A student evaluation method based on personalized information

Abstract

The present invention discloses a student evaluation method based on personalized information, including the following: S1: Construct a data set; S2: Construct an evaluation model, import the student data in S1 into a feature encoder, and respectively obtain personal information feature representations, exercise information feature representations, and personalized information feature representations; S3: Import the basic information feature representations in S2 into a cognitive diagnosis module to obtain cognitive diagnosis feature representations; S4: Import the personalized information feature representations in S2 into a personalized information enhancement module to obtain personalized feature representations; S5: Import the cognitive diagnosis feature representations in S3 and the personalized feature representations in S4 into a prediction module to obtain predicted answer results, construct a loss function to optimize the evaluation model, and minimize the loss function to update the parameters of the evaluation model. The present invention significantly improves the performance of the evaluation model for cognitive diagnosis in terms of accuracy by integrating and utilizing the personalized information of students.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and more particularly, to a student evaluation method based on personalized information. Background Art

[0002] Cognitive diagnosis is one of the core tasks in the field of intelligent education. Its main goal is to diagnose the mastery level of each knowledge point of students by revealing the interaction pattern between students and each knowledge point during the process of answering questions. The current diagnostic models mainly focus on the interaction data between students and questions, that is, to evaluate the mastery of each knowledge point of students according to the answering records of students to questions.

[0003] With the rapid development of deep learning, especially the outstanding achievements in the fields of computer vision and natural language processing, the computer's ability to understand and analyze non-interactive information has been greatly enhanced, which provides the possibility for improving the accuracy of cognitive diagnosis models.

[0004] The existing cognitive diagnosis models are mainly divided into probability models and neural network models. These models mainly study the relationship between users and questions, and quantify the interaction between users and questions, the association between questions and knowledge points, and the network structure of knowledge points. However, the information utilization methods of these models are relatively single, and they cannot fully explore the high-dimensional potential features of students during the interaction process on the learning platform.

[0005] As a popular method in the current field of machine learning, the neural network can theoretically fit any continuous function by using a multi-layer perceptron for non-linear transformation of input features, so it is very suitable for modeling the complex relationship between students and questions. However, the "black box" nature of the neural network makes its parameters lack interpretability in most cases. Although the existing neural network models can predict the answering results of students, the student vectors they generate cannot truly reflect the mastery level of students on each specific knowledge point, which causes difficulties in implementing accurate cognitive diagnosis, and inaccurate diagnostic results will waste students' energy on repeated exercises.

[0006] Although the existing cognitive diagnosis models have made remarkable progress in the interaction between students and exercises, there are still some defects and deficiencies. In practical applications, the learning data of students is often sparse. A student may only complete a part of the learning tasks or tests, resulting in insufficient model training data. Data sparsity will cause many cognitive diagnosis models to be unable to accurately capture the learning characteristics of students for modeling, affect the accuracy of diagnostic results, and it is difficult to explain the reasons for the wrong results predicted by the cognitive diagnosis models. Many cognitive diagnosis models are still insufficient in using students' personalized information and lack effective methods to synthesize and utilize this information, which will cause the models to be unable to fully consider the individual differences of students, and thus it is difficult to provide truly personalized learning suggestions.

[0007] Existing traditional non-neural network models are difficult to meet the modeling requirements of students' interaction behaviors in complex scenarios due to their functional limitations. Neural networks, on the other hand, have the potential to explore rich information between exercises and students and perform high-dimensional analysis on this information. Reasonably using personalized information as a signal to guide the model in student modeling can uncover the influence of the commonalities among individual students on their knowledge mastery levels. The present invention can better utilize students' personalized information to assist in formulating customized strategies for students' learning, and enhancing students' cognitive diagnosis using personalized information will lead to more accurate and interpretable results. Summary of the Invention

[0008] To overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a student evaluation method based on personalized information to solve the problems raised in the above background art.

[0009] A student evaluation method based on personalized information includes the following steps:

[0010] Step S1: Construct a data set, which includes preprocessed student data. The student data includes basic information composed of personal information and exercise information and students' personalized information;

[0011] Step S2: Construct an evaluation model. The evaluation model includes a feature encoder, a cognitive diagnosis module, a personalized information enhancement module, and a prediction module. Import the student data in Step S1 into the feature encoder to respectively obtain the feature representations of personal information, exercise information, and personalized information; among them, the feature representations of personal information and exercise information constitute the feature representation of basic information;

[0012] Step S3: Import the feature representation of basic information in Step S2 into the cognitive diagnosis module to obtain the cognitive diagnosis feature representation;

[0013] Step S4: Import the feature representation of personalized information in Step S2 into the personalized information enhancement module to obtain the personalized feature representation;

[0014] Step S5: Import the cognitive diagnosis feature representation in Step S3 and the personalized feature representation in Step S4 into the prediction module to obtain the predicted answer results, construct a loss function to optimize the evaluation model, and minimize the loss function to update the parameters of the evaluation model.

