WO2025086355A1 - 学生情感感知的个性化在线试题推荐方法 - Google Patents

学生情感感知的个性化在线试题推荐方法 Download PDF

Info

Publication number
WO2025086355A1
WO2025086355A1 PCT/CN2023/131750 CN2023131750W WO2025086355A1 WO 2025086355 A1 WO2025086355 A1 WO 2025086355A1 CN 2023131750 W CN2023131750 W CN 2023131750W WO 2025086355 A1 WO2025086355 A1 WO 2025086355A1
Authority
WO
WIPO (PCT)
Prior art keywords
students
student
test
response
question
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
PCT/CN2023/131750
Other languages
English (en)
French (fr)
Inventor
魏昕
周亮
陈铭子
戴文婷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to JP2024571904A priority Critical patent/JP2026511279A/ja
Publication of WO2025086355A1 publication Critical patent/WO2025086355A1/zh
Anticipated expiration legal-status Critical
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/20—Education
    • G06Q50/205—Education administration or guidance

Definitions

  • the present invention relates to the technical field of online learning big data mining, and in particular to a personalized online test question recommendation method based on student emotion perception.
  • online learning platforms provide students with a convenient way to practice test questions. Students can obtain rich question bank resources through online learning platforms and learn knowledge anytime and anywhere. Online learning platforms provide teachers and students with a wide range of convenient and intelligent learning methods. In recent years, many online learning platforms have emerged to provide students with test questions, such as ASSISTment, Coursera, EdX, etc. However, due to the large number of test question resources, students often find it difficult to effectively find test questions that suit them, which affects their learning interest and efficiency. Therefore, it is very necessary to use educational big data mining technology to accurately recommend suitable test question resources to students based on the personal characteristics of specific students.
  • Recommendation systems in the field of education can provide students with a variety of learning resources and test resources, such as video courses, practice test questions, etc.
  • Recommendation systems are an important branch of information retrieval in educational big data mining. Their goal is to provide students with learning materials that meet their needs and assist them in practicing test questions.
  • Traditional online resource recommendation methods are mainly based on collaborative filtering algorithms, which construct a user-resource matrix and then recommend learning resources based on the similarity of users or resources. However, in online learning scenarios, these methods have some defects: they do not fully consider the personalized characteristics of students; the recommended content lacks diversity and rationality, resulting in the test resources provided by the recommendation system failing to meet the personalized needs of students and failing to conduct targeted training.
  • the emotion-aware test question recommendation can improve the accuracy of diagnosis on the one hand; on the other hand, it can utilize more personalized knowledge of students.
  • CDM does not consider the commonality of students' answers, and the recommended test questions have a small coverage, which leads to sparse data in the subsequent recommendation process and certain limitations of the recommended test questions.
  • the purpose of the present invention is to provide a personalized online test question recommendation method based on student emotion perception, introduce student emotional information, design an online test question recommendation scheme based on the individuality and commonality of students' test questions, recommend test questions suitable for students, and improve students' practice effects.
  • a method for recommending personalized online test questions based on student emotion perception comprising the following steps:
  • PCA is used to classify students’ complex emotions, and the influence of emotions on students is adjusted through the weight of the self-attention module to model students’ external cognition;
  • step S1 collects the online learning answering behavior records including the student's emotions and extracts effective information; specifically:
  • S11 collect student ID, test question ID, knowledge point ID, student's emotional data E on the test question, and student's response R on the test question;
  • W contains six emotional dimensions: bored, concentrated, confused, frustrated, offtask, and gaming; the emotional element A ij of each dimension belongs to [0,1]. The closer it is to 1, the stronger the student's performance in this emotion.
  • the test question matrix Q contains the relationship between all test questions and knowledge points.
  • step S2 models students' external cognition through PCA algorithm module and hierarchical self-attention module based on students' emotions, students' answering behavior and test matrix, specifically:
  • E is a matrix containing six emotion dimensions
  • ⁇ represents the dot product operation of the vector
  • V 1 and V 2 represent the feature vectors respectively;
  • the self-attention module takes the sequence pj of knowledge points tested in the test as the key in the attention mechanism, the emotion representation and the student’s potential knowledge point mastery sequence s i are used as the query and value in the attention mechanism respectively;
  • the cosine similarity is used to calculate the student answer question vector pj and the positive emotion vector
  • the mutual influence weight of the interaction is ⁇ p ; then, after considering the feature interaction, the positive emotion vector after the interaction, OP , is obtained by weighting ⁇ p and the student's potential mastery of the knowledge point s i , and the expression is:
  • the interaction layer simulates the interaction between students' negative emotions, students' knowledge mastery and answering questions through self-attention, and obtains the negative emotion vector o n after interaction;
  • the aggregation layer uses another self-attention module to aggregate the different interactive emotions o p and o n calculated;
  • the interactive emotion results of different categories, o p and o n, are used as the query and key in the attention mechanism respectively, and the students’ knowledge mastery is used as the value in the attention mechanism.
  • Different interactive emotions are aggregated to obtain the students’ external cognition ⁇ o with integrated emotional features.
  • the expression is:
  • step S3 integrates external cognition and students' inherent cognition, introduces the NCB-IRM framework, and comprehensively predicts students' personalized responses; given the student response r ij , two extreme cases in which students' emotional characteristics affect their responses are considered respectively; specifically:
  • ⁇ o , ⁇ j , and b j are the students' external cognitive level, the test item discrimination, and the test item difficulty, respectively;
  • r ij represents the response of student i to test j
  • a ij represents the emotion expressed by student i in answering test j
  • ⁇ i , a j , d j are the student's intrinsic cognitive level, test item discrimination and test item difficulty respectively;
  • the student response matrix is a comprehensive response to the test question prior and the student prior.
  • the item response function of the student response is as follows:
  • the item response function that maximizes the student response is obtained by a Markov chain Monte Carlo method based on Metropolis-Hastings to obtain the optimal solution of ⁇ i , a j , ⁇ j , b j , d j ;
  • step S4 is based on the actual responses of students, uses PMF to decompose the commonality of students' answers, combines students' personalized predicted responses, and jointly predicts students' responses to test questions; specifically:
  • Rp is the predicted student response
  • ⁇ is the average score of all students
  • r′ij is the predicted student personalized response
  • R′ is the student common response predicted by PMF
  • the ratio between students’ individuality and commonality is adjusted by parameter ⁇ .
  • step S5 provides students with test questions of appropriate difficulty based on the predicted student responses to form a final recommendation list; specifically:
  • the difficulty of the test question is calculated by the mean of the students' responses:
  • r ij is the response of student i to test question j
  • n is the number of records of answering test question j
  • the present invention uses PMF to mine the common characteristics of students' answers and combines them with students' personalized prediction responses, which not only improves the prediction ability of the model but also ensures the comprehensiveness of the recommendation plan.
  • the present invention designs the difficulty of test questions according to the students' mastery of the test questions after fully considering the individual characteristics of the students and the common characteristics of the test answers, thereby improving the rationality of the recommended test questions.
  • a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the student emotion-aware personalized online test question recommendation method of the present invention are implemented.
  • a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the student emotion-aware personalized online test question recommendation method of the present invention are implemented.
  • the present invention has at least the following beneficial effects:
  • NCB-IRM Based on PMF, this paper introduces NCB-IRM and proposes a personalized online test question recommendation method based on student emotional perception.
