Multi-factor interactive learning effect evaluation method based on learning emotion embedding

By building a multi-factor interactive network embedded in learning emotions, the evaluation problem of learning emotions in the existing technology is solved, and high accuracy and interpretive learning effect evaluation is achieved, and personalized chemical industry intervention is supported.

CN118941143APending Publication Date: 2024-11-12SHAANXI NORMAL UNIV
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410978585.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing learning effect evaluation model is too simplified by students' individual characteristics and does not fully consider the impact of learning emotions on evaluation, which leads to the inability to effectively reveal individual differences and hinders the implementation of personalized intervention strategies.

Method used

Using a multi-factor interactive network based on learning emotion embedding, through the emotional and cognitive feature embedding module, personalized emotional feature extraction module and learning effect evaluation module, a multi-factor interactive network is built, cross-entropy loss function is trained, and student exercises are realized. Evaluate.

Benefits of technology

It improves the accuracy and interpretability of learning effect evaluation, can deeply analyze the complex interactions of learning emotional characteristics, provide personalized feedback from multiple perspectives, and improves the integrity and reliability of evaluation results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941143A_ABST
    Figure CN118941143A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-factor interactive learning effect evaluation method based on learning emotion embedding. The multi-factor interactive learning effect evaluation method comprises the steps of preprocessing a data set, constructing a multi-factor interactive network, training the multi-factor interactive network, verifying the multi-factor interactive network and testing the multi-factor interactive network. According to the method, a multi-factor interaction network is adopted, personalized features of learners are embedded, meanwhile, a personalized emotion feature extraction module is used, the complex interaction relation between different learning emotion features is deeply analyzed, key emotion features influencing the learning effect are extracted, and learning effect evaluation of learning emotion feature enhancement is achieved. Compared with an existing learning effect evaluation model, the method has good learning effect evaluation performance and high exercise answer prediction accuracy, the evaluation result is high in interpretability, and the method can be used for learning effect evaluation and personalized academic intervention.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of learning effect evaluation and artificial intelligence technology, and specifically relates to student learning effect evaluation and personalized intervention. Background Art

[0002] Learning effectiveness assessment is an activity that aims to evaluate learners’ achievement of educational goals in the educational process through quantitative means. Its core goal is to provide a scientific and objective basis for educational decision-making through systematic collection and analysis of evidence, thereby ensuring and promoting the all-round development of learners.

[0003] At present, the mainstream learning effect evaluation research is influenced by the traditional evaluation concept, focusing on the evaluation indicators with cognitive level as the core, and conducting evaluation and analysis based on the learners' answers to exercises. This type of research focuses on the three core factors of exercises, students and knowledge and their interactions, and has developed a series of learning effect evaluation models, such as Item Response Theory (IRT), Matrix Factorization (MF), and Neural Cognitive Diagnosis (NeuralCD). Although cognitive level is the most intuitive manifestation of learning effect evaluation, the existing models mostly start from the single perspective of cognitive characteristics, oversimplify the individual characteristics of students, and do not fully consider the impact of students' different characteristics on learning effect evaluation. When facing a group of students with completely consistent exercise responses, since such models focus on the interaction results between students and knowledge, they ignore the complex mechanism of the role of students' individual characteristics in the interaction process between students and knowledge. Therefore, such models cannot effectively reveal individual differences between students, hinder teachers from implementing personalized academic intervention strategies, and thus cannot achieve multi-perspective and personalized feedback on learning effect evaluation results, leading to the problem of flattening the evaluation object and homogenizing feedback.

[0004] As a multidimensional descriptive feature, learning emotion covers key psychological factors such as students' learning motivation, learning attitude, and academic anxiety. These factors together constitute a collection of emotional experiences that students experience during the learning process. Existing studies have not only confirmed that emotional characteristics such as learning motivation, academic confidence, and academic anxiety have significant and diverse effects on students' learning outcomes, but also revealed that learning emotion characteristics of different dimensions interact with each other when affecting learning outcomes, and that the effects of emotional characteristics on students are significantly different. Therefore, incorporating students' learning emotions into the learning outcome evaluation framework and using the complexity and diversity of learning emotions to incorporate learners' individual differences into the evaluation model can not only comprehensively explore the causes of learning outcomes, but also provide a reference for teachers to deeply analyze the results of students' interaction with knowledge through the extraction and analysis of key emotional characteristics. Therefore, conducting research on learning outcome evaluation with learning emotion embedded has important prospects and significance. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the deficiencies of the above-mentioned prior art and provide a multi-factor interactive learning effect evaluation method based on learning emotion embedding with high learning effect evaluation accuracy and strong interpretability.

