Recommendations for personalized online test questions to improve students' emotional recognition

Abstract

The present invention discloses a method for recommending personalized online test questions for student emotion recognition, comprising: step S1 of collecting student response behavior and corresponding emotion records from online learning, performing data cleaning, and extracting useful information; step S2 of classifying the student's complex emotions using PCA based on the student's emotion data, adjusting the influence of emotions on the student through the weighting of the self-attention module, and modeling the student's external perception; step S3 of comprehensively predicting the student's personalized response by introducing the NCB-IRM framework, combining the student's external perception and the student's unique perception; step S4 of jointly predicting the student's response to the test questions by decomposing commonalities in the student's responses using PMF and combining them with the student's personalized predicted response; and step S5 of providing the student with test questions of appropriate difficulty based on the predicted student response and forming a final recommendation list. Based on PMF and NCB-IRM, the present invention combines student individuality and commonalities in responses to achieve accuracy and rationality in recommended test questions based on the student's test question mastery.

Description

[Technical Field]

[0001] The present invention relates to the technology of online learning big data mining, and more particularly to a method for recommending personalized online test questions for student sentiment recognition. [Background technology]

[0002] The development of online learning platforms provides students with a convenient way to practice test questions. Through these platforms, students can access a rich resource bank of questions and learn knowledge anytime, anywhere. Online learning platforms offer teachers and students a broad, convenient, and intelligent learning approach. In recent years, numerous online learning platforms have emerged, such as ASSISTment, Coursera, and EdX, providing students with test question practice. However, the sheer volume of test question resources often makes it difficult for students to effectively find suitable test questions, impacting their learning interest and efficiency. Therefore, it is crucial to utilize educational big data mining technology to accurately recommend appropriate test question resources to students based on their individual characteristics.

[0003] Recommendation systems in the field of education can provide students with various forms of learning and test resource resources, such as video courses and practice test questions. Recommendation systems are a crucial area of ​​information retrieval in educational big data mining, and their goal is to provide learning materials that meet students' needs and to support students' practice with test questions. Traditional online resource recommendation methods primarily rely on collaborative filtering algorithms to construct user-resource matrices and then recommend learning resources based on the similarity between users or resources. However, in online learning scenarios, these methods have several shortcomings: they do not adequately consider students' personal characteristics, and the recommendations lack diversity and rationality. As a result, the test question resources provided by recommendation systems cannot meet students' personal needs, and they cannot provide training that is tailored to the purpose.

[0004] To mine students' practice patterns for test questions, most conventional online test question resource recommendation methods employ artificial intelligence algorithms such as neural networks, deep learning, and reinforcement learning. However, these methods often overemphasize the accuracy of the recommendation model and ignore the personal characteristics exhibited by students during practice. They are unable to design personalized recommendation methods based on students' own learning characteristics, resulting in recommended test questions lacking directionality and interpretability. Cognitive diagnostics provide a theoretical basis for providing personalized recommendation methods to various students by analyzing students' answer logs, assessing their knowledge acquisition status, and determining their learning state. Using cognitive diagnostic methods (CDM) for test question recommendations allows for diagnosing students' knowledge acquisition status, inferring characteristics such as the difficulty and distinguishability of test questions, and considering students' personal characteristics to a considerable extent. Since the diagnostic effect of CDM directly impacts the accuracy of test question recommendations, the performance of the recommendation method deteriorates if the CDM diagnostic results are not sufficiently accurate.

[0005] In cognitive assessments, increasing the data input dimension can effectively improve the effectiveness of the assessment. Furthermore, by utilizing multi-source information to enhance data semantic integration in the recommendation system, the representation of users and objects can be enriched. Therefore, after increasing the emotional dimension, recommendation of emotion recognition test questions can improve diagnostic accuracy while utilizing more personalized knowledge from students. However, CDM does not consider the commonality of student responses, resulting in limited coverage of recommended test questions. Consequently, data becomes sparse in the subsequent recommendation process, and there are certain limitations to the recommended test questions. [Overview of the project] [Problems that the invention aims to solve]

[0006] To address the above problems, the object of the present invention is to provide a personalized online test question recommendation method for recognizing students' emotions, to introduce students' emotional information, to design an online test question recommendation method based on the individuality and commonalities of students' test question responses, to recommend test questions suitable for students, and to improve students' practice effectiveness. [Means for solving the problem]