  • NCB-IRM is used to combine the external cognition of student emotional perception with the internal cognition of students, and the personalized needs of students are considered more comprehensively.
  • PMF is used to explore the commonality of students' answers, which not only ensures the comprehensiveness of the recommendation plan, but also improves the rationality of the recommended test questions.
  • NCB-IRM an external cognitive modeling method based on students' emotional perception is designed.
  • PCA is used to reduce the complexity of sentiment analysis.
  • a hierarchical self-attention method is used to adjust the impact on students according to different sentiment weights. Based on students' emotions, an effective method is provided for mining students' personalized needs.
  • Fig. 1 is a flow chart of the method of the present invention
  • Figure 2 is a schematic diagram of the external cognitive modeling structure of students' emotion perception
  • FIG3 is a graph showing the accuracy of students' answers to questions of different difficulty levels according to the present invention.
  • Figure 4 shows examples of two students’ cognitive results and recommended test questions
  • 4(a) shows the enhanced cognition of the diagnosed students A and B
  • 4(b) shows the test questions recommended for students A and B and the relationship between the test questions and knowledge points.
  • the data set in this embodiment comes from the online learning platform ASSISTment. Data from 2009 and 2017 were collected from the platform. ASSIST2009 contains 401,756 student test practice records, and ASSIST2017 adds student emotional data, including 942,816 student test practice records. When each student answers a test question, his or her response result and emotional data will be recorded.
  • a personalized online test question recommendation method based on student emotion perception is described in this embodiment.
  • the steps include:
  • Student id test question id, knowledge point id, student's emotion E on the test question, student's response R on the test question;
  • E contains six emotional dimensions: bored, concentrated, confused, frustrated, offtask, and gaming. Each dimension belongs to [0,1]. The closer the value of a certain dimension is to 1, the stronger the student's emotion is.
  • the potential knowledge mastery of student i at knowledge point k is represented by ⁇ ik , which is randomly initialized, ⁇ ik ⁇ [0,1];
  • test matrix Q contains the relationship between all test questions and knowledge points.
  • Data cleaning refers to deleting empty entries and redundant answer behavior records.
  • the cleaned data sets are divided into training set, test set, and validation set in a ratio of 7:2:1.
  • the statistical information is shown in Table 1.
  • S21 uses the PCA algorithm to cluster students' complex emotions into two categories, which are positive emotions and negative emotions, respectively.
  • the expression is:
  • S22 uses two self-attention modules in the interaction layer to simulate the clustered emotions (with For example) interact with students' potential mastery of knowledge points ⁇ ik and test questions' knowledge points q jk ;
  • the self-attention module takes the test question knowledge point sequence p j as the key in the attention mechanism and the emotion representation The sequence si of students’ potential mastery of knowledge points is used as the query and value in the attention mechanism respectively;
  • the cosine similarity is used to calculate the student answer question vector pj and the positive emotion vector
  • the mutual influence weight ⁇ p of the interaction is used to assign weights to students' positive emotions based on the characteristics of the test questions.
  • the positive emotion vector op after interaction is obtained by weighting ⁇ p with the potential mastery of students' knowledge points si , and the influence of positive emotions on students' knowledge mastery is adjusted.
  • the expression is:
  • the interaction layer understands the interactive relationship between students' negative emotions, students' knowledge mastery and answering questions through self-attention, and obtains the negative emotion vector o n after interaction;
  • the S23 aggregation layer uses another self-attention module to aggregate the different calculated interaction emotions o p and o n ;
  • Step S2 can solve the problem that the modeling mechanism of students' external cognition is unclear.
  • ⁇ o , ⁇ j , and bj are the students' external cognitive level, the test item discrimination, and the test item difficulty, respectively;
  • r ij represents the response of student i to test j
  • a ij represents the emotion expressed by student i in answering test j
  • ⁇ i , a j , d j are the student's intrinsic cognitive level, test item discrimination and test item difficulty respectively;
  • ⁇ ij and ⁇ ij represent the probability that student i correctly answers question j based on the test questions practiced and the sentiment characteristics, respectively;
  • the student response matrix is a comprehensive response to the test question prior and the student's emotional prior; given the response r ij , student cognition ⁇ i , ⁇ o , test question variables a j , ⁇ j , b j , d j , the item response function of the student response is as follows:
  • S35 maximizes the item response function of student responses and obtains the optimal solution of ⁇ i , a j , ⁇ j , d j , b j through a Markov chain Monte Carlo method based on Metropolis-Hastings:
  • S351 first randomizes all parameters to initial values
  • S352 starts from the first number and sequentially traverses the samples at T moments;
  • S353 collects a sample from ⁇ i(t) ⁇ U( ⁇ i(t-1) - ⁇ ⁇ i , ⁇ i(t-1) + ⁇ ⁇ i ) and calculates the probability Accept ⁇ i ;
  • S354 collects a sample from a j(t) ⁇ U(a j(t-1) - ⁇ aj ,a j(t-1) + ⁇ aj ) and generates a sample with probability Accept a j ;
  • S355 collects a sample from ⁇ j(t) ⁇ U( ⁇ j(t-1) - ⁇ ⁇ j , ⁇ j(t-1) + ⁇ ⁇ j ) and generates a sample with probability Accept ⁇ j ;
  • S356 collects a sample from d j(t) ⁇ U(d j(t-1) - ⁇ dj ,d j(t-1) + ⁇ dj ) and generates a sample with probability Accept d j ;
  • S357 collects a sample from bj (t) ⁇ U( bj(t-1) - ⁇ bj , bj (t-1) + ⁇ bj ) and generates a sample with probability Accept b j ;
  • S358 traverses all T moments, and if the convergence condition is met, outputs the estimates of ⁇ i , a j , ⁇ j , d j , b j ;
  • step S3 can solve the problem of inaccurate personalized cognitive modeling of students.
  • S4 is based on students’ real responses, uses PMF to decompose the common characteristics of students’ answers, and combines it with students’ personalized predicted responses to jointly predict students’ responses to test questions.
  • S41 performs probability matrix decomposition based on the student response R through PMF to obtain the characteristic matrices U i and V j of the student and the test questions;
  • S411 assumes that the known response data and the distribution of student and test question feature vectors satisfy Gaussian distribution, and defines the conditional distribution on the observed response:
  • U i represents the potential characteristics of students
  • V j represents the potential characteristics of test questions
  • N is the Gaussian distribution probability density function with mean ⁇ and variance ⁇ 2 .
  • S412 calculates the posterior probability of the potential characteristics of students and test questions as follows:
  • R is the true value
  • Ui and Vj represent the latent feature vectors of students and test questions respectively
  • I ij is the indicator function (if student i answers If the test question j is passed, the value is 1; otherwise, it is 0);
  • S415 updates U i ,V j using stochastic gradient descent method:
  • S43 calculates the weighted sum of the student's personalized prediction response and the student's common prediction response, as follows:
  • R p is the predicted student response
  • ⁇ is the average score of all students
  • r′ ij is the predicted student personalized response
  • R′ is the student common response predicted by PMF
  • the ratio between students’ individuality and commonality is adjusted by parameter ⁇ ;
  • Rp is the predicted student response
  • R is the true value
  • Ui and Vj represent the D-dimensional latent feature vectors of students and test questions, respectively.
  • S432 estimates r′ ij by using a Markov chain Monte Carlo method based on Metropolis-Hastings, estimates the latent factor vectors U i , V j by maximizing the posterior probability, and outputs the trained predicted response R p .
  • step S4 can solve the problem that the traditional recommendation scheme has weak prediction ability and the recommendation method The problem of incomplete case.
  • S5 provides students with test questions of appropriate difficulty based on the predicted student responses to form a final recommendation list.
  • the difficulty of the test question is calculated by the mean of the students' responses:
  • r ij is the response of student i to test question j
  • n is the number of records of answering test question j
  • Step S52 Filter out a list of recommended test questions R p ⁇ [1-d 2 , 1-d 1 ] according to the test question difficulty d j .
  • Step S5 Recommendations are made according to the students' mastery of the test questions, which can solve the problem of unreasonable test question recommendation schemes.
  • Xavier is used to initialize the parameters. Specifically, these parameters are filled with random values sampled from N(0,std 2 ), where n i represents the input dimension of the neural network, and n i+1 represents the output dimension of the neural network.
  • precision indicators include accuracy (ACC), area under the ROC curve (AUC), root mean square error (RMSE) and mean absolute error (MAE), the purpose of which is to measure whether the cognition of emotional perception is more in line with the students' actual mastery