[0006] The technical solution adopted to solve the above technical problems consists of the following steps:

[0007] (1) Dataset preprocessing

[0008] The students' science question answer records and learning emotion intensity data were extracted from the PISA2015 dataset. The student samples with less than 20 exercise answer records and missing learning emotion intensity data were eliminated. The PISA2015 dataset included 44161 student samples, involving 126 exercises and covering 53 knowledge points. The dataset was divided into training set, validation set and test set according to the ratio of 70:10:20.

[0009] (2) Constructing a multi-factor interaction network

[0010] The multi-factor interaction network is composed of an emotion and cognitive feature embedding module, a personalized emotion feature extraction module, a learning effect evaluation module, and an exercise answer prediction module connected in series.

[0011] (3) Training multi-factor interaction networks

[0012] Construct the cross entropy loss function Loss as follows:

[0013]

[0014] in, Represents the predicted value of the i-th data, y iis the true value of the ith data, n represents the total number of data, and n is a finite positive integer.

[0015] The training set is input into the multi-factor interaction network for training. The server graphics card used for training is RTX 2080Ti, the initial learning rate is 0.08, the number of training rounds is 100, the training batch size is 32, and the Adam optimizer is used for training until the cross entropy loss function Loss converges.

[0016] (4) Verify the multi-factor interaction network

[0017] The validation set is input into the trained multi-factor interaction network for verification, the probability of students correctly answering the exercises is predicted, and the area under the receiver operating characteristic curve, prediction accuracy, and root mean square error index values ​​for evaluating the network performance are output.

[0018] (5) Testing multi-factor interaction networks

[0019] The test set is input into the trained multi-factor interaction network for testing, the probability of students answering the exercises correctly is predicted, and the multi-factor interaction learning effect is output.

[0020] In step (2) of the present invention, in constructing a multi-factor interactive network, the emotion and cognitive feature embedding module is composed of a learning emotion feature embedding layer and a cognitive feature encoding layer in parallel; the construction method of the learning emotion feature embedding layer is as follows:

[0021] h f =h w ⊙h u

[0022] h w =g w ×v em

[0023] h u =g u ×v sc

[0024]

[0025] Among them, h f represents the learning sentiment feature embedding layer, ⊙ represents element-wise multiplication, and h w A high-dimensional embedding vector representing the feature index, h u represents the dynamically adjusted eigenvalue vector, g w represents the sentiment feature index, g u represents the sentiment feature value, v em ∈R 1 ×N represents a trainable embedding vector, v sc ∈R1×N Represents a trainable scaling vector, N represents the total number of categories of learned emotions, and N is a finite positive integer.

[0026] The method for constructing the cognitive feature encoding layer is as follows:

[0027] x=[x1,x2,…,x k ]

[0028]

[0029] Among them, x represents the cognitive feature encoding layer, x k Indicates the encoding position, and id indicates the feature index of the encoding object.

[0030] In step (2) of constructing a multi-factor interactive network of the present invention, the personalized emotion feature extraction module is composed of a self-attention layer and a layer normalization layer connected in series.

[0031] In step (2) of the present invention, in constructing a multi-factor interaction network, the learning effect evaluation module is composed of an average pooling layer, a cognitive feature extraction layer, and a learning effect evaluation layer connected in series in sequence. The construction method of the average pooling layer is as follows:

[0032] s f =τ(W×s att +b)

[0033] Among them, s f represents the average pooling layer, τ(·) represents the average pooling function, W is the weight matrix, b is the bias parameter, and s att Represents the personalized sentiment feature vector.