[0007] To solve at least one of the above technical problems, according to one aspect of the present invention, Step S1 involves collecting records of students' online learning response behavior and corresponding emotions, performing data cleaning, and extracting useful information. Step S2 involves classifying students' complex emotions using PCA based on student emotional data, adjusting the influence of emotions on students through the weighting of the self-attention module, and modeling students' external perceptions. Step S3 involves combining students' external perceptions and their intrinsic perceptions to introduce the NCB-IRM framework and comprehensively predict students' individualized responses. Step S4 of jointly predicting a student's response to a test question by decomposing the commonality of the student's answers using PMF and combining it with the student's personalized predicted response; Step S5 of providing test questions of appropriate difficulty to the student based on the predicted student response and forming a final recommendation list, to provide a method for recommending personalized online test questions for student emotion recognition.

[0008] In step S1 of collecting the online learning answer behavior records including the student's emotions and extracting valid information, specifically: S11, collect the student id, test question id, key point id, the student's emotion data E for the test question, and the student's response R to the test question. E includes six emotional dimensions of bored, concentrating, confused, frustrated, offtask, gaming, and the emotional element A of each dimension ij belongs to [0,1], and the closer to 1, the higher the student's performance for this emotion. The response of student i to test question j is r ij is represented by r ij =1 means that student i answered test question j correctly, and r ij =0 means that student i answered test question j incorrectly. S12, perform data cleaning, delete the answer behavior records including empty entries and redundancy, and extract valid information as follows: Extract the potential knowledge acquisition degree of student i for key point k from the student's behavior, represented by α ik and α ik ∈[0,1], The test question matrix Q includes the relationship between all test questions and key points, and the element q jk represents the situation where each test question q tests a key point, and q jk =1 means that test question j does not test key point k, and q jk =0 means that test question j does not test key point k, characterized by the method according to claim 1.

[0009] In step S2, which models students' external perceptions through the PCA algorithm module and the hierarchical self-attention module based on student emotions, student response behavior, and the test question matrix, specifically, S21, using the PCA algorithm, students' complex emotions were clustered into two types, represented as positive and negative emotions, and the expression formula was:

number

number

number

number

number

number

number

number

number

number

[0010] Step S3, which combines external perception and the student's inherent perception to introduce the NCB-IRM framework and comprehensively predict the student's individualized response, involves the student's response r ij It is known that two extreme situations in which students' emotional characteristics influence their responses are considered, and specifically, In S31, student i answered test question j correctly, influenced only by emotional characteristics, and the student's learning response function was modeled.

number

number

number

number

number

[0011] In step S4, which uses PMF to break down commonalities in students' responses based on their actual responses and combines this with students' individualized predicted responses to collaboratively predict students' responses to test questions, specifically, S41, Probability matrix decomposition is performed on the actual responses of students through PMF, and the feature matrix U of the students and test questions is obtained. i , V j Having obtained, S42, predicting responses that include commonalities among students.

number

number

number

[0012] In step S5, which provides students with test questions of appropriate difficulty based on their predicted responses and forms a final recommendation list, specifically, S51, Predicted student response R p Based on this, the difficulty levels [d1,d2] of the test questions were designed. The difficulty level of the test questions is calculated from the average student response.

number

[0013] This invention mines commonalities in students' responses through PMF and combines them with students' personalized predictive responses to improve the predictive power of the model while ensuring comprehensiveness of the recommendation method.

[0014] This invention improves the rationality of recommended test questions by designing the difficulty level of test questions according to the students' level of proficiency, while taking into full consideration the individual characteristics of students and the commonalities in their answers.

[0015] According to another aspect of the present invention, a computer-readable storage medium is provided in which a computer program is stored, and when the program is executed on a processor, steps of the personalized online test question recommendation method for student emotion recognition of the present invention are realized.

[0016] According to yet another aspect of the present invention, a computer device is provided which includes memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the program, steps of the present invention's method for recommending personalized online test questions for student emotion recognition are realized. [Effects of the Invention]

[0017] The present invention has at least the following beneficial effects compared to the prior art.

[0018] 1. This invention proposes a personalized online test question recommendation method for student emotion recognition by introducing NCB-IRM based on PMF. First, it uses NCB-IRM to combine the external perception and internal perception of students' emotions to more comprehensively consider students' personal needs. Second, it mines commonalities in students' responses through PMF to ensure comprehensiveness of recommendation methods and improve the rationality of the recommended test questions.