[0015] Further, the preprocessing of the data in Step S1 is specifically as follows:

[0016] Perform encoding and vectorization representation on the student data, including:

[0017] Step S11: Perform encoding and vectorization representation on personal information and exercise information;

[0018] Step S12: Preprocess the personalized information of students and encode it into the format of a weight matrix;

[0019] Step S13: Repeat steps S11 - S12 to obtain N pieces of preprocessed student data, and divide the data set according to the categories of the training set, validation set, and test set to obtain the preprocessed data set.

[0020] Further, step S2 is specifically as follows:

[0021] Step S21: The feature encoder obtains the knowledge - point correlation vector and exercise discrimination vector of the personal information feature representation and exercise information feature representation of student data respectively by constructing a learning matrix and through an activation function; expressed as:

[0022] ;

[0023] ;

[0024] ;

[0025] Among them, represents the personal information feature representation, represents the knowledge - point correlation vector, represents the exercise discrimination vector, are all learnable matrices, A represents the knowledge - point mastery matrix, B represents the exercise knowledge - point examination difficulty matrix, D represents the exercise discrimination matrix, represents personal information, represents the exercise information feature representation;

[0026] Step S22: Combine matrix theory to perform function fitting. During this process, obtain the exercise knowledge - point correlation matrix; expressed as:

[0027] ;

[0028] ;

[0029] Among them, represents the exercise knowledge - point correlation matrix obtained through matrix theory, is the knowledge - point matrix corresponding to each exercise information feature representation marked by the q - matrix; represents element - by - element multiplication, represents the basic information feature representation;

[0030] Step S23: The feature encoder obtains the student's personalized information in the format of the preprocessed weight matrix, that is, the personalized information feature representation.

[0031] Further, step S3 is specifically as follows:

[0032] The monotonicity assumption is realized through a fully connected layer; assuming the number of fully connected layers is 3, it is expressed as:

[0033] ;

[0034] ;

[0035] Among them, represents the output of the first layer of the fully connected layer, and respectively represent the weight matrix and bias of the first layer of the fully connected layer, represents the output of the second layer of the fully connected layer, and respectively represent the weight matrix and bias of the second layer of the fully connected layer; represents the sigmoid activation function, making the features more linearly separable or the classification more accurate;

[0036] The third layer of the fully connected layer is used to fit the answer prediction, which is expressed as:

[0037] ;

[0038] Among them, represents the probability prediction value of the student answering the exercise correctly, and respectively represent the weight matrix and bias of the third layer of the fully connected layer;

[0039] Based on this, the cognitive diagnosis feature representation is obtained; it is expressed as:

[0040] ;

[0041] Among them, represents the cognitive diagnosis feature representation.

[0042] Further, step S4 is specifically as follows:

[0043] Step S41: The personalized information enhancement module learns the potential influence between different personalized information of the student through the self-attention mechanism, highlights important information and filters out unimportant information, shares weights through the self-attention mechanism, and obtains the potential influence between information; uses the cosine similarity function to calculate the similarity between the personal information feature representation and the personalized information feature representation, and obtains the weight vector corresponding to the th personalized information of the student , expressed as:

[0044] ;

[0045] ;

[0046] Among them, represents the personalized information feature representation, represents the cosine function similarity calculation function, represents the discrete normalization function, represents the weight vector of the row weight, represents the th information weight of the th student, represents the influence intensity of the personalized information on the student, represents the total number of personalized information,

[0047] Step S42: Use the attention mechanism to aggregate the contribution weights of students with other similar personalized information, expressed as:

[0048] ;

[0049] ;

[0050] Among them, represents the influence weight of the personalized information on the th student, represents the personalized information weight matrix;

[0051] Step S43: While fusing additional personalized information, in order to adaptively perceive the influence relationship of the personalized information on the cognitive diagnosis result, adaptively update the weight matrix and the hyperparameter threshold of the influence of the information on the student individual modeling through the loss function , thereby enhancing the final diagnosis result, as shown in the following formula:

[0052] ;

[0053] Among them, represents the degree of mastery of the knowledge points predicted by the personalized information enhancement module for the student, that is, the personalized feature representation.