[0034] The method for constructing the cognitive feature extraction layer is as follows:

[0035] s c =sigmoid(x s ×B)

[0036]

[0037] Among them, s c represents the cognitive feature extraction layer, x s Represents the student's cognitive feature encoding vector, B∈R M×T is a trainable parameter matrix, M represents the total number of students, and T represents the total number of knowledge points.

[0038] The construction method of the learning effect evaluation layer is as follows:

[0039] p=α×s f +(1-α)×s c

[0040] Among them, p represents the learning effect evaluation layer, and α∈(0,1) is a trainable parameter.

[0041] In step (2) of constructing a multi-factor interactive network of the present invention, the exercise answer prediction module is composed of a multi-factor interactive layer and a fully connected layer 1, a fully connected layer 2, and a fully connected layer 3 connected in series in sequence.

[0042] The method for constructing the multi-factor interaction layer of the present invention is as follows:

[0043] a=k e ×(pe diff )×e disc

[0044] k e =x e ×R

[0045] e diff =sigmoid(x e ×C)

[0046] e disc =sigmoid(x e ×d)

[0047]

[0048]

[0049] d=[d1,d2,…,d L ] T

[0050] Among them, a represents the multi-factor interaction layer, x e Represents the cognitive feature encoding vector of the exercise, R∈{0,1} L×T Represents the correlation matrix between exercises and knowledge points, C∈R L×T is the trainable parameter matrix, d∈R L×1 is a trainable parameter vector, L represents the total number of exercises, and T represents the total number of knowledge points.

[0051] The output dimension of the fully connected layer 1 of the present invention is 512, the output dimension of the fully connected layer 2 is 256, and the output dimension of the fully connected layer 3 is 1.

[0052] Since the present invention adopts a multi-factor interactive network, embeds learners' emotions and cognitive characteristics, and uses a personalized emotion feature extraction module, it solves the problems of flattening of evaluation objects and homogenization of evaluation feedback in existing learning effect evaluation models, deeply analyzes the complex interactive relationship between different learning emotion characteristics, and realizes learning effect evaluation and exercise answer prediction enhanced by learning emotion characteristics. Compared with the existing learning effect evaluation model, the AUC value and prediction accuracy of the present invention are higher than those of the IRT method, MIRT method, MF method, NeuralCD method, and KaNCD method, and the root mean square error is lower than that of the above methods. It has the advantages of good learning effect evaluation performance, high exercise answer prediction accuracy, strong interpretability, and complete and reliable evaluation results, and can be used for learning effect evaluation and personalized academic intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flow chart of embodiment 1 of the present invention.

[0054] Figure 2 It is a structural diagram of a multi-factor interaction network.

[0055] Figure 3 yes Figure 2 Schematic diagram of the structure of the emotion and cognitive feature embedding module.

[0056] Figure 4 yes Figure 2 Schematic diagram of the structure of the personalized sentiment feature extraction module.

[0057] Figure 5 yes Figure 2 Schematic diagram of the structure of the learning effect evaluation module.

[0058] Figure 6 yes Figure 2 Schematic diagram of the structure of the exercise answer prediction module. Specific implementation methods

[0060] The present invention will be further described in detail below in conjunction with the accompanying drawings and examples, but the present invention is not limited to the following embodiments.

[0061] Example 1

[0062] like Figure 1 As shown, the multi-factor interactive learning effect evaluation method based on learning emotion embedding of this embodiment consists of the following steps.

[0063] (1) Dataset preprocessing

[0064] The students' science question answer records and learning emotion intensity data were extracted from the PISA2015 dataset. The student samples with less than 20 exercise answer records and missing learning emotion intensity data were eliminated. The PISA2015 dataset included 44161 student samples, involving 126 exercises and covering 53 knowledge points. The dataset was divided into training set, validation set and test set according to the ratio of 70:10:20.

[0065] (2) Constructing a multi-factor interaction network

[0066] Figure 2 A schematic diagram of the structure of the multi-factor interactive network of this embodiment is given. Figure 2 In the embodiment, the multi-factor interaction network is composed of an emotion and cognitive feature embedding module, a personalized emotion feature extraction module, a learning effect evaluation module, and an exercise answer prediction module connected in series.