[0019] 2. NCB-IRM designs a modeling method for external cognition based on students' emotion perception, reducing the complexity of emotion analysis using PCA, while adjusting the impact on students according to the weight of different emotions using a hierarchical self-attention method, providing an effective method for mining students' personal needs based on their emotions. [Brief explanation of the drawing]

[0020] [Figure 1] This is a flowchart of the method according to the present invention. [Figure 2] This is a schematic diagram illustrating the structure of modeling the external perception of students' emotional recognition. [Figure 3] This invention represents the performance of students' answer accuracy rates corresponding to different difficulty levels of test questions. [Figure 4] Figure 4(a) shows the cognitive results of two students and examples of recommended test questions. Figure 4(b) shows the reinforcement cognitions of diagnosed students A and B, and the recommended test questions for students A and B, as well as the relationship between the test questions and key points. [Modes for carrying out the invention]

[0021] The dataset in this embodiment comes from the online learning platform ASSISTment, from which data from both 2009 and 2017 were collected. Of these, ASSIST2009 contains 401,756 student test practice records, while ASSIST2017 includes additional student sentiment data and contains 942,816 student test practice records. When each student answers a test question, their response and sentiment data are all recorded.

[0022] As shown in Figure 1, the recommended method for personalized online test questions for student emotion recognition described in this embodiment includes the following steps S1 to S5.

[0023] S1. Collect learning behavior records of students watching educational videos, perform data cleaning on the learning behavior records, and extract useful information.

[0024] Collect student ID, test question ID, key point ID, student sentiment E regarding the test question, and student response R to the test question. E includes six emotional dimensions: bored, concentrated, confused, frustrated, offtask, and gaming. Each dimension belongs to the range [0,1], and the closer the emotional value of a particular dimension is to 1, the stronger the student's emotion is. Student i's response to test question j is r ij It is represented as, r ijWhen = 1, it means that student i answered test question j correctly, and r ij If = 0, it means that student i answered test question j incorrectly. Data cleaning is performed to remove empty entries and redundant response records, and valid information is extracted as follows: The potential level of knowledge acquisition of student i for key point k is α ik Represented by α ik Initialize it randomly, α ik ∈[0,1], The test question matrix Q contains the relationships between all test questions and key points. q represents the situation where a test question q tests key point k. jk Represented by q jk If = 1, it means that test question j does not test point k, and q jk If = 0, it means that test question j does not test point k.

[0025] Data cleaning involves removing empty entries and response behavior records that contain redundancy. In the two datasets of this embodiment, the cleaned datasets are split into a training set, test set, and validation set in a ratio of 7:2:1, and the information is statistically analyzed and shown in Table 1.

[0026] [Table 1]

[0027] Based on S2, student emotions, student response behavior, and the test question matrix, students' external perceptions are modeled through the PCA algorithm module and the hierarchical self-attention module.

[0028] Using the PCA algorithm in S21, students' complex emotions were clustered into two types, represented as positive and negative emotions, respectively, and the expression formula was:

number

number

number

[0029] S22, in the interaction layer, two self-attention modules are used to simulate emotions obtained through clustering.

number

number

number

number

Number

Number

Number

[0030] [[ID=5C]]S23, the aggregation layer uses another self-attention module to aggregate the calculated different interactive emotions o p , o n . Specifically the results of different categories of interactive emotions o p and n are used as query and key in the attention mechanism respectively, adjust the influence of emotions on students based on attention weights, use the acquisition status of students for knowledge as value in the attention mechanism, aggregate different interactive emotions, and calculate the external cognition θ of students that fuses the emotional characteristics obtained based on students' characteristics. The expression is o .

Number

[0031] S3. Combine external cognition and the student's unique cognition, introduce the NCB-IRM framework, and comprehensively predict the student's personalized response. The response r of the student ij is known, and two extreme situations where the student's emotional characteristics affect the student's response are considered respectively.

