data, providing favorable conditions for the development of educational data mining, among which knowledge tracking is an important research direction in this field. Knowledge tracking can evaluate students' current knowledge status based on their historical learning records and track changes in their knowledge status. Teachers can provide more targeted teaching based on students' knowledge status, and personalized education is also the main development direction of future education.
[0003] Existing deep knowledge tracking systems directly use students' answer records as input to the knowledge tracking system, and then predict the students' current knowledge status. Most deep knowledge tracking systems assume that there is only one knowledge point behind an exercise, but this is not a realistic assumption. Most exercises contain multiple knowledge points, such as an elementary school arithmetic problem containing addition and subtraction. In fact, students often learn the knowledge points contained in the exercises rather than the exercises themselves. After mastering these knowledge points, students can correctly answer new questions that have never been seen but test the knowledge points. At the same time, these deep knowledge tracking systems do not take into account the impact of forgetting behavior in the learning process on the knowledge status. The learning process is a process of memory and forgetting. According to the forgetting theory in educational psychology: forgetting behavior will occur in the learning process, and the time interval and number of times students learn the same knowledge point will affect the degree of forgetting.
[0004] Therefore, the current deep knowledge tracking solutions do not meet the needs of actual scenarios. Summary of the invention
[0005] In view of this, in order to solve the technical problem that the existing deep knowledge tracking method cannot consider the situation that one exercise data contains multiple knowledge points, which leads to low prediction accuracy, the present invention proposes a deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions, and the method includes the following steps:
[0006] Obtain answer data, cut the answer sequence to a preset length, and construct a training set;
[0007] Set up input module, state update module and prediction module to build knowledge tracking model;
[0008] Input the training set into the knowledge tracking model, perform training, update the model parameters, and obtain the optimal knowledge tracking model;
[0009] Predict answers based on the optimal knowledge model.
[0010] In some embodiments, the step of training the knowledge tracking model based on the training set, updating parameters, and obtaining the optimal knowledge tracking model specifically includes:
[0011] Embed the questions in the answer data, represent the questions with vectors, and obtain the question vectors;
[0012] Use knowledge points to represent questions and get the relevant weights of questions and knowledge points;
[0013] For the question represented by the knowledge points, calculating forgetting behavior data of multiple knowledge points to obtain forgetting features;
[0014] Considering the influence of the updated knowledge state and forgetting characteristics, the answer prediction is performed in combination with the question vector to obtain the prediction data;
[0015] Update the knowledge state in reverse according to the prediction results;
[0016] By comparing the predicted data with the actual corresponding answer records in the training set, the loss value is calculated and the parameters are updated using back propagation to obtain the optimal knowledge tracking model.
[0017] In some embodiments, it also includes:
[0018] The relevant weights are filtered, and the values greater than the threshold are set to 1, indicating that the question examines this knowledge point; the values less than the threshold are set to 0, indicating that the question does not examine this knowledge point.
[0019] The present invention also proposes a deep knowledge tracking system based on forgetting behavior in multi-knowledge point questions, the system comprising:
[0020] A data acquisition unit, which acquires answer data, cuts the answer sequence into a preset length, and constructs a training set;
[0021] The model building unit sets the input module, the state update module and the prediction module to build the knowledge tracking model;
[0022] The model training unit inputs the training set into the knowledge tracking model, performs training, updates the model parameters, and obtains the optimal knowledge tracking model;
[0023] The model prediction unit predicts the answer based on the optimal knowledge model.
[0024] Based on the above scheme, the present invention provides a deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions, builds a prediction network, represents the input student exercises with the knowledge points contained in the exercises, and calculates the forgetting information of these knowledge points (such as how many times the students have learned the same knowledge point, and what is the time interval between the last time they learned the knowledge point). These forgetting information will play an important role in the deep knowledge tracking system; representing the questions with knowledge points and calculating the forgetting behavior data of the knowledge points, this deep knowledge tracking model can more accurately predict the students' current knowledge status. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 It is a flowchart of the steps of a deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions of the present invention;
[0026] Figure 2 It is a structural block diagram of a deep knowledge tracking model according to a specific embodiment of the present invention;
[0027] Figure 3 It is a schematic diagram of a process of representing a problem with multiple knowledge points in a specific embodiment of the present invention;
[0028] Figure 4 This is an example diagram of the present invention for calculating the forgetting behavior data of knowledge points. DETAILED DESCRIPTION
[0029] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0030] It should be noted that, for the convenience of description, only the parts related to the relevant invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other.
[0031] It should be understood that the "system", "device", "unit" and / or "module" used in this application is a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the word can be replaced by other expressions.
[0032] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "a kind" and / or "the" do not refer to the singular, but also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements. The elements defined by the sentence "includes a..." do not exclude the existence of other identical elements in the process, method, commodity or device that includes the elements.
[0033] In the description of the embodiments of the present application, "plurality" means two or more than two. The following terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features.
[0034] In addition, flow charts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, the various steps may be processed in reverse order or simultaneously. At the same time, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.