[0054] Furthermore, Step S5 is specifically as follows:

[0055] The prediction module sets hyperparameters to enhance the final diagnosis result, as shown in the following formula:

[0056] ;

[0057] ;

[0058] ;

[0059] Among them, represents an evaluation model, represents the preliminary predicted answer result, represents hyperparameters, represents the relevant parameters of the exercise, including the difficulty of the question , the discrimination of the question , the suspicion factor and the sliding factor , represents that when the parameters of the exercise are the final predicted answer result.

[0060] Furthermore, in step S5, a loss function is constructed to optimize the evaluation model, and the loss function is minimized to update the parameters of the evaluation model. Specifically:

[0061]

[0062] Among them, represents the hyperparameter for balancing the losses of the two modules, represents the total loss function, represents the loss function of the cognitive diagnosis module, represents the loss function of the personalized information enhancement module.

[0063] Technical effects and advantages of the present invention:

[0064] By integrating and utilizing the personalized information of students, the present invention significantly improves the performance of the evaluation model for cognitive diagnosis in terms of accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is a flowchart of the steps of a method for evaluating students based on personalized information according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0067] Refer to Figure 1 , a method for evaluating students based on personalized information. In one example, it includes the following steps:

[0068] Step S1: Construct a dataset, which includes preprocessed student data. The student data includes basic information composed of personal information and exercise information, and student personalized information;

[0069] Step S2: Construct an evaluation model. The evaluation model includes a feature encoder, a cognitive diagnosis module, a personalized information enhancement module, and a prediction module. Import the student data in Step S1 into the feature encoder to obtain the personal information feature representation, the exercise information feature representation, and the personalized information feature representation respectively; among them, the personal information feature representation and the exercise information feature representation constitute the basic information feature representation;

[0070] Step S3: Import the basic information feature representation in Step S2 into the cognitive diagnosis module to obtain the cognitive diagnosis feature representation;

[0071] Step S4: Import the personalized information feature representation in Step S2 into the personalized information enhancement module to obtain the personalized feature representation;

[0072] Step S5: Import the cognitive diagnosis feature representation in Step S3 and the personalized feature representation in Step S4 into the prediction module to obtain the predicted answer results, construct a loss function to optimize the evaluation model, and minimize the loss function to update the parameters of the evaluation model.

[0073] Further, in one example, preprocess the data in Step S1, specifically:

[0074] Perform encoded vector representation on the student data, including:

[0075] Step S11: Perform encoded vector representation on the personal information and exercise information;

[0076] Step S12: Preprocess the student personalized information and encode it into a weight matrix format;

[0077] Step S13: Repeat Steps S11 - S12 to obtain N pieces of preprocessed student data, and divide the dataset according to the categories of the training set, validation set, and test set to obtain the preprocessed dataset.

[0078] Further, in one example, Step S2 is specifically:

[0079] Step S21: The feature encoder obtains the knowledge point correlation vector and the exercise discrimination vector of the personal information feature representation and the exercise information feature representation of the student data respectively by constructing a learning matrix and through an activation function; expressed as:

[0080] ;

[0081] ;

[0082] ;

[0083] Among them, represents the personal information feature representation, represents the knowledge point relevance vector, represents the exercise discrimination vector, are all learnable matrices, A represents the knowledge point mastery matrix, B represents the exercise knowledge point examination difficulty matrix, and D represents the exercise discrimination matrix, represents personal information, represents the exercise information feature representation;

[0084] Step S22: Combine matrix theory to perform function fitting. During this process, obtain the exercise knowledge point correlation matrix; denoted as:

[0085] ;

[0086] ;

[0087] Among them, represents the exercise knowledge point correlation matrix obtained through matrix theory, is the knowledge point matrix corresponding to each exercise information feature representation marked by the q matrix; represents element-wise multiplication, represents the basic information feature representation;

[0088] Step S23: The feature encoder obtains the student's personalized information in the format of a preprocessed weight matrix, that is, the personalized information feature representation.