[0067] Figure 3 A schematic diagram of the structure of the emotion and cognitive feature embedding module of this embodiment is given. Figure 3 In the present embodiment, the emotion and cognitive feature embedding module is composed of a learning emotion feature embedding layer and a cognitive feature encoding layer in parallel. The construction method of the learning emotion feature embedding layer of the present embodiment is as follows:

[0068] h f =h w ⊙h u

[0069] h w =g w ×v em

[0070] h u =g u ×v sc

[0071]

[0072] Among them, h f represents the learning sentiment feature embedding layer, ⊙ represents element-wise multiplication, and h w A high-dimensional embedding vector representing the feature index, h u represents the dynamically adjusted eigenvalue vector, g w represents the sentiment feature index, g u represents the sentiment feature value, v em ∈R 1 ×N represents a trainable embedding vector, v sc ∈R 1×N represents a trainable scaling vector, N represents the total number of categories of learned emotions, and the value of N is a finite positive integer. In this embodiment, the value of N is 6.

[0073] The method for constructing the cognitive feature coding layer of this embodiment is as follows:

[0074] x=[x1,x2,…,x k ]

[0075]

[0076] Among them, x represents the cognitive feature encoding layer, x k Indicates the encoding position, and id indicates the feature index of the encoding object.

[0077] Figure 4 Given Figure 2 Schematic diagram of the structure of the personalized emotional feature extraction module. Figure 4 In the embodiment, the personalized emotion feature extraction module is composed of a self-attention layer and a layer normalization layer connected in series.

[0078] Figure 5 Given Figure 2 The structural diagram of the learning effect evaluation module. Figure 5 In the embodiment, the learning effect evaluation module is composed of an average pooling layer, a cognitive feature extraction layer, and a learning effect evaluation layer connected in series. The construction method of the average pooling layer of the embodiment is as follows:

[0079] s f =τ(W×s att +b)

[0080] Among them, s f represents the average pooling layer, τ(·) represents the average pooling function, W is the weight matrix, b is the bias parameter, and s att Represents the personalized sentiment feature vector.

[0081] The construction method of the cognitive feature extraction layer in this embodiment is as follows:

[0082] s c =sigmoid(x s ×B)

[0083]

[0084] Among them, s c represents the cognitive feature extraction layer, x s Represents the cognitive feature encoding vector of students, B∈R M×T is a trainable parameter matrix, M represents the total number of students, and T represents the total number of knowledge points.

[0085] The construction method of the learning effect evaluation layer in this embodiment is as follows:

[0086] p=α×sf +(1-α)×s c

[0087] Wherein, p represents the learning effect evaluation layer, α∈(0,1) is a trainable parameter, and the value of α in this embodiment is 0.5.

[0088] Figure 6 Given Figure 2 The schematic diagram of the structure of the exercise answer prediction module. Figure 6 In the embodiment, the exercise answer prediction module is composed of a multi-factor interaction layer and a fully connected layer 1, a fully connected layer 2, and a fully connected layer 3 connected in series in sequence.

[0089] The method for constructing the multi-factor interaction layer of this embodiment is as follows:

[0090] a=k e ×(pe diff )×e disc

[0091] k e =x e ×R

[0092] e diff =sigmoid(x e ×C)

[0093] e disc =sigmoid(x e ×d)

[0094]

[0095] d=[d1,d2,…,d L ] T

[0096] Among them, a represents the multi-factor interaction layer, x e Represents the cognitive feature encoding vector of the exercise, R∈{0,1} L×T Represents the correlation matrix between exercises and knowledge points, C∈R L×T is the trainable parameter matrix, d∈R L×1 is a trainable parameter vector, L represents the total number of exercises, and T represents the total number of knowledge points.

[0097] In this embodiment, the output dimension of the fully connected layer 1 is 512, the output dimension of the fully connected layer 2 is 256, and the output dimension of the fully connected layer 3 is 1.

[0098] (3) Training multi-factor interaction networks

[0099] Construct the cross entropy loss function Loss as follows:

[0100]

[0101] in, Represents the predicted value of the i-th data, y i is the true value of the ith data, n represents the total number of data, and the value of n is a finite positive integer. The value of n in this embodiment is 3895000.