  • non-precision indicators include the rationality of the recommended test questions, that is, whether the recommended test questions meet the needs of students and whether they are in line with the students' cognitive situation. The specific meanings of these five indicators are as follows:
  • this embodiment uses student responses to supervise model performance. Specifically, depending on whether the predicted result is a score rate or an answer result (0 or 1), this embodiment can indirectly evaluate the performance of emotion perception cognitive diagnosis from two perspectives: regression and classification.
  • regression and classification When the problem is treated 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.
  • RMSE root mean square error
  • MAE mean absolute error
  • y' represents the observed value of the i-th observation
  • y represents the actual value of the i-th observation
  • n represents the total number of observations. The smaller the RMSE and MAE values are, the better the prediction effect of the model is.
  • the predicted result (1, 0) represents positive and negative instances
  • the evaluation indicators are usually the area under the ROC curve (AUC) and the prediction accuracy (Accuracy, ACC).
  • M represents the number of positive samples
  • N represents the number of negative samples
  • CorrectPair represents the correctly ordered sample pairs among all sample pairs
  • TP True Positive
  • TN True Negative
  • FP False Positive
  • FN False Negative
  • AUC and ACC are between 0 and 1. The closer the value is to 1, the better the prediction result of the model.
  • the test question recommendation in this embodiment does not consider recommending the most difficult/easiest test questions to students, but tries to recommend test questions that meet their cognitive level. Therefore, the evaluation index is set as the correct answer rate of the recommended test questions, that is, the ratio of the number of correctly answered test questions to the number of recommended test questions:
  • the numerator represents the number of correct answers given by students in the recommended test questions, and the denominator represents the number of recommended test questions. The larger the RACC, the closer the recommended test questions are to the students' cognitive level.
  • IRT It uses a logistic function to describe students’ knowledge mastery. It can provide interpretable parameters (such as students’ latent characteristics, problem discrimination and difficulty) to analyze students’ performance. It is a continuous one-dimensional CDM.
  • MIRT It has a logistic item response function and is an extension of the unidimensional IRT. It is a continuous multidimensional CDM.
  • DINA It introduces the test knowledge point matrix Q, and also pays attention to the influence of students' sliding and guessing behaviors. It uses discrete binary vectors to represent students' knowledge mastery and is a discrete multidimensional CDM.
  • the input structures of CDM are: IRT (maximum student id + 1, maximum question id + 1), MIRT (maximum student id + 1, maximum question id + 1, latent_number), and DINA (maximum student id + 1, maximum question id + 1, knowledge_number).
  • latent_number knowledge_number, where, in the ASSIST2009 data set, the input structure of IRT is (4164, 17747), the input structure of MIRT is (4164, 17747, 102), and the input structure of DINA is (4164, 17747, 102); similarly, in the ASSIST2017 data set, the input structure of IRT is (7784, 3164), the input structure of MIRT is (7784, 3164, 123), and the input structure of DINA is (7784, 3164, 123).
  • A-CDM and AE-CDM are added on the basis of ASSIST2009.
  • A-CDM refers to adding students' emotional characteristics as model input on the basis of CDM, and conducting experiments in the NCB-IRM framework;
  • AE-CDM is the cognitive diagnosis model of emotion perception proposed in the embodiment, which uses PCA and hierarchical self-attention modeling of students' emotions as external cognition on the basis of NCB-IRM.
  • the prediction results of students' answers in ASSIST2017 are shown in Table 3.
  • AE-CDM also has a certain degree of improvement over A-CDM.
  • the evaluation indicators compared with A-IRT, the AUC value of AE-IRT increased by 0.040 and the MAE value decreased by 0.019; compared with A-MIRT, the AUC value of AE-MIRT increased by 0.034 and the MAE value decreased by 0.034; compared with A-DINA, the AUC value of AE-DINA increased by 0.023 and the MAE value decreased by 0.017.
  • IRT has the best prediction results and EA-IRT has the most obvious improvement effect.
  • the AUC of EA-IRT is 38.64% higher than that of IRT, and the ACC of EA-IRT is 45.12% higher than that of the baseline IRT.
  • this embodiment explores the relationship between students' mastery of test questions and the recommended difficulty.
  • the RACC of the AE-PMFDINA (applying the present invention to the DINA framework) proposed in the present invention can exceed PMF-DINA.
  • the present invention can more accurately model students' cognition, more accurately predict students' scores, and recommend questions within the set recommended difficulty range to each student.
  • Figure 4 shows the mastery of six knowledge points by student A and student B on ASSISIT2017. It can be seen that student A has a good grasp of knowledge points such as square-root and pattern-finding, but a poor grasp of knowledge points such as area and probability; student B has a good grasp of knowledge points such as area and pattern-finding, but a poor grasp of knowledge points such as square-root, probability, and equation-solving.
  • the correct answer rates of student A and student B are both 50%.
  • the present invention recommends test questions No. 4558, No. 2115, and No. 1624 to student A, and recommends test questions No. 894, No. 2401, and No. 1597 to student B.
  • Figure 4(a) shows that student A has a weak grasp of the knowledge points area and probability.
  • Figure 4(b) shows the correspondence between some test questions and knowledge points.
  • the recommended test question 4558 tests the knowledge point area
  • the recommended test question 2115 tests the knowledge point probability.
  • Student B has a weak grasp of the knowledge points square-root and probability.
  • the test questions recommended to student B are recommended question 894, which tests the knowledge point square-root, and question 2401, which tests the knowledge point probability. Since student A has a high degree of mastery of square-root, no related test questions were recommended to the student.
  • the personalized test question recommendation method proposed in the present invention can recommend corresponding test questions to each student based on their personalized learning status when recommending test questions with higher difficulty, and the test question recommendation results are highly interpretable.
  • Embodiment 2 is a diagrammatic representation of Embodiment 1:
  • the computer-readable storage medium of this embodiment stores a computer program thereon, and when the program is executed by a processor, the steps in the student emotion-aware personalized online test question recommendation method of embodiment 1 are implemented.
  • the computer-readable storage medium of this embodiment may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal; the computer-readable storage medium of this embodiment may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card, a secure digital card, a flash memory card, etc. equipped on the terminal; further, the computer-readable storage medium may also include both an internal storage unit of the terminal and an external storage device.
  • the computer-readable storage medium of this embodiment is used to store computer programs and other programs and data required by the terminal.
  • the computer-readable storage medium can also be used to temporarily store data that has been output or is to be output.
  • Embodiment 3 is a diagrammatic representation of Embodiment 3
  • the computer device of this embodiment includes a memory, a processor, and a computer program stored in the memory and executable on the processor.
  • the processor executes the program, the steps in the student emotion-aware personalized online test question recommendation method of Embodiment 1 are implemented.
  • the processor may be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, readily available programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
  • the memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.
  • the disclosed content of the embodiments may be provided as methods, systems, or computer program products. Therefore, the present solution may adopt the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Moreover, the present solution may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program codes.
  • a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program codes.
  • These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.
  • These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.
  • the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Educational Administration (AREA)
  • Educational Technology (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • General Business, Economics & Management (AREA)
  • Primary Health Care (AREA)
  • Marketing (AREA)
  • Human Resources & Organizations (AREA)
  • General Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明公开了一种学生情感感知的个性化在线试题推荐方法，包括如下步骤：S1、采集学生进行在线学习的答题行为以及相应的情感记录，进行数据清洗，提取有效信息；S2、基于学生情感数据，利用PCA将学生的复杂情感分类，并通过自注意力模块权重调整情感对学生的影响，建模学生外在认知；S3、综合学生外在认知和学生固有认知，引入NCB-IRM框架，全面地预测学生个性化响应；S4、利用PMF分解学生答题共性，结合学生个性化预测响应，共同预测学生对试题的响应；S5、根据预测的学生响应，为学生提供合适难度的试题，形成最终的推荐列表。本发明基于PMF和NCB-IRM，结合学生个性和答题共性，根据学生的试题掌握情况，实现了试题推荐的准确性和合理性。