[0089] Furthermore, in one example, step S3 is specifically:

[0090] Implement the monotonicity assumption through the fully connected layer; assume that the number of fully connected layers is 3 layers, denoted as:

[0091] ;

[0092] ;

[0093] Among them, represents the output of the first layer of the fully connected layer, and represent the weight matrix and bias of the first layer of the fully connected layer respectively, represents the output of the second layer of the fully connected layer, and respectively represent the weight matrix and bias of the second layer of the fully connected layer; represents the sigmoid activation function, making the features more linearly separable or the classification more accurate;

[0094] The third layer of the fully connected layer is used to fit the answer prediction, expressed as:

[0095] ;

[0096] where represents the probability prediction value of the student answering the exercise correctly, and respectively represent the weight matrix and bias of the third layer of the fully connected layer;

[0097] Based on this, the cognitive diagnosis feature representation is obtained; expressed as:

[0098] ;

[0099] where represents the cognitive diagnosis feature representation.

[0100] Furthermore, in one example, step S4 is specifically:

[0101] Step S41: The personalized information enhancement module learns the potential influence between different personalized information received by the student through the self-attention mechanism, highlights important information and filters out unimportant information, shares weights through the self-attention mechanism, and obtains the potential influence between information; calculates the similarity between the personal information feature representation and the personalized information feature representation using the cosine similarity function, and obtains the weight vector corresponding to the th personalized information of the student , expressed as:

[0102] ;

[0103] ;

[0104] where represents the personalized information feature representation, represents the cosine function similarity calculation function, represents the discrete normalization function, represents the weight vector the th row weight of the th student's th information weight, represents the influence intensity of the personalized information on the student, represents the total number of personalized information, represents the shared weight;

[0105] Step S42: Aggregate the contribution weights of students with other similar personalized information using the attention mechanism, expressed as:

[0106] ;

[0107] ;

[0108] wherein, represents the influence weight of personalized information on the th student, represents the personalized information weight matrix;

[0109] Step S43: While fusing additional personalized information, in order to adaptively perceive the influence relationship of personalized information on the cognitive diagnosis result, adaptively update the weight matrix and the hyperparameter threshold of the influence of information on the student individual modeling through the loss function , thereby enhancing the final diagnosis result, as shown in the following formula:

[0110] ;

[0111] wherein, represents the degree of mastery of knowledge points predicted by the personalized information enhancement module for students, that is, the personalized feature representation.

[0112] Furthermore, Step S5 is specifically as follows:

[0113] The prediction module sets hyperparameters to enhance the final diagnosis result, as shown in the following formula:

[0114] ;

[0115] ;

[0116] ;

[0117] wherein, represents the evaluation model, represents the preliminary predicted answer result, represents the hyperparameter, represents the relevant parameters of the exercise, including the difficulty of the question , the discrimination of the question , the suspicion factor and the sliding factor , represents that when the parameter of the exercise is the final predicted answer result.

[0118] Further, in one example, in step S5, a loss function is constructed to optimize the evaluation model, and the loss function is minimized to update the parameters of the evaluation model, specifically:

[0119]

[0120] where represents a hyperparameter for balancing the losses of the two modules, represents the total loss function, represents the loss function of the cognitive diagnosis module, represents the loss function of the personalized information enhancement module.

[0121] The present invention significantly improves the performance of the evaluation model for cognitive diagnosis in terms of accuracy by integrating and utilizing the personalized information of students.