[0035] Reference Figure 1 , is a flowchart of an optional example of a deep knowledge tracking method based on forgetting behavior in a multi-knowledge point question proposed by the present invention. The method can be applied to a computer device. The knowledge tracking method proposed in this embodiment may include but is not limited to the following steps:
[0036] Step S1, obtaining answer data and constructing a training set;
[0037] Step S2: constructing a knowledge tracking model, wherein the knowledge tracking model includes an input module, a state update module and a prediction module;
[0038] Step S3: training the knowledge tracking model based on the training set, updating parameters, and obtaining an optimal knowledge tracking model;
[0039] Step S4: predicting answers based on the optimal knowledge model.
[0040] Knowledge tracing is usually expressed as a time series prediction problem, based on the student's past interaction records with exercises X = {x1, x2, ..., x t-1}, predict the probability p(r t =1|q t , X), where x t =(qt ; r t ) represents the question q answered by the student at time t t and the answer result r t ∈(0,1), 0 means the answer is wrong, 1 means the answer is correct. In the model of this paper, it is assumed that each question can be represented by the knowledge point space C = {c1, c2, ..., c n}, that is, q t =C t =(c1,c i , ..., c j ), c i is the i-th knowledge point in the knowledge point space. Then, the student forgetting behavior data is calculated based on the knowledge points contained in the question.
[0041] In some feasible embodiments, the step S1 specifically includes:
[0042] First of all, it should be clear that the input of the deep knowledge tracking system is a student's historical answer sequence. The data obtained is a large number of historical answer sequences of different students. The lengths of these sequences are not the same. In order to facilitate the parallel processing efficiency of the knowledge tracking system, these different sequences are cut into sequences of the same length, which is generally about 200. Based on the sequences of the same length, the training set is constructed.
[0043] In some feasible embodiments, the step S2 specifically includes:
[0044] The knowledge tracking model of the present invention includes an input module, a state updating module and a prediction module.
[0045] The input module calculates the knowledge points and forgotten data contained in the question; the state update module updates the knowledge state; and the prediction module predicts the answer to the next question. Figure 2 .
[0046] In some feasible embodiments, in the input module, the specific working process is as follows:
[0047] In the input stage, we input the questions answered by students. t In this embodiment, q t Perform embedding encoding to represent the answer question as a vector, and obtain a dimension of d k The vector k t ; Then k t With key matrix Multiply them and normalize them through the Softmax function to find the weight w of each knowledge point in the question t , N is the number of knowledge points, the specific formula is as follows:
[0048] w t (i) = Softmax(M k (i)k t )#(1)
[0049] Among them, i represents the i-th knowledge point, w t (i) represents the weight of the i-th knowledge point, k t Represents the question vector.
[0050] After obtaining the weights of each knowledge point in each question, we can represent the question with knowledge points. Assume that there are N knowledge points in the knowledge point space of the questions answered by students in the dataset, but a question does not take all the knowledge points into account, so we have to use the weight w t For screening, the value greater than the threshold τ is set to 1 to indicate that the question examines this knowledge point, and the value less than the threshold τ is set to 0 to indicate that the question does not examine this knowledge point. Figure 3 As shown, C t It is to use knowledge points to express the new result of the question.
[0051] In some feasible embodiments, in the state update module, the specific working process is as follows:
[0052] According to the attached Figure 4 The method shown in the figure calculates the forgetting behavior data of the knowledge points contained in these questions. Figure 4 The time interval f is calculated in i (the time interval between learning two identical knowledge points) and the number of times f c (The number of times the same knowledge point has been learned up to the current moment) are two typical forgetting feature data. Other forgetting behavior data calculated by knowledge points are also acceptable. The present invention only lists these two data. According to educational psychology theory, these two forgetting features will have marginal effects, f c With f i The larger the value of , the greater the impact on the update of the knowledge state. However, when it reaches a certain value, the impact of the forgetfulness feature on the knowledge state will not become more obvious as the forgetfulness feature increases. Therefore, we need to perform log2 x scaling on the extracted forgetfulness feature. The formula for the forgetfulness feature f is as follows:
[0053] f = log2[f c , f i ]#(2)
[0054] In the prediction stage, we need to use the relevant weights of the questions and knowledge points to get the state matrix d v is the dimension size of the state matrix) to read the knowledge state h at the current moment t , is a comprehensive state of knowledge, ht The calculation formula is as follows:
[0055]
[0056] v represents the dimension of the matrix, which can be set to any value, but is generally set to 50.
[0057] In some feasible embodiments, in the prediction module, the specific working process is as follows:
[0058] In the prediction stage, considering the influence of forgetting factors, we first pass the forgotten features through a simple linear layer and activate them using the Sigmoid function to obtain f in The formula is as follows, where
[0059] f in =Sigmoid(W1f+b1)#(4)
[0060] The input question information will also affect the final prediction result. Taking these factors into consideration, the final prediction state f is obtained. t , connect them and activate them using the Tanh function, the formula is as follows, where
[0061] f t =Tanh(W2[h t , f in , k t ]+b2)#(5)
[0062] Finally, we use the prediction layer to predict the probability p that the student will answer the question correctly. t , the calculation formula is as follows, So p t is a scalar.