Description

学生情感感知的个性化在线试题推荐方法 技术领域
本发明涉及在线学习大数据挖掘技术领域，特别涉及一种学生情感感知的个性化在线试题推荐方法。
背景技术
在线学习平台的发展为学生提供了便捷的试题练习方式。学生可以通过在线学习平台获取丰富的题库资源，随时随地学习知识。在线学习平台为师生提供了一种广泛的、便利的、智能的学习方式。近年来，出现了许多为学生提供试题练习的在线学习平台，例如ASSISTment、Coursera、EdX等。但是，由于试题资源过于庞大，学生往往难以有效地找到适合自己的试题，从而影响了他们的学习兴趣和效率。因此，利用教育大数据挖掘技术，根据特定学生的个人特征，精确推荐适合的试题资源给学生是非常必要的。
教育领域中的推荐系统可以为学生提供多种形式的学习资源和试题资源，例如视频课程、练习试题等。推荐系统是教育大数据挖掘中信息检索的一个重要分支，它的目标是为学生提供与满足他们需求的学习材料，辅助他们进行试题练习。传统的在线资源推荐方法主要基于协同过滤算法，通过构建用户-资源矩阵，然后根据用户或资源的相似性来推荐学习资源。但是，在在线学习场景中，这些方法存在一些缺陷：它们没有充分考虑学生的个性化特征；推荐的内容缺乏多样性和合理性，导致推荐系统提供的试题资源不能满足学生的个性化需求，不能进行有针对性地训练。
为了挖掘学生的试题练习规律，现有的在线试题资源推荐方案多数采用了神经网络、深度学习、强化学习等人工智能算法。然而，这些方法往往过于强调推荐模型的精度，忽视了学生练习过程中表现出地个性化特征。它们未能根据学生自身的学习特点来设计个性化的推荐方案，导致推荐的试题缺乏针对性和解释性。认知诊断通过分析学生的答题日志，评估学生知识掌握情况，衡量学生的学习状态，为不同的学生提供个性化的推荐方案的提供理论基础。将认知诊断方法(CDM)用于试题推荐时既能诊断出学生对知识的掌握情况，又能推断试题的难度、区分度等特性，很大程度考虑了学生的个性化特征。由于CDM的诊断效果将直接影响试题推荐的精准程度，当CDM的诊断结果不够准确时，将造成推荐方案的性能下降。
在认知诊断中，增加数据输入维度可以有效提高认知诊断的效果。此外，利用多源信息来增强推荐系统的数据语义融合也可以丰富用户和物品的表示。因此，在增加情感维度后，情感感知的试题推荐一方面可以提升诊断的准确性；另一方面可以利用更多的学生个性化知识。然而，CDM没有考虑学生答题共性，所推荐的试题覆盖范围小，从而导致后续推荐过程中数据稀疏，所推荐的试题有一定局限性。
发明内容
发明目的：针对上述问题，本发明目的是提供一种学生情感感知的个性化在线试题推荐方法，引入学生情感信息，从学生作答试题的个性和共性出发设计在线试题推荐方案，推荐适合学生的试题，提高学生练习效果。
为了解决上述技术问题至少之一，根据本发明的一方面，提供一种学生情感感知的个性化在线试题推荐方法，包括如下步骤：
S1、采集学生进行在线学习的答题行为以及相应的情感记录，进行数据清洗，提取有效信息；
S2、基于学生情感数据，利用PCA将学生的复杂情感分类，并通过自注意力模块权重调整情感对学生的影响，建模学生外在认知；
S3、综合学生外在认知和学生固有认知，引入NCB-IRM框架，全面地预测学生个性化响应；
S4、利用PMF分解学生答题共性，结合学生个性化预测响应，共同预测学生对试题的响应；
S5、根据预测的学生响应，为学生提供合适难度的试题，形成最终的推荐列表。
2、根据权利要求1所述的方法，其特征在于，步骤S1采集包括学生情感的在线学习的答题行为记录，提取有效信息；具体为：
S11、采集包括学生id、试题id、知识点id、学生在试题上的情感数据E，学生在试题上的响应R；
W包含六个情感维度bored、concentrating、confused、frustrated、offtask、gaming；每个维度的情感元素Aij属于[0,1]，越接近1，表示学生在这种情感上的表现越强；
学生i在试题j的响应用rij表示；若rij＝1表示学生i答对试题j；若rij＝0，则表示学生i答错试题j；
S12、对数据清洗，删除包含空条目与冗余的答题行为记录，提取有效信息如下：
由学生行为提取学生i在知识点k的潜在知识掌握，用αik表示,αik∈[0,1]；
试题矩阵Q为包含了所有试题和知识点的关系，元素qjk表示每个试题q考察知识点的情况；若qjk＝1，表示试题j未考察知识点k；若qjk＝0，表示试题j未考察知识点k。
3、根据权利要求1所述的方法，其特征在于，步骤S2基于学生情感、学生答题行为和试题矩阵，通过PCA算法模块和分层的自注意力模块建模学生外在认知，具体为：
S21、利用PCA算法，将学生的复杂情感进行聚类为两类，分别表示为积极情感和消极情感，表达式为：
其中，E是包含六个情感维度的矩阵，·表示向量的点积运算，是情感样本在前两个主成分方向上的投影值，V1、V2分别表示特征向量；
S22、在交互层使用两个自注意力模块模拟聚类得到的情感，以为例，与学生知识点潜在掌握情况αik以及试题考察知识点情况qjk交互；
输入积极情感时，自注意力模块将试题考察知识点的序列pj作为注意力机制中的键，将情感表示与学生知识点潜在掌握情况序列si分别作为注意力机制中的查询与值；
首先，通过余弦相似度计算学生回答试题向量pj与积极情感向量交互的相互影响权重ωp；然后，在考虑特征交互作用后，通过ωp与学生知识点潜在掌握情况si的加权和，得到交互后的积极情感向量op，表达式为：
其中pj＝(qj1,qj2,…qjk)，si＝(αi1,αi2,…αik)；
类似地，输入消极情感时，交互层通过自注意力模拟学生消极情感、学生知识掌握与作答试题的交互关系，得到交互后的消极情感向量on；
S23、聚合层利用另一个自注意力模块，将计算的不同交互情感op、on聚合；
将不同类别的交互情感结果op和on分别作为注意力机制中的查询和键，学生的知识掌握情况作为注意力机制中的值，聚合不同的交互情感，得到融合情感特征的学生外在认知θo，表达式为：
4、根据权利要求1所述的方法，其特征在于，步骤S3综合外在认知和学生固有认知，引入NCB-IRM框架，全面地预测学生个性化响应；已知学生响应rij，分别考虑学生情感特征影响学生响应的两种极端情况；具体为：
S31、仅受到情感特征的影响，学生i正确回答了试题j，将学生学习反应函数建模：
其中，θo、αj、bj分别是学生的外在认知水平、试题区分度和试题难度；
S32、未受到情感特征的影响，学生i正确回答了试题j，将学生学习反应函数 建模为：
其中，rij表示学生i在试题j的响应，Aij表示学生i回答试题j表现的情感，θi、aj、dj分别是学生的内在认知水平、试题区分度和试题难度；
S33、假设每个问题的响应在统计上独立于学生的认知，使用伯努利分布对所有学生的响应进行建模：
其中ηij和ζij分别代表学生i根据通过试题练习和情感特征正确回答问题j的概率；
S34、学生的响应矩阵是对试题先验、学生先验的综合响应；给定响应rij，学生认知θi、θo，试题变量aj、αj、bj、dj，学生响应的项目反应函数如下式：
S35、最大化学生响应的项目反应函数，通过一种基于Metropolis-Hastings的马尔科夫链蒙特卡洛方法得到θi、aj、αj、bj、dj的最优解；
S36、预测个性化学生响应
5、根据权利要求1所述的方法，其特征在于，步骤S4基于学生真实响应，利用PMF分解学生答题共性，结合学生个性化预测响应，共同预测学生对试题的响应；具体为：
S41、通过PMF对学生真实响应进行概率矩阵分解，得到学生和试题的特征矩阵Ui、Vj；
S42、预测包含学生共性的响应
S43、计算学生个性化预测响应和学生共性预测响应的加权和，如下式：
其中，Rp是预测的学生响应，μ是所有学生的均分，r′ij为预测的学生个性化响应，R′为PMF预测的学生共性响应，通过参数ρ调节学生的个性和共性之间的比例。
6、根据权利要求1所述的学生情感感知的个性化在线试题推荐方法，其特 征在于，步骤S5根据预测的学生响应，为学生提供合适难度的试题，形成最终的推荐列表；具体为：
S51、根据预测的学生响应Rp设计试题难度[d1，d2]；
试题难度由学生响应的均值计算：
其中，rij为学生i对试题j的响应，n为作答试题j的记录数；
S52、根据试题难度dj，筛选出推荐试题的列表Rp∈[1-d2,1-d1]。