[0122] The above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A student evaluation method based on personalized information, characterized in that, It includes the following steps: Step S1: Construct a dataset, which includes preprocessed student data. The student data includes basic information composed of personal information and exercise information, and student personalized information. Step S2: Construct an evaluation model. The evaluation model includes a feature encoder, a cognitive diagnosis module, a personalized information enhancement module, and a prediction module. Import the student data in Step S1 into the feature encoder to obtain the personal information feature representation, the exercise information feature representation, and the personalized information feature representation respectively. Among them, the personal information feature representation and the exercise information feature representation constitute the basic information feature representation. Step S3: Import the basic information feature representation in Step S2 into the cognitive diagnosis module to obtain the cognitive diagnosis feature representation. Step S4: Import the personalized information feature representation in Step S2 into the personalized information enhancement module to obtain the personalized feature representation. Step S5: Import the cognitive diagnosis feature representation in Step S3 and the personalized feature representation in Step S4 into the prediction module to obtain the predicted answer results. Construct a loss function to optimize the evaluation model, and minimize the loss function to update the parameters of the evaluation model. Step S4 is specifically as follows: Step S41: The personalized information enhancement module learns the potential influence between different personalized information of students through the self-attention mechanism, highlights important information and filters out unimportant information, shares weights through the self-attention mechanism, and obtains the potential influence between information; uses the cosine similarity function to calculate the similarity between the personal information feature representation and the personalized information feature representation, and obtains the weight vector corresponding to the th personalized information of the student , which is expressed as: ； ； in, Indicates the characteristics of personal information. Indicates personalized information feature representation, Indicates the use of cosine function similarity calculation function, represents the discrete normalization function, Represents the weight vector No. row weights, Indicates student's Information weight, Indicates the impact of personalized information on students, Indicates the total amount of personalized information, represents the shared weight; Step S42: Use the attention mechanism to aggregate the contribution weights of other students with similar personalized information, expressed as: ； ； ； Among them, represents the influence weight of personalized information on the th student, represents the personalized information weight matrix; Step S43: While integrating additional personalized information, in order to adaptively perceive the influence relationship of personalized information on the cognitive diagnosis result, the weight matrix is adaptively updated through the loss function, so as to enhance the final diagnosis result, as shown in the following formula: ； Among them, represents the personalized information enhancement module predicting the student's mastery of knowledge points, that is, the personalized feature representation; Step S5 is specifically as follows: The prediction module sets hyperparameters , thus enhancing the final diagnosis result, as shown in the following formula: ； ； ； Among them, represents the evaluation model, represents the preliminary predicted answer result, represents the cognitive diagnosis feature representation, represents the hyperparameter, represents the relevant parameters of the exercise, including the difficulty of the question , the discrimination of the question , suspicion factor and sliding factor , represents that when the parameters of the exercise are the final predicted answer result.

2. The method for evaluating students based on personalized information according to claim 1, wherein Preprocess the data in Step S1, specifically: Encode and vectorize the student data, including: Step S11: Encode and vectorize the personal information and exercise information. Step S12: Preprocess the student personalized information and encode it into the format of a weight matrix. Step S13: Repeat Steps S11 - S12 to obtain N pieces of preprocessed student data, and divide the dataset according to the categories of the training set, validation set, and test set to obtain the preprocessed dataset.

3. The student evaluation method based on personalized information according to claim 2, wherein Step S2 is specifically as follows: Step S21: The feature encoder obtains the knowledge point correlation vector and the exercise discrimination vector of the personal information feature representation and the exercise information feature representation of the student data respectively by constructing a learning matrix and through an activation function, expressed as: ； ； ； Among them, represents the personal information feature representation, represents the knowledge point relevance vector, represents the exercise discrimination vector, are all learnable matrices. A represents the knowledge point mastery matrix, B represents the exercise knowledge point examination difficulty matrix, and D represents the exercise discrimination matrix. represents the personal information, represents the exercise information feature representation; Step S22: Combine with matrix theory to perform function fitting. During this process, obtain the exercise knowledge point correlation matrix, which is expressed as: ； ； Among them, represents the exercise knowledge point correlation matrix obtained through matrix theory, which is the knowledge point matrix corresponding to the information feature representation of each exercise marked by the q matrix; represents element-wise multiplication, represents the basic information feature representation; Step S23: The feature encoder obtains the preprocessed student personalized information in the format of a weight matrix, that is, the personalized information feature representation.

4. The student evaluation method based on personalized information according to claim 3, wherein Step S3 is specifically as follows: Implement the monotonicity assumption through a fully connected layer. Assume that the number of fully connected layers is 3 layers, expressed as: ； ； Among them, represents the output of the first layer of the fully connected layer, and respectively represent the weight matrix and bias of the first layer of the fully connected layer, represents the output of the second layer of the fully connected layer, and respectively represent the weight matrix and bias of the second layer of the fully connected layer; represents the sigmoid activation function, making the features more linearly separable or the classification more accurate; The third layer of the fully connected layer is used to fit the answer prediction, expressed as: ； wherein, represents the predicted probability value of the student answering the exercise correctly, and respectively represent the weight matrix and bias of the third layer of the fully connected layer; Based on this, the cognitive diagnosis feature representation is obtained, expressed as: ； Among them, represents a cognitive diagnostic feature representation.

5. A student evaluation method based on personalized information according to claim 1, characterized in that, In Step S5, a loss function is constructed to optimize the evaluation model, and the loss function is minimized to update the parameters of the evaluation model, specifically: ； Among them, represents the hyperparameter for balancing the losses of two modules, represents the total loss function, represents the loss function of the cognitive diagnosis module, represents the loss function of the personalized information enhancement module.

Images