[0102] The training set is input into the multi-factor interaction network for training. The server graphics card used for training is RTX 2080Ti, the initial learning rate is 0.08, the number of training rounds is 100, the training batch size is 32, and the Adam optimizer is used for training until the cross entropy loss function Loss converges.

[0103] (4) Verify the multi-factor interaction network

[0104] The validation set is input into the trained multi-factor interaction network for verification, the probability of students correctly answering the exercises is predicted, and the area under the receiver operating characteristic curve, prediction accuracy, and root mean square error index values ​​for evaluating the network performance are output.

[0105] (5) Testing multi-factor interaction networks

[0106] The test set is input into the trained multi-factor interaction network for testing, the probability of students answering the exercises correctly is predicted, and the multi-factor interaction learning effect is output.

[0107] Complete a multi-factor interactive learning effect evaluation method based on learning emotion embedding.

[0108] Example 2

[0109] The multi-factor interactive learning effect evaluation method based on learning emotion embedding of this embodiment consists of the following steps.

[0110] (1) Dataset preprocessing

[0111] This step is the same as in Example 1.

[0112] (2) Constructing a multi-factor interaction network

[0113] The structure of the multi-factor interactive network of this embodiment is the same as that of the first embodiment.

[0114] The construction method of the learning emotion feature embedding layer in this embodiment is as follows:

[0115] h f =h w ⊙h u

[0116] h w =g w ×v em

[0117] h u =g u ×v sc

[0118]

[0119] Wherein, N represents the total number of categories of learning emotions, and the value of N is a finite positive integer. The value of N in this embodiment is 39. The meanings and value ranges of other parameters in the above expression are the same as those in Embodiment 1.

[0120] The construction method of the learning effect evaluation layer in this embodiment is as follows:

[0121] p=α×s f +(1-α)×s c

[0122] Wherein, p represents the learning effect evaluation layer, α∈(0,1) is a trainable parameter, and the value of α in this embodiment is 0.1.

[0123] The construction method of other modules and structural layers in the multi-factor interaction network of this embodiment is the same as that of embodiment 1.

[0124] The other steps are the same as those in Example 1. A multi-factor interactive learning effect evaluation method based on learning emotion embedding is completed.

[0125] Example 3

[0126] The multi-factor interactive learning effect evaluation method based on learning emotion embedding of this embodiment consists of the following steps.

[0127] (1) Dataset preprocessing

[0128] This step is the same as in Example 1.

[0129] (2) Constructing a multi-factor interaction network

[0130] The structure of the multi-factor interactive network of this embodiment is the same as that of the first embodiment.

[0131] The construction method of the learning emotion feature embedding layer in this embodiment is as follows:

[0132] h f =h w ⊙h u

[0133] h w =g w ×v em

[0134] h u =g u ×v sc

[0135]

[0136] Wherein, N represents the total number of categories of learning emotions, and the value of N is a finite positive integer. The value of N in this embodiment is 64. The meanings and value ranges of other parameters in the above expression are the same as those in Embodiment 1.

[0137] The construction method of the learning effect evaluation layer in this embodiment is as follows:

[0138] p=α×s f +(1-α)×s c

[0139] Wherein, p represents the learning effect evaluation layer, α∈(0,1) is a trainable parameter, and the value of α in this embodiment is 0.9.

[0140] The construction method of other modules and structural layers in the multi-factor interaction network of this embodiment is the same as that of embodiment 1.

[0141] The other steps are the same as those in Example 1. A multi-factor interactive learning effect evaluation method based on learning emotion embedding is completed.

[0142] In order to verify the beneficial effects of the present invention, the inventors conducted comparative experiments using the multi-factor mutual learning effect evaluation model based on learning emotion embedding of Example 1 of the present invention and the learning effect evaluation model based on item response theory (hereinafter referred to as IRT model), the learning effect evaluation model based on multidimensional item response theory (hereinafter referred to as MIRT model), the learning effect evaluation model based on matrix decomposition (hereinafter referred to as MF model), the learning effect evaluation model based on neural network (hereinafter referred to as NeuralCD model), and the neural network learning effect evaluation model based on knowledge association (hereinafter referred to as KaNCD model).