本发明通过PMF挖掘学生答题共性，结合学生个性化预测响应，既提升了模型的预测能力，又保证推荐方案的全面性。
本发明在充分考虑学生个性化特征和答题共性后，根据学生对试题的掌握程度设计试题难度，提高了推荐试题的合理性。
根据本发明的另一方面，提供了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现本发明的学生情感感知的个性化在线试题推荐方法中的步骤。
根据本发明的又一方面，提供了一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现本发明的学生情感感知的个性化在线试题推荐方法中的步骤。
相比于现有技术，本发明至少具有如下有益效果：
1、本发明在PMF的基础上引入NCB-IRM，提出了一种学生情感感知的个性化在线试题推荐方法，首先利用NCB-IRM将学生情感感知的外在认知和学生内在认知结合，更全面地考虑学生的个性化需求；其次，通过PMF挖掘学生答题共性，既保证推荐方案的全面性，又提高了推荐试题的合理性；
2、在NCB-IRM中，设计了基于学生情感感知的外在认知建模方法，一方面利用PCA降低情感分析的复杂度，另一方面使用分层的自注意力方法根据不同情感权重来调整其对学生的影响，基于学生情感，为学生个性化需求的挖掘提供有效的方法。
附图说明
图1为本发明方法流程图；
图2为学生情感感知的外在认知建模结构示意图；
图3为本发明不同试题难度对应的学生答题准确率的表现；
图4为两名学生认知结果和推荐试题的案例；
4(a)为诊断的学生A和学生B的增强认知，4(b)为学生A和学生B推荐的试题以及试题和知识点的关系。
具体实施方式
本实施例中的数据集来源于在线学习平台ASSISTment，共从该平台上采集了2009年和2017年的数据，其中ASSIST2009中包含401756条学生试题练习记录，ASSIST2017中加入学生情感数据，包含942816条学生试题练习记录。每名学生作答试题时，其响应结果、情感数据都将被记录。
如图1所示，本实施例所述的一种学生情感感知的个性化在线试题推荐方法， 包括如下步骤：
S1、采集学生观看教育视频的学习行为记录，对学习行为记录进行数据清洗，提取有效信息。
学生id、试题id、知识点id、学生在试题上的情感E，学生在试题上的响应R；
E包含六个情感维度bored、concentrating、confused、frustrated、offtask、gaming；每个维度属于[0,1]，某一维情感的数值越接近1，表示学生这种情感越强；
学生i在试题j的响应用rij表示；若rij＝1表示学生i答对试题j；若rij＝0，则表示学生i答错试题j；
对数据清洗，删除包含空条目与冗余的答题行为记录，提取有效信息如下：
学生i在知识点k的潜在知识掌握，用αik表示，随机初始化αik，αik∈[0,1]；
试题矩阵Q为包含了所有试题和知识点的关系。试题q考察知识点k的情况，用qjk表示；若qjk＝1，表示试题j未考察知识点k；若qjk＝0，表示试题j未考察知识点k。
数据清洗是指删除包含空条目与冗余的答题行为记录。本实施例的两个数据集中，将清洗完的数据集按照7：2：1的比例分割为训练集、测试集、验证集，统计信息如表1。
表1数据集的统计描述
S2、基于学生情感、学生答题行为和试题矩阵，通过PCA算法模块和分层的自注意力模块建模学生外在认知。
S21利用PCA算法，将学生的复杂情感进行聚类为两类，分别表示为积极情感和消极情感，表达式为：
其中，E＝Ebored,Econcentrate,Econfused,Efrustrated,Eofftast,Egaming，·表示向量的点积运算，降维后的n_component＝2，V1,V2分别是特征向量，是情感样本在前两个主成分方向上的投影值。
S22在交互层使用两个自注意力模块模拟聚类得到的情感(以为例)与学生知识点潜在掌握情况αik以及试题考察知识点情况qjk交互；
输入积极情感时，自注意力模块将试题考察知识点序列pj作为注意力机制中的键(key)，将情感表示与学生知识点潜在掌握情况序列si分别作为注意力机制中的查询(query)与值(value)；
首先，通过余弦相似度计算学生回答试题向量pj与积极情感向量交互的相互影响权重ωp，基于试题特性为学生积极情感分配权重；然后，在考虑特征交互作用后，通过ωp与学生知识点潜在掌握情况si的加权和，得到交互后的积极情感向量op，调整积极情感对学生知识掌握的影响，表达式为：
其中pj＝(qj1,qj2,…qjk)，si＝(αi1,αi2,…αik)；
类似地，输入消极情感时，交互层通过自注意力理解学生消极情感、学生知识掌握与作答试题的交互关系，得到交互后的消极情感向量on；
S23聚合层利用另一个自注意力模块，将计算的不同交互情感op、on聚合；
具体而言，将不同类别的交互情感结果op和on分别作为注意力机制中的query和key，根据注意力权重调整情感对学生的影响；将学生的知识掌握情况作为注意力机制中的value，聚合不同的交互情感，计算基于学生特性得到融合情感特征的学生外在认知θo，表达式为：
当建模情感影响的外在认知时，学生的潜在知识掌握αik取为0.001。步骤S2可以解决学生外在认知建模机理不清晰的问题。
S3综合外在认知和学生固有认知，引入NCB-IRM框架，全面地预测学生个性化响应。已知学生响应rij，分别考虑学生情感特征影响学生响应的两种极端情况；
S31仅受到情感特征的影响，学生i正确回答了试题j，将学生学习反应函数建模：
其中，θo、αj、bj分别是学生的外在认知水平、试题区分度和试题难度；
S32未受到情感特征的影响，学生i正确回答了试题j，将学生学习反应函数建模为：
其中，rij表示学生i在试题j的响应，Aij表示学生i回答试题j表现的情感，θi、aj、dj分别是学生的内在认知水平、试题区分度和试题难度；
S33假设每个问题的响应在统计上独立于学生的认知，使用伯努利分布对所有学生的响应进行建模：
其中ηij和ζij分别代表学生i根据通过试题练习和情感特征正确回答问题j的概率；
S34学生的响应矩阵是对试题先验、学生情感先验的综合响应；给定响应rij，学生认知θi、θo，试题变量aj、αj、bj、dj，学生响应的项目反应函数如下式：
S35最大化学生响应的项目反应函数，通过一种基于Metropolis-Hastings的马尔科夫链蒙特卡洛方法得到θi,aj,αj,dj,bj的最优解：
S351首先将所有参数随机化为初始值；
S352从第一个数开始，依次遍历T个时刻的样本；
S353从θi(t)～U(θi(t-1)-δθi,θi(t-1)+δθi)中采集一个样本，并以概率接受θi；
S354从aj(t)～U(aj(t-1)-δaj,aj(t-1)+δaj)中采集一个样本，并以概率接受aj；
S355从αj(t)～U(αj(t-1)-δαj,αj(t-1)+δαj)中采集一个样本，并以概率 接受αj；
S356从dj(t)～U(dj(t-1)-δdj,dj(t-1)+δdj)中采集一个样本，并以概率接受dj；
S357从bj(t)～U(bj(t-1)-δbj,bj(t-1)+δbj)中采集一个样本，并以概率接受bj；
S358遍历完所有的T个时刻，若满足收敛条件，则输出θi,aj,αj,dj,bj的估计；
S36预测个性化学生响应
结合学生情感数据，步骤S3可以解决学生个性化认知建模不够准确的问题。
S4基于学生真实响应，利用PMF分解学生答题共性，结合学生个性化预测响应，共同预测学生对试题的响应。
S41基于学生响应R，通过PMF进行概率矩阵分解，得到学生和试题的特征矩阵Ui、Vj；
S411假设已知的响应数据和学生、试题特征向量分布满足高斯分布，定义在观测的响应上的条件分布：
其中Ui表示学生潜在特征，Vj表示试题潜在特征，N是均值为μ，方差为σ2的高斯分布概率密度函数。
S412求出学生和试题潜在特征的后验概率如下式：
S413要使得含有超参数的学生和试题潜在特征的后验概率最大，就需要最小化目标函数中含有二次正则项的平方误差和：
其中，其中，R为真实值，是正则化参数，Ui和Vj分别表示学生与试题隐特征向量，表示矩阵范数，Iij是指标函数(如果学生i作答 了试题j，该值为1；否则为0)；
S414分别对Ui,Vj求导：