[0032] S31. Only affected by emotional characteristics, student i answers test question j correctly, model the student's learning response function,

Equation

[0033] S32. Not affected by emotional characteristics, student i answers test question j correctly, model the student's learning response function,

Equation

[0034] S33. Assuming that the response to each question is statistically independent of the student's cognition, use the Bernoulli distribution to model the responses of all students,

number

[0035] S34, the student response matrix is ​​a comprehensive response to the test questions a priori and the student emotions a priori, and the response r ij , student cognitive θ i θ o , Test question variable a j , α j , b j d j When this is set, the item response function of student responses is,

number

[0036] S35, maximize the item response function of student responses and use a Markov chain Monte Carlo method based on Metropolis-Hastings to find θ i a j , α j d j , b j By obtaining the optimal solution, S351, First, randomize all parameters to their initial values, S352, traversing the T time point samples sequentially from the first value, S353,

number

number

number

number

number

number

number

number

number

number

[0037] S36, Predicting Personalized Student Responses

number

[0038] By combining it with student emotional data, step S3 can solve the problem that the modeling of students' individualized cognitions is not sufficiently accurate.

[0039] S4. Based on actual student responses, PMF is used to decompose commonalities in student responses and, combined with students' individualized predictive responses, collaboratively predict students' responses to test questions.

[0040] S41. Based on student responses R, perform probability matrix decomposition through PMF and the feature matrix U of the student and test questions. i , V j Having obtained, S411, assuming that the distribution of the feature vectors of the known response data and students and test questions satisfies a Gaussian distribution, we define the conditional distribution in the observed responses.

number

number

number

number

number

number

number

[0041] S42. Predicting student responses, including commonalities among students.

number

[0042] In S43, the weighted sum of the individualized predicted response of each student and the common predicted response of each student is calculated using the following formula:

number

number

number

number

number

number

[0043] By combining students' individuality and commonalities, Step S4 can address the problems of traditional recommended methods, which have low predictive power and are not comprehensive.

[0044] In S5, based on predicted student responses, provide students with test questions of appropriate difficulty and form a final recommendation list.

[0045] S51, Predicted student response R p Based on this, the difficulty levels of the test questions [d1,d2] were designed. The difficulty level of the test questions is calculated from the average student response.

number

[0046] The experimental method for this embodiment will be further described below.

[0047] During the training process, the parameters are initialized using Xavier. Specifically, these parameters are N(0,std 2 The weights are filled with random values ​​sampled from )

number

[0048] In this embodiment, the performance metrics for evaluating the recommended means proposed by the present invention are divided into two types: accuracy metrics and non-accuracy metrics. Accuracy metrics include accuracy rate (ACC), area under the ROC curve (AUC), root mean square error (RMSE), and mean absolute error (MAE), and their purpose is to determine whether the perception of emotion recognition is more consistent with the actual learning status of the students. Non-accuracy metrics include the rationality of the recommended test questions, i.e., whether the recommended test questions meet the students' needs and are consistent with the students' cognitive status. The specific meanings of these five metrics are as follows.

[0049] Because it is difficult to accurately obtain the actual values ​​of students' student cognition, it is challenging to directly assess the performance of cognitive diagnostics for student emotion recognition. Therefore, based on existing research, this embodiment monitors the model's performance using student responses. Specifically, depending on whether the predicted result is a percentage score or a response result (0 or 1), this embodiment can indirectly assess the performance of cognitive diagnostics for emotion recognition from two aspects: regression and classification. When treating the problem as a regression task, the root mean square error (RMSE) and mean absolute error (MAE) are used to quantify the distance between the predicted score and the actual score.

number

[0050] y' represents the i-th observed value, y is the i-th observed actual value, and n represents the total number of observations. Smaller RMSE and MAE values ​​indicate a higher predictive effect of the model.

[0051] When treating a problem as a classification task, the prediction results (1, 0) represent positive and negative instances, and evaluation metrics such as Area Under Curve (AUC) and Accuracy (ACC) are commonly used.

number

[0052] M represents the number of positive samples, N represents the number of negative samples, CorrectPair represents the number of precisely ordered sample pairs out of all sample pairs, TP (True Positives) represents the number of samples that the model accurately predicted as positive, TN (True Negatives) represents the number of samples that the model accurately predicted as negative, FP (False Positives) represents the number of samples that the model incorrectly predicted as positive, and FN (False Positives) represents the number of samples that the model incorrectly predicted as negative. The values ​​of AUC and ACC are between 0 and 1, and the closer the value is to 1, the better the model's prediction results.

[0053] The test question recommendations in this embodiment do not aim to recommend the most difficult / easiest test questions to students, but rather attempt to recommend test questions that are appropriate to the students' cognitive levels. Therefore, the evaluation metric is set as the percentage of correct answers to the recommended test questions, i.e., the ratio of correctly answered test questions to the number of recommended test questions.

number

[0054] In this example, three frameworks in CDM—IRT, MIRT, and DINA—were used for comparison, and E-CDM represents a CDM that uses the attention layer. Their predictive effects in ASSIST2009 are shown in Table 2.