[0143] Determine the evaluation index: AUC (B1) is obtained by calculating the area under the receiver operating characteristic (ROC) curve, where the horizontal and vertical coordinates of the ROC curve represent the false positive rate (FPR) and the true positive rate (TPR). The larger the AUC value, the stronger the model's ability to distinguish between positive and negative samples under different threshold settings. The ROC curve is constructed as follows:

[0144]

[0145] Among them, TP means that both the predicted value and the true value point to the number of exercises answered correctly by the students, TN means that both the predicted value and the true value point to the number of exercises answered incorrectly by the students, FP means that the predicted value points to the number of exercises answered incorrectly and the true value points to the number of exercises answered correctly, and FN means that the predicted value points to the number of exercises answered correctly and the true value points to the number of exercises answered incorrectly.

[0146] The prediction accuracy (B2) is used to measure the proportion of correct classifications by the model, that is, the proportion of all correctly predicted instances to the total number of instances. The larger the prediction accuracy value, the better the performance of the model. The prediction accuracy is calculated as follows:

[0147]

[0148] Among them, ACC represents the prediction accuracy.

[0149] The root mean square error (B3) is an indicator for evaluating the prediction accuracy of the model, reflecting the average level of the difference between the predicted value and the true value. The smaller the root mean square error value, the better the performance of the model. The root mean square error is calculated as follows:

[0150]

[0151] Among them, RMSE represents the root mean square error, n is the total number of data, and y i For real student exercises answer situation, is the probability that the student correctly answers the exercise predicted by the model.

[0152] The comparative experimental results of the method of Example 1 and the comparative method are shown in Table 1.

[0153] Table 1 Experimental results of the method of Example 1 and the comparative experimental method

[0154]

[0155]

[0156] As can be seen from Table 1, the AUC (B1) of the method in Example 1 is 79.51%, which is higher than the IRT method, MIRT method, MF method, NeuralCD method, and KaNCD method; the prediction accuracy (B2) of the method in Example 1 is 73.21%, which is higher than the IRT method, MIRT method, MF method, NeuralCD method, and KaNCD method. The root mean square error (B3) of the method in Example 1 is 43.28%, which is lower than the IRT method, MIRT method, MF method, NeuralCD method, and KaNCD method. Experiments show that the method of the present invention has better learning effect evaluation accuracy. In addition, the method of the present invention can visualize the multidimensional learning emotion characteristics that affect the learning effect, revealing that there are significant individual differences in the effects of different learning emotion characteristics on the learning effect, improving the interpretability of the learning effect evaluation model, and providing an objective basis for formulating personalized academic intervention strategies based on the emotion dimension.

Claims

1. A multi-factor interactive learning effect evaluation method based on learning emotion embedding, characterized by It consists of the following steps: (1) Dataset preprocessing The students' science question answer records and learning emotion intensity data were extracted from the PISA2015 dataset. The student samples with less than 20 question answer records and missing learning emotion intensity data were eliminated. The PISA2015 dataset includes 44161 student samples, involving 126 questions and covering 53 knowledge points. The dataset is divided into training set, validation set and test set according to the ratio of 70:10:

20. (2) Constructing a multi-factor interaction network The multi-factor interaction network is composed of the emotion and cognitive feature embedding module, the personalized emotion feature extraction module, the learning effect evaluation module, and the exercise answer prediction module in series. (3) Training multi-factor interaction networks Construct the cross entropy loss function Loss as follows: in, Represents the predicted value of the i-th data, y i is the true value of the i-th data, n represents the total number of data, and n is a finite positive integer; The training set is input into the multi-factor interaction network for training. The server graphics card used for training is RTX 2080Ti, the initial learning rate is 0.08, the number of training rounds is 100, the training batch size is 32, and the Adam optimizer is used for training until the cross entropy loss function Loss converges. (4) Verify the multi-factor interaction network The validation set is input into the trained multi-factor interaction network for validation, the probability of students correctly answering the exercises is predicted, and the area under the receiver operating characteristic curve, prediction accuracy, and root mean square error index values ​​for evaluating the network performance are output; (5) Testing multi-factor interaction networks The test set is input into the trained multi-factor interaction network for testing, the probability of students answering the exercises correctly is predicted, and the multi-factor interaction learning effect is output.