S415用随机梯度下降方法更新Ui,Vj：

直到收敛或达到最大迭代次数，得到Ui,Vj的估计。
S42预测包含学生共性的学生响应
S43计算学生个性化预测响应和学生共性预测响应的加权和，如下式：
其中，Rp是预测的学生响应，μ是所有学生的均分，r′ij为预测的学生个性化响应，R′为PMF预测的学生共性响应，通过参数ρ调节学生的个性和共性之间的比例；
S431要使预测值与真实值接近，优化目标为最小化函数：
其中，Rp为预测的学生响应，R为真实值，是正则化参数，Ui和Vj分别表示D维的学生与试题隐特征向量，表示矩阵范数，Iij为指示函数，当学生i作答试题j时，Iij＝1，否则为0；
S432通过基于Metropolis-Hastings的马尔科夫链蒙特卡洛方法估计r′ij，通过最大化后验概率来估计潜在因子向量Ui、Vj，输出训练的预测响应Rp。
综合学生个性和共性，步骤S4可以解决传统推荐方案预测能力弱，推荐方 案不全面的问题。
S5根据预测的学生响应，为学生提供合适难度的试题，形成最终的推荐列表。
S51根据预测的学生响应Rp设计试题难度[d1，d2]；
试题难度由学生响应的均值计算：
其中，rij为学生i对试题j的响应，n为作答试题j的记录数；
S52根据试题难度dj，筛选出推荐试题的列表Rp∈[1-d2,1-d1]。步骤S5根据学生试题掌握情况进行推荐，可以解决试题推荐方案不合理的问题。
下面对本实施例的实验方法作进一步说明。
训练过程中，使用哈维尔(Xavier)来初始化参数。确切地说，这些参数用从N(0,std2)中采样的随机值填充权重，其中ni表示神经网络的输入维度，ni+1表示神经网络的输出维度。此外，设置epoch＝5，lr(学习率)＝0.001。所有模型都是用Python语言在PyTorch中实现的，实验的硬件配置为2.3GHz Dual-Core Intel Core i5，运行内存为8G，操作系统为macOS Big Sur11.3.1。
本实施例中，用于评价本发明提出的推荐方案的性能指标分为两类：精确度指标与非精确度指标：精确度指标包括准确率(ACC)、ROC曲线下面积(AUC)、均方根误差(RMSE)与平均绝对误差值(MAE)，目的是衡量情感感知的认知是否更贴合学生的真实掌握情况；非精确度指标包括推荐试题的合理性，即推荐的试题是否满足学生的需求，是否符合学生的认知情况。这五个指标的具体含义如下：
由于无法准确获取学生的学生认知的真实值，学生情感感知认知诊断的表现难以进行直接评估。于是根据已有工作的基础，本实施例利用学生响应来监督模型性能。具体的，根据预测结果为得分率还是答题结果(0或1)，本实施例可从回归和分类两个角度对情感感知认知诊断的性能进行了间接的评估。将问题处理为回归任务时，采用均方根误差(Root Mean Square Error，RMSE)和平均绝对误差(Mean Absolute Error，MAE)来量化预测分数和实际分数之间的距离。

其中y'表示第i次观察的观测值，y表示第i次观察的实际值，n表示总观察次数。RMSE和MAE的值越小，表示模型的预测效果越好。
将问题处理为分类任务时，预测的结果(1，0)的表示正、负实例，评价指标常采用ROC曲线下面积(Area Under Curve，AUC)和预测准确率(Accuracy， ACC)。