[0055] IRT: A continuous, one-dimensional CDM employs a Logistic function to describe students' knowledge acquisition, providing interpretable parameters (e.g., students' latent characteristics, question distinguishability, and difficulty) to analyze student performance.

[0056] MIRT is a continuous, multidimensional CDM that has a Logistic Item Response Function and is obtained by extending a one-dimensional IRT.

[0057] DINA: The test question key matrix Q is introduced, and attention is paid to the influence of student slides and inference behavior. Discretized binary vectors are used to represent students' knowledge acquisition, making it a discretized multidimensional CDM.

[0058] In CDM, the distinction level of test questions and the difficulty level of test questions in IRT are both a j =4, d j = 0, and in MIRT, a j =0, d j = 0, and the DINA slide, inference, and step size are max slip =0.4, max guess =0.4, max step = is set to 1000. The input structures for CDM are IRT(maximum student id+1, maximum test question id+1), MIRT(maximum student id+1, maximum test question id+1, latent_number), and DINA(maximum student id+1, maximum test question id+1, knowledge_number). In this example, latent_number = knowledge_number, where in the ASSIST2009 dataset the input structure for IRT is (4164,17747), the input structure for MIRT is (4164,17747,102), and the input structure for DINA is (4164,17747,102). Similarly, in the ASSIST2017 dataset the input structure for IRT is (7784,3164), the input structure for MIRT is (7784,3164,123), and the input structure for DINA is (7784,3164,123).

[0059] Below, we will analyze the experimental results by combining diagrams and tables.

[0060] Since ASSIST2009 does not contain data related to emotions, we will first examine the effect of the self-attention layer using this dataset. We will use student response logs as input, p j Let be both the query and key of the self-attention, s iBy using this as the value of self-attention, the weighting of the influence of test questions on student learning is adjusted.

[0061] [Table 2]

[0062] (1) After the ASSIST2009 dataset was modified based on the CDM and the self-attention module was added, overall, the ACC of all E-CDMs increased compared to the CDM, and the MAE of E-CDMs decreased compared to the CDM. This means that the self-attention module can more effectively capture the internal relationships between students and test questions.

[0063] (2) As can be seen from the model results, the effect of E-IRT (ACC=0.659, MAE=0.375) is optimal in these models, which means that IRT is relatively well-suited to the ASSIST2009 dataset.

[0064] To investigate whether emotional input further enhances the model effect, this embodiment conducted experiments on the ASSIST2017 dataset, which includes emotional data. A-CDM and AE-CDM were added to ASSIST2009. Here, A-CDM refers to adding students' emotional characteristics as model input to CDM and conducting experiments using the NCB-IRM framework, while AE-CDM is the cognitive diagnostic model of emotion recognition proposed in this embodiment, which models students' emotions as external perceptions using PCA and hierarchical self-attention based on NCB-IRM. The prediction results of students' responses in ASSIST2017 are shown in Table 3.

[0065] [Table 3]

[0066] (1) Compared to models that only pay attention to student response logs (CDM, E-CDM), CDM models that integrate sentiment features (A-CDM, AE-CDM) showed significantly improved performance. When only sentiment features were added as input, the results were IRT (AUC=0.535, RMSE=0.575), MIRT (AUC=0.551, RMSE=0.633), DINA (AUC=0.508, RMSE=0.572), and A-IRT (AUC=0.701, RMSE=0.465), A-MIRT (AUC=0.608, RMSE=0.533), A-DINA (AUC=0.543, RMSE=0.554), which means that the addition of sentiment features has a significant impact on the predictive ability of the model.

[0067] (2) AE-CDM also showed some improvement compared to A-CDM. Based on the evaluation index, the AUC value of AE-IRT increased by 0.040 and the MAE value decreased by 0.019 compared to A-IRT; the AUC value of AE-MIRT increased by 0.034 and the MAE value decreased by 0.034 compared to A-MIRT; and the AUC value of AE-DINA increased by 0.023 and the MAE value decreased by 0.017 compared to A-DINA. From this, it can be seen that the self-attention module plays a modal role in students' cognition of emotional characteristics.