2. The multi-factor interactive learning effect evaluation method based on learning emotion embedding according to claim 1 is characterized by: In step (2) of constructing a multi-factor interaction network, the emotion and cognitive feature embedding module is composed of a learning emotion feature embedding layer and a cognitive feature encoding layer in parallel; the construction method of the learning emotion feature embedding layer is as follows: h f =h w ⊙h u h w =g w ×v em h u =g u ×v sc Among them, h f represents the learning sentiment feature embedding layer, ⊙ represents element-wise multiplication, and h w A high-dimensional embedding vector representing the feature index, h u represents the dynamically adjusted eigenvalue vector, g w represents the sentiment feature index, g u represents the sentiment feature value, v em ∈R 1×N represents a trainable embedding vector, v sc ∈R 1×N represents a trainable scaling vector, N represents the total number of categories of learned emotions, and N is a finite positive integer; The method for constructing the cognitive feature encoding layer is as follows: x=[x1,x2,…,x k ] Among them, x represents the cognitive feature encoding layer, x k Indicates the encoding position, and id indicates the feature index of the encoding object.

3. The multi-factor interactive learning effect evaluation method based on learning emotion embedding according to claim 1 is characterized in that: In step (2) of constructing a multi-factor interaction network, the personalized emotion feature extraction module is composed of a self-attention layer and a layer normalization layer connected in series.

4. The multi-factor interactive learning effect evaluation method based on learning emotion embedding according to claim 1 is characterized in that: In step (2) of constructing a multi-factor interaction network, the learning effect evaluation module is composed of an average pooling layer, a cognitive feature extraction layer, and a learning effect evaluation layer connected in series in sequence. The construction method of the average pooling layer is as follows: s f =τ(W×s att +b) Among them, s f represents the average pooling layer, τ(·) represents the average pooling function, W is the weight matrix, b is the bias parameter, and s att Represents personalized sentiment feature vector; The method for constructing the cognitive feature extraction layer is as follows: s c =sigmoid(x s ×B) Among them, s c represents the cognitive feature extraction layer, x s Represents the cognitive feature encoding vector of students, B∈R M×T is a trainable parameter matrix, M represents the total number of students, and T represents the total number of knowledge points; The construction method of the learning effect evaluation layer is as follows: p=α×s f +(1-a)×s c Among them, p represents the learning effect evaluation layer, and α∈(0,1) is a trainable parameter.

5. The multi-factor interactive learning effect evaluation method based on learning emotion embedding according to claim 1 is characterized in that: In step (2) of constructing a multi-factor interaction network, the exercise answer prediction module is composed of a multi-factor interaction layer and a fully connected layer 1, a fully connected layer 2, and a fully connected layer 3 connected in series in sequence.

6. The multi-factor interactive learning effect evaluation method based on learning emotion embedding according to claim 5 is characterized by: The method for constructing the multi-factor interaction layer is as follows: a=k e ×(p-e diff )×e disc k e =x e ×R e diff =sigmoid(x e ×C) e disc =sigmoid(x e ×d) d=[d1,d2,…,d L ] T Among them, a represents the multi-factor interaction layer, x e Represents the cognitive feature encoding vector of the exercise, R∈{0,1} L×T Represents the correlation matrix between exercises and knowledge points, C∈R L×T is the trainable parameter matrix, d∈R L×1 is a trainable parameter vector, L represents the total number of exercises, and T represents the total number of knowledge points.

7. The multi-factor interactive learning effect evaluation method based on learning emotion embedding according to claim 5 is characterized by: The output dimension of the fully connected layer 1 is 512, the output dimension of the fully connected layer 2 is 256, and the output dimension of the fully connected layer 3 is 1.

Citation Information

Cited By

  • Knowledge tracking method and system based on hypergraph neural network and emotion perception

    CN119311809A

  • Method for evaluating thinking ability in bilingual writing based on deep learning

    CN119721160A