其中M表示正样本数，N表示负样本数，CorrectPair表示所有样本对中被正确排序的样本对；TP(真正例)表示模型正确预测为正例的样本数量，TN(真负例)表示模型正确预测为负例的样本数量，FP(假正例)表示模型错误预测为正例的样本数量，FN(假负例)表示模型错误预测为负例的样本数量。AUC和ACC的值在0到1之间，数值越接近1，表示模型的预测结果越好。
本实施例的试题推荐不是考虑向学生推荐最难/最简单的试题，而是尝试推荐符合他们认知程度的试题。因此，评价指标设定为被推荐试题的正确作答率，即正确作答试题数占推荐试题数的比例：
其中，分子表示推荐试题中学生答对的个数，分母表示推荐的试题个数。RACC越大，推荐的试题越接近学生的认知水平。
本实施例中，将CDM中的三个框架IRT、MIRT、DINA用于对比，其中与E-CDM表示使用注意力层的CDM，他们在ASSIST2009中的预测效果如表2所示。
IRT：采用一个Logistic函数来描述学生的知识掌握，它可以提供可解释的参数(例如学生潜在特征、问题区分度和难度)来分析学生的表现，是一个连续的单维CDM。
MIRT：具有Logistic项目反应函数，由单维IRT扩展而来，是一个连续的多维CDM。
DINA：引入了试题知识点矩阵Q，还关注学生的滑动和猜测行为的影响，采用离散二进制向量来表示学生的知识掌握，是一个离散的多维CDM。
CDM中，IRT中试题区分度、试题难度分别为aj＝4，dj＝0；MIRT中aj＝0，dj＝0；DINA的滑动、猜测、步长设置为maxslip＝0.4，maxguess＝0.4，maxstep＝1000。CDM的输入结构分别为：IRT(最大学生id+1，最大试题id+1)，MIRT(最大学生id+1，最大试题id+1，latent_number)，DINA(最大学生id+1，最大试题id+1，knowledge_number)。本实施例中，令latent_number＝knowledge_number，其中，在ASSIST2009数据集中，IRT的输入结构为(4164，17747)，MIRT的输入结构为(4164，17747，102)，DINA的输入结构为(4164，17747，102)；同理在ASSIST2017数据集中，IRT的输入结构为(7784，3164)，MIRT的输入结构为(7784，3164，123)，DINA的输入结构为(7784，3164，123)。
下面结合附图与表格分析实验结果。
由于ASSIST2009中没有记录情感相关的数据，首先在这个数据集中验证自 注意力层的效果。将学生答题日志作为输入，通过将pj同时作为自注意力中的query和key，将si作为自注意力中的value，调整试题对学生掌握的影响权重。
表2 ASSIST2009中学生认知预测结果
(1)ASSIST2009数据集在CDM的基础上增加的自注意力模块后，总体而言所有E-CDM的ACC较CDM有所增加，E-CDM的MAE较CDM有所下降，这表明自注意力模块可以更有效地捕捉学生与试题之间的内部相关性。
(2)从模型的结果看，E-IRT(ACC＝0.659，MAE＝0.375)在这些模型中的效果最佳，表明IRT对ASSIST2009数据集的适应性较好。
为探究情感输入是否会进一步增强模型效果，本实施例对包含情感数据的ASSIST2017数据集进行实验。在ASSIST2009的基础上增加了A-CDM和AE-CDM。其中A-CDM指在CDM的基础上，增加学生的情感特征作为模型输入，在NCB-IRM框架中进行实验；AE-CDM为实施例中提出的情感感知的认知诊断模型，在NCB-IRM的基础上使用PCA和分层的自注意力建模学生情感为外在认知。ASSIST2017中学生答题预测结果如表3所示。
表3 ASSIST2017中学生认知预测结果
(1)与仅关注学生答题日志的模型(CDM、E-CDM)相比，融合了情感特征的CDM(A-CDM、AE-CDM)在性能上有了较大的提升。在仅增加情感特征作为输入的情况下，IRT(AUC＝0.535，RMSE＝0.575)、MIRT(AUC＝0.551，RMSE＝0.633)、DINA(AUC＝0.508，RMSE＝0.572)，而A-IRT(AUC＝0.701，RMSE＝0.465)、A-MIRT(AUC＝0.608，RMSE＝0.533)、A-DINA(AUC＝0.543， RMSE＝0.554)，这表明加入情感特征对于模型的预测能力有着显著的影响。
(2)AE-CDM的较A-CDM也有一定程度上的提升。根据评价指标，与A-IRT相比，AE-IRT的AUC值增加了0.040，MAE值减少0.019；与A-MIRT相比，AE-MIRT的AUC值增加了0.034，MAE值减少0.034；与A-DINA相比，AE-DINA的AUC值增加了0.023，MAE值减少了0.017。由此可见，自注意力模块在情感特征对学生认知的调节作用。
(3)在三个CDM中，IRT的预测结果最佳且EA-IRT的提升效果最为明显。EA-IRT的AUC比IRT提升38.64％，EA-IRT的ACC比基线IRT提升45.12％。
如图3所示，本实施例探究了学生试题掌握与推荐难度的关系。
ASSIST2017数据集中，随着推荐试题的难度的增加，学生对推荐问题的RACC不断下降，表明这些模型能够根据学生的需求推荐适当难度水平的试题。根据实验结果，本发明提出的AE-PMFDINA(将本发明应用于DINA框架)的RACC能够超过PMF-DINA。本发明在引入学生情感特征后，能够更精准地建模学生的认知，更准确地预测学生的得分，向每个学生推荐所设定推荐难度范围内的试题。
本实施例还通过案例分析展示了对于两个学生的试题推荐情况。图4表示了在ASSISIT2017上学生A、学生B对于6个知识点的掌握程度。可以看出,学生A在square-root、pattern-finding等知识点上掌握较好，知识点area、probability上掌握较差；学生B在知识点area、pattern-finding上掌握较好，在知识点square-root、probability、equation-solving上掌握较差。
在测试数据集中，学生A和学生B的正确作答率均为50％，当选取难度在0.3～0.5之间的试题进行推荐时，本发明向学生A推荐的试题为推荐第4558题2115题和1624题，向学生B推荐的试题为推荐第894题、2401题和1597题。
其中，由图4(a)可知学生A对知识点area、probability的掌握较弱，图4(b)展示了部分试题-知识点的对应关系。推荐的试题4558题考查的是area知识点，2115题考查的是probability知识点。学生B对知识点square-root、probability的掌握较弱，向学生B推荐的试题为推荐第894题考查的是square-root知识点、2401题考查的是probability知识点。由于学生A对square-root的掌握程度较高，该生没有被推荐相关的试题。
从该案例分析中可看出，本发明提出的个性化试题推荐方法在推荐难度较高的试题时，可以针对每个学生的个性化学习状态向其推荐相应的试题，试题推荐结果具有很强的可解释性。
实施例2：
本实施例的计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现实施例1的学生情感感知的个性化在线试题推荐方法中的步骤。
本实施例的计算机可读存储介质可以是终端的内部存储单元，例如终端的硬盘或内存；本实施例的计算机可读存储介质也可以是所述终端的外部存储设备，例如终端上配备的插接式硬盘，智能存储卡，安全数字卡，闪存卡等；进一步地，计算机可读存储介质还可以既包括终端的内部存储单元也包括外部存储设备。
本实施例的计算机可读存储介质用于存储计算机程序以及终端所需的其他程序和数据，计算机可读存储介质还可以用于暂时地存储已经输出或者将要输出的数据。
实施例3：
本实施例的计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现实施例1的学生情感感知的个性化在线试题推荐方法中的步骤。
本实施例中，处理器可以是中央处理单元，还可以是其他通用处理器、数字信号处理器、专用集成电路、现成可编程门阵列或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等，通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等；存储器可以包括只读存储器和随机存取存储器，并向处理器提供指令和数据，存储器的一部分还可以包括非易失性随机存取存储器，例如，存储器还可以存储设备类型的信息。
本领域内的技术人员应明白，实施例公开的内容可提供为方法、系统、或计算机程序产品。因此，本方案可采用硬件实施例、软件实施例、或结合软件和硬件方面的实施例的形式。而且，本方案可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器和光学存储器等)上实施的计算机程序产品的形式。
本方案是参照根据本方案实施例的方法、和计算机程序产品的流程图和/或方框图来描述的，应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合；可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，所述的存储介质可为磁碟、光盘、只读存储记忆体(Read-OnlyMemory，ROM)或随机存储记忆体(RandomAccessMemory，RAM)等。
本发明所述实例仅仅是对本发明的优选实施方式进行描述，并非对本发明构思和范围进行限定，在不脱离本发明设计思想的前提下，本领域工程技术人员对本发明的技术方案作出的各种变形和改进，均应落入本发明的保护范围。

Claims (8)