[0068] (3) Of the three CDMs, the IRT prediction results were optimal, and the improvement effect of EA-IRT was the most significant. The AUC of EA-IRT increased by 38.64% compared to IRT, and the ACC of EA-IRT increased by 45.12% compared to baseline IRT.

[0069] As shown in Figure 3, this embodiment investigated the relationship between students' mastery of test questions and the recommended difficulty level.

[0070] In the ASSIST2017 dataset, as the difficulty of the recommended test questions increased, students' RACC for the recommended questions continued to decrease, meaning that these models can recommend test questions at an appropriate difficulty level according to the students' needs. Experimental results show that the RACC of AE-PMFDINA (applying the invention to the DINA framework) proposed by the present invention can exceed that of PMF-DINA. The present invention can more accurately model students' cognition, more accurately predict students' scores, and recommend test questions within a set recommended difficulty range to each student after introducing students' emotional characteristics.

[0071] This example further demonstrates the test question recommendations for two students through case analysis. Figure 4 shows the mastery levels of students A and B for the six key points in ASSISIT2017. From this, it can be seen that student A has a relatively good grasp of key points such as square-root and pattern-finding, but has not a very good grasp of key points such as area and probability. Student B has a relatively good grasp of key points such as area and pattern-finding, but has not a very good grasp of key points such as square-root, probability, and equation-solving.

[0072] In the test dataset, if we select and recommend test questions with a difficulty level of 0.3 to 0.5, where both student A and student B have a 50% correct answer rate, then in this invention, the recommended test questions for student A are questions 4558, 2115, and 1624, and the recommended test questions for student B are questions 894, 2401, and 1597.

[0073] As is clear from Figure 4(a), student A has not mastered the key points area and probability, and Figure 4(b) shows the correspondence between some test questions and key points. Recommended test questions 4558 test the area key point, and question 2115 tests the probability key point. Student B has not mastered the key points square-root and probability, and recommended test questions 894 test the square-root key point, and question 2401 tests the probability key point. Since student A has mastered square-root well, no relevant test questions are recommended for that student.

[0074] As can be seen from this case analysis, the personalized test question recommendation method proposed by the present invention recommends test questions of a high difficulty level, and can recommend appropriate test questions according to each student's individual learning status, and the test question recommendation results have a very high degree of interpretability.

[0075] Example 2 A computer program is stored in the computer-readable storage medium according to this embodiment, and when the program is executed by the processor, the steps of the personalized online test question recommendation method for student emotion recognition according to Embodiment 1 are implemented.

[0076] The computer-readable storage medium in this embodiment may be an internal storage unit of the terminal, such as the terminal's hard disk or memory; the computer-readable storage medium in this embodiment may be an external storage device of the terminal, such as a plug-in hard disk, smart memory card, secure digital card, or flash card; and furthermore, the computer-readable storage medium may include both the terminal's internal storage unit and the external storage device.

[0077] The computer-readable storage medium in this embodiment is used to store computer programs and other programs and data necessary for the terminal, and the computer-readable storage medium may also be used to temporarily store output or output data.

[0078] Example 3: The computer device according to this embodiment includes memory, a processor, and a computer program stored in memory and executable on the processor, and when the processor executes the program, it implements the steps of the personalized online test question recommendation method for student emotion recognition according to Embodiment 1.

[0079] In this embodiment, the processor may be a central processing unit, or other general-purpose processor, digital signal processor, application-specific integrated circuit, field-programmable gate array signal or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc. The general-purpose processor may be a microprocessor, or the processor may be any ordinary processor, etc. The memory may include read-only memory and random access memory and provide instructions and data to the processor, and a portion of the memory may include non-volatile random access memory. For example, the memory may store device type information.

[0080] Those skilled in the art should understand that the contents disclosed in the examples may be provided as methods, systems, or computer program products. Accordingly, the present invention may take the form of hardware examples, software examples, or examples combining software and hardware. Furthermore, the solution may take the form of a computer program product that can be implemented on one or more computer-compatible storage media (including, but not limited to, disk memory and optical memory) containing computer-compatible program code.

[0081] This solution will be described with reference to the methods and flowcharts and / or block diagrams of computer program products relating to embodiments of this solution, but it should be understood that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions are provided to the processor of a general-purpose computer, dedicated computer, embedded processor, or other programmable data processing device to generate a device, thereby generating a device for realizing one flow in the flowchart or one or more blocks in multiple flows and / or block diagrams.