[0063] p t =Sigmoid(W3f t +b3)#(6)
[0064] Finally, it is necessary to update the students' knowledge status at every moment in time. The update of students' knowledge status is mainly based on the students' answer records. t =(q t ; r t ) to update, and still use the embedding function to encode the answer record, encoding the answer record into a vector v t , dimension d v .
[0065] v tThrough the input gate and the forget gate, the input gate determines which information in the knowledge state matrix to save, and the forget gate determines which information in the knowledge state matrix to forget.
[0066] In some feasible embodiments, the step S3 specifically includes:
[0067] At this point, we can calculate the student's knowledge state at each moment based on the input student's historical answer sequence, and calculate the probability p that the student will answer the question correctly at the next moment t , and then compare it with the real answer records in the data set, calculate the loss value, and backpropagate to update the various parameters of the deep knowledge tracking of this model and retain the optimal parameters.
[0068] A deep knowledge tracking system based on forgetting behavior in multi-knowledge point questions, including:
[0069] Data acquisition unit, which acquires answer data and builds a training set;
[0070] A model building unit, building a knowledge tracking model, wherein the knowledge tracking model includes an input module, a state update module and a prediction module;
[0071] A model training unit, which trains the knowledge tracking model based on the training set, updates parameters, and obtains an optimal knowledge tracking model;
[0072] The model prediction unit predicts the answer based on the optimal knowledge model.
[0073] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0074] A deep knowledge tracking device based on forgetting behavior in multi-knowledge point questions:
[0075] at least one processor;
[0076] at least one memory for storing at least one program;
[0077] When the at least one program is executed by the at least one processor, the at least one processor implements the deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions as described above.
[0078] The contents of the above method embodiments are all applicable to the present device embodiments. The functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0079] A storage medium stores processor-executable instructions, which are used to implement a deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions when executed by the processor.
[0080] The contents of the above method embodiments are all applicable to the present storage medium embodiments. The functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0081] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.
Claims
1. A deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions, characterized by: The following steps are involved: Get the answer data and build a training set; Constructing a knowledge tracking model, wherein the knowledge tracking model includes an input module, a state update module and a prediction module; Training the knowledge tracking model based on the training set, updating parameters, and obtaining an optimal knowledge tracking model; Answer prediction is performed based on the optimal knowledge model.
2. According to claim 1, a deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions is characterized in that: The step of obtaining the answer data and constructing the training set also includes: The answer sequence in the answer data is trimmed to a preset length.
3. According to claim 2, a deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions is characterized in that: The step of training the knowledge tracking model based on the training set, updating parameters, and obtaining the optimal knowledge tracking model specifically includes: Encode the questions in the answer data to obtain a question vector; Use knowledge points to represent questions and get the relevant weights of questions and knowledge points; Based on the questions and knowledge points, forgetting behavior data is calculated to obtain forgetting features; Performing answer prediction according to the knowledge state, the forgetting feature and the question vector to obtain prediction data; Update the knowledge state based on the prediction results; The predicted data is compared with the answer records in the answer data, the loss value is calculated and the parameters are updated to obtain the optimal knowledge tracking model.
4. According to claim 3, a deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions is characterized in that: The formula for expressing the relevant weights of the questions and knowledge points is as follows: w t (i)=Softmax(M k (i)k t ) Among them, i represents the i-th knowledge point, w t (i) represents the weight of the i-th knowledge point, M k represents the preset key matrix, k t Represents the question vector.
5. According to claim 3, a deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions is characterized in that: The formula for the potential knowledge state is as follows: Where N represents the total number of knowledge points. Represents the state matrix.
6. According to claim 3, a deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions is characterized in that: The step of using knowledge points to represent questions and obtaining the relevant weights of questions and knowledge points also includes: The relevant weights are screened through the weight threshold, and the knowledge points less than the threshold are used to indicate that the question does not contain this knowledge point, and the knowledge points greater than the threshold are used to indicate that the question contains this knowledge point, and the question is represented by the relevant knowledge points.
7. According to claim 3, a deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions is characterized in that: The forgetting feature is calculated using multiple knowledge points contained in the question. The calculation formula of the forgetting feature is as follows: f=log2[f c ,f i ] Among them, f c Indicates the number of times the same knowledge point is learned, f i Indicates the time interval between learning two identical knowledge points.
8. A deep knowledge tracking system based on forgetting behavior in multi-knowledge point questions, characterized by: include: Data acquisition unit, which acquires answer data and builds a training set; A model building unit, building a knowledge tracking model, wherein the knowledge tracking model includes an input module, a state update module and a prediction module; A model training unit, which trains the knowledge tracking model based on the training set, updates parameters, and obtains an optimal knowledge tracking model; The model prediction unit predicts the answer based on the optimal knowledge model.
9. A deep knowledge tracking device based on forgetting behavior in multi-knowledge point questions, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a deep knowledge tracking method based on forgetting behavior in multi-knowledge point questions as described in any one of claims 1 to 7.