  1. 一种学生情感感知的个性化在线试题推荐方法，其特征在于，包括如下步骤：
    S1、采集学生进行在线学习的答题行为以及相应的情感记录，进行数据清洗，提取有效信息；
    S2、基于学生情感数据，利用PCA将学生的复杂情感分类，并通过自注意力模块权重调整情感对学生的影响，建模学生外在认知；
    S3、综合学生外在认知和学生固有认知，引入NCB-IRM框架，全面地预测学生个性化响应；
    S4、利用PMF分解学生答题共性，结合学生个性化预测响应，共同预测学生对试题的响应；
    S5、根据预测的学生响应，为学生提供合适难度的试题，形成最终的推荐列表。
  2. 根据权利要求1所述的方法，其特征在于，步骤S1采集包括学生情感的在线学习的答题行为记录，提取有效信息；具体为：
    S11、采集包括学生id、试题id、知识点id、学生在试题上的情感数据E，学生在试题上的响应R；
    E包含六个情感维度bored、concentrating、confused、frustrated、offtask、gaming；每个维度的情感元素Aij属于[0,1]，越接近1，表示学生在这种情感上的表现越强；
    学生i在试题j的响应用rij表示；若rij＝1表示学生i答对试题j；若rij＝0，则表示学生i答错试题j；
    S12、对数据清洗，删除包含空条目与冗余的答题行为记录，提取有效信息如下：
    由学生行为提取学生i在知识点k的潜在知识掌握，用αik表示,αik∈[0,1]；
    试题矩阵Q为包含了所有试题和知识点的关系，元素qjk表示每个试题q考察知识点的情况；若qjk＝1，表示试题j未考察知识点k；若qjk＝0，表示试题j未考察知识点k。
  3. 根据权利要求1所述的方法，其特征在于，步骤S2基于学生情感、学生答题行为和试题矩阵，通过PCA算法模块和分层的自注意力模块建模学生外在认知，具体为：
    S21、利用PCA算法，将学生的复杂情感进行聚类为两类，分别表示为积极情感和消极情感，表达式为：
    其中，E是包含六个情感维度的矩阵，·表示向量的点积运算，是情感样本在前两个主成分方向上的投影值，V1、V2分别表示特征向量；
    S22、在交互层使用两个自注意力模块模拟聚类得到的情感，以为例，与 学生知识点潜在掌握情况αik以及试题考察知识点情况qjk交互；
    输入积极情感时，自注意力模块将试题考察知识点的序列pj作为注意力机制中的键，将情感表示与学生知识点潜在掌握情况序列sj分别作为注意力机制中的查询与值；
    首先，通过余弦相似度计算学生回答试题向量pj与积极情感向量交互的相互影响权重ωp；然后，在考虑特征交互作用后，通过ωp与学生知识点潜在掌握情况si的加权和，得到交互后的积极情感向量op，表达式为：
    其中pj＝(qj1,qj2,…qjk)，si＝(αi1,αi2,…αik)；
    类似地，输入消极情感时，交互层通过自注意力模拟学生消极情感、学生知识掌握与作答试题的交互关系，得到交互后的消极情感向量on；
    S23、聚合层利用另一个自注意力模块，将计算的不同交互情感op、on聚合；
    将不同类别的交互情感结果op和on分别作为注意力机制中的查询和键，学生的知识掌握情况作为注意力机制中的值，聚合不同的交互情感，得到融合情感特征的学生外在认知θo，表达式为：
  4. 根据权利要求1所述的方法，其特征在于，步骤S3综合外在认知和学生固有认知，引入NCB-IRM框架，全面地预测学生个性化响应；已知学生响应rij，分别考虑学生情感特征影响学生响应的两种极端情况；具体为：
    S31、仅受到情感特征的影响，学生i正确回答了试题j，将学生学习反应函数建模：
    其中，θo、αj、bj分别是学生的外在认知水平、试题区分度和试题难度；
    S32、未受到情感特征的影响，学生i正确回答了试题j，将学生学习反应函数建模为：
    其中，rij表示学生i在试题j的响应，Aij表示学生i回答试题j表现的情感，θi、aj、dj分别是学生的内在认知水平、试题区分度和试题难度；
    S33、假设每个问题的响应在统计上独立于学生的认知，使用伯努利分布对所有学生的响应进行建模：
    其中ηij和ζij分别代表学生i根据通过试题练习和情感特征正确回答问题j的概率；
    S34、学生的响应矩阵是对试题先验、学生先验的综合响应；给定响应rij，学生认知θi、θo，试题变量aj、αj、bj、dj，学生响应的项目反应函数如下式：
    S35、最大化学生响应的项目反应函数，通过一种基于Metropolis-Hastings的马尔科夫链蒙特卡洛方法得到θi、aj、αj、bj、dj的最优解；
    S36、预测个性化学生响应
  5. 根据权利要求1所述的方法，其特征在于，步骤S4基于学生真实响应，利用PMF分解学生答题共性，结合学生个性化预测响应，共同预测学生对试题的响应；具体为：
    S41、通过PMF对学生真实响应进行概率矩阵分解，得到学生和试题的特征矩阵Ui、Vj；
    S42、预测包含学生共性的响应
    S43、计算学生个性化预测响应和学生共性预测响应的加权和，如下式：
    其中，Rp是预测的学生响应，μ是所有学生的均分，r′ij为预测的学生个性化 响应，R′为PMF预测的学生共性响应，通过参数ρ调节学生的个性和共性之间的比例。
  6. 根据权利要求1所述的学生情感感知的个性化在线试题推荐方法，其特征在于，步骤S5根据预测的学生响应，为学生提供合适难度的试题，形成最终的推荐列表；具体为：
    S51、根据预测的学生响应Rp设计试题难度[d1，d2]；
    试题难度由学生响应的均值计算：
    其中，rij为学生i对试题j的响应，n为作答试题j的记录数；
    S52、根据试题难度dj，筛选出推荐试题的列表Rp∈[1-d2,1-d1]。
  7. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于：该程序被处理器执行时实现如权利要求1～6中任一项所述的学生情感感知的个性化在线试题推荐方法中的步骤。
  8. 一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其特征在于，所述处理器执行所述程序时实现如权利要求1～6中任一项所述的学生情感感知的个性化在线试题推荐方法中的步骤。
PCT/CN2023/131750 2023-10-24 2023-11-15 学生情感感知的个性化在线试题推荐方法 Pending WO2025086355A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2024571904A JP2026511279A (ja) 2023-10-24 2023-11-15 生徒の感情認識のための個人化されたオンラインテスト問題の推奨方法

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202311381649.8 2023-10-24
CN202311381649.8A CN117370665A (zh) 2023-10-24 2023-10-24 学生情感感知的个性化在线试题推荐方法

Publications (1)

Publication Number Publication Date
WO2025086355A1 true WO2025086355A1 (zh) 2025-05-01

Family

ID=89400003

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/131750 Pending WO2025086355A1 (zh) 2023-10-24 2023-11-15 学生情感感知的个性化在线试题推荐方法

Country Status (3)

Country Link
JP (1) JP2026511279A (zh)
CN (1) CN117370665A (zh)
WO (1) WO2025086355A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN121029961A (zh) * 2025-11-03 2025-11-28 湘江实验室 一种法律知识智能问答方法、设备及介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN119295275B (zh) * 2024-12-09 2025-03-25 快上云(上海)网络科技有限公司 一种智能化学习行为分析方法及系统
CN120123742A (zh) * 2025-03-10 2025-06-10 安徽大学 一种基于情感感知的认知诊断方法

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109919810A (zh) * 2019-01-22 2019-06-21 山东科技大学 在线学习系统中的学生建模与个性化课程推荐方法
KR102465768B1 (ko) * 2021-10-18 2022-11-15 주식회사 아이스크림에듀 개인화 맞춤형 학습분석 제공 시스템

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6712027B1 (ja) * 2019-12-27 2020-06-17 有限会社ソリューションゲート 学習支援システム
WO2022181104A1 (ja) * 2021-02-25 2022-09-01 日本電気株式会社 分析装置、分析方法及び非一時的なコンピュータ可読媒体

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109919810A (zh) * 2019-01-22 2019-06-21 山东科技大学 在线学习系统中的学生建模与个性化课程推荐方法
KR102465768B1 (ko) * 2021-10-18 2022-11-15 주식회사 아이스크림에듀 개인화 맞춤형 학습분석 제공 시스템

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
ABDUL ASHU, CHEN JENHUI, LIAO HUA-YUAN, CHANG SHUN-HAO: "An Emotion-Aware Personalized Music Recommendation System Using a Convolutional Neural Networks Approach", APPLIED SCIENCES, vol. 8, no. 7, 1 October 2018 (2018-10-01), Basel, pages 1 - 16, XP093309281, ISSN: 2076-3417, DOI: 10.3390/app8071103 *
DU JING, GAO BOJUN;ZHOU WEI;CAI ZHENYU: "A Comparative Study on Context Awareness Model in Adaptive Learning System ", E-EDUCATION RESEARCH, vol. 41, no. 8, 1 August 2020 (2020-08-01), JP , pages 58 - 66+81, XP093309285, ISSN: 