[0082] These computer program instructions may be stored in computer-readable memory that can guide a computer or other programmable data processing device to operate in a particular manner, thereby generating a product including an instruction unit, which implements the functions specified in one or more flows of a flowchart and / or one or more blocks of a block diagram.

[0083] These computer program instructions may be loaded into a computer or other programmable data processing device to cause the computer or other programmable device to perform a series of operational steps to generate processing implemented by the computer, thereby providing steps for implementing a function specified in one or more flows of a flowchart and / or one or more blocks of a block diagram.

[0084] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by issuing instructions to the relevant hardware via a computer program, which may be stored in a computer-readable storage medium, and when the program is executed, it may include the processes of each embodiment of the above-described method. Here, the storage medium may be a magnetic disk, an optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0085] The examples described in this invention are merely to illustrate preferred embodiments of the invention and do not limit the concept or scope of the invention. Those skilled in the art can make various modifications and improvements to the technical solutions of the invention without departing from the design concept of the invention, and these should fall within the scope of protection of the invention.

Claims

1. A recommended method for personalized online test questions for students' emotional recognition, Step S1 involves collecting records of students' online learning response behavior and corresponding emotions, performing data cleaning, and extracting useful information. Step S2 involves classifying students' complex emotions using PCA based on student emotional data, adjusting the influence of emotions on students through the weighting of the self-attention module, and modeling students' external perceptions. Step S3 involves combining students' external perceptions and their intrinsic perceptions to introduce the NCB-IRM framework and comprehensively predict students' individualized responses. Step S4 involves using PMF to decompose commonalities in students' responses and combining them with students' individualized predictive responses to collaboratively predict students' responses to test questions. A personalized online test question recommendation method for student sentiment recognition, comprising step S5, which involves providing students with test questions of appropriate difficulty based on predicted student responses and forming a final recommendation list.

2. Step S1, which involves collecting online learning response records including students' emotions and extracting useful information, specifically involves: S11, collect student ID, test question ID, key point ID, student sentiment data E regarding the test question, and student response R to the test question. E includes six emotional dimensions: bored, concerned, confused, frustrated, offtask, and gaming, and each dimension has an emotional element A ｉｊ It belongs to the range [0, 1], and the closer it is to 1, the higher the student's performance in relation to this emotion. Student i's response to test question j is r ｉｊ It is represented as, r ｉｊ When = 1, it means that student i answered test question j correctly, and r ｉｊ If the value is 0, it means that student i answered test question j incorrectly. S12. Data cleaning is performed to remove empty entries and redundant response records, and valid information is extracted as follows: The potential level of knowledge acquisition of student i regarding key point k is extracted from the student's behavior, and α ｉｋ It is represented as α ｉｋ ∈[0,1], The test problem matrix Q includes the relationships between all test problems and key points, and the element q ｊｋ represents the situation where each test problem q tests a key point. When q ｊｋ = 1, it means that test problem j does not test key point k. When q ｊｋ = 0, it means that test problem j does not test key point k. The method according to claim 1, characterized in that.

3. In step S2, which models students' external perceptions through the PCA algorithm module and the hierarchical self-attention module based on student emotions, student response behavior, and the test question matrix, specifically, S21, using the PCA algorithm, students' complex emotions were clustered into two types, represented as positive and negative emotions, and the expression formula was [Number 65] And, E is a matrix containing six emotional dimensions, and ・ represents the dot product operation of vectors. [Number 66] is the projected value of the emotion sample in the direction of the first two principal components, V １ , V ２ Each of these represents a feature vector, S22, in the interaction layer, two self-attention modules are used to simulate emotions obtained by clustering ( [Number 67] (For example) Students' potential level of acquisition of key points α ｉｋ And the test questions test the key points q ｊｋ Interact with it, Positive emotions [Number 68] When inputting, the self-attention module will input the sequence of key points to be tested in the test question. ｊ Using this as a key in the attention mechanism, emotional performance [Number 69] and sequences of students' potential learning status for key points ｉ These are the query and value in the attention mechanism, respectively. First, the student's test question answer vector p ｊ and positive emotional vector [Number 70] The weights of the interaction's mutual influence ω ｐ We calculate this using cosine similarity, and then, after considering the interactive effects of the features, ω ｐ and the students' potential level of understanding of the key points ｉ Through a weighted sum, the positive emotional vector after the interaction is obtained. ｐ Having obtained, the expression is [Number 71] And, p ｊ = (q ｊ１， q ｊ２， ...q ｊｋ ), s ｉ = (α ｉ１， α ｉ２， …α ｉｋ ) and Similarly, negative emotions [Number 72] When inputting, the interaction layer simulates the interactive relationship between the student's negative emotions, the student's acquisition of knowledge, and the test question answers through self-attention, and the negative emotion vector after the interaction. ｎ Having obtained, [Number 73] S23, the collective layer uses another self-attention module to calculate different interactive emotions. ｐ , o ｎ Gather them together, Interactive emotion results from different categories ｐ and o ｎ These are used as queries and keys in the attention mechanism, respectively, and the student's acquisition status of knowledge is used as a value in the attention mechanism. By aggregating different interactive emotions, the student's external perception θ integrates emotional characteristics. ｏ Having obtained this, the expression is, [Number 74] The method according to claim 1, characterized in that it is the same.

4. Step S3, which combines external perception and the student's inherent perception to introduce the NCB-IRM framework and comprehensively predict the student's individualized response, involves the student's response r ｉｊ It is known that students' emotional characteristics influence their responses in two extreme situations, and specifically, In S31, under the influence of only emotional characteristics, student i correctly answers test question j, and the student's learning response function is modeled. [Number 75] θ ｏ , α ｊ , b ｊ These represent the student's level of external perception, their ability to distinguish between test questions, and the difficulty level of the test questions, respectively. In S32, student i answered test question j correctly, unaffected by emotional characteristics, and the student's learning response function was modeled. [Number 76] r ｉｊ This represents student i's response to test question j, and A ｉｊ θ represents the emotions of student i when performing to answer test question j, ｉ a ｊ d ｊ These represent the student's internal cognitive level, the ability to distinguish between test questions, and the difficulty level of the test questions, respectively. S33, assuming that the responses to each question are statistically independent of the students' cognition, we model the responses of all students using a Bernoulli distribution. [Number 77] η ｉｊ and ζ ｉｊ Each of these represents the probability that student i correctly answered question j based on their practice with the test questions and their emotional characteristics. S34, the student response matrix is ​​a comprehensive response to the test questions a priori and the students a priori, and the response r ｉｊ , student cognitive θ ｉ θ ｏ , Test question variable a ｊ , α ｊ , b ｊ d ｊ When this is set, the item response function of student responses is, [Number 78] And, S35. Maximize the item response function of the student response and use the Markov chain Monte Carlo method based on Metropolis-Hastings to determine θ ｉ a ｊ , α ｊ , b ｊ d ｊ By obtaining the optimal solution, S36. Predicting individualized student responses. [Number 79] The method according to feature 1.

5. In step S4, which uses PMF to decompose commonalities in students' responses based on their actual responses and combines this with students' individualized predicted responses to collaboratively predict students' responses to test questions, specifically, S41, Probability matrix decomposition is performed on the actual responses of students through PMF, and the characteristic matrix U of the students and test questions is obtained. ｉ , V ｊ Having obtained, S42, Predict responses that include commonalities among students. [Number 80] In S43, the weighted sum of the individualized predicted response of each student and the common predicted response of each student is calculated using the following formula. [Number 81] R ｐ is the predicted student response, and μ is the average score of all students. [Number 82] The method according to claim 1, wherein R' is the predicted individualized response of the student, R' is the student commonality response predicted by PMF, and the ratio of student individuality to commonality is adjusted by the parameter ρ.

6. In step S5, which provides students with test questions of appropriate difficulty based on predicted student responses and forms a final recommendation list, specifically, S51, Predicted student response R ｐ Based on this, the difficulty level of the test questions [d １ d ２ Designed ] The difficulty level of the test questions is calculated from the average student response. [Number 83] r ｉｊ is student i's response to test question j, and n is the number of times student i has answered test question j. S52, difficulty level of the test questions: d ｊ Based on this, a recommended list of test questions R ｐ ∈[1-d ２ , 1-d １ A method for recommending personalized online test questions for student emotion recognition according to claim 1, characterized by screening for ].

7. A computer-readable storage medium in which a computer program is stored, characterized in that, when the program is executed by a processor, it implements the steps of the personalized online test question recommendation method for student emotion recognition described in any of claims 1 to 6.

8. A computer device comprising memory, a processor, and a computer program stored in memory and executable on the processor, wherein when the processor executes the program, it implements the steps of a personalized online test question recommendation method for student emotion recognition as described in any one of claims 1 to 6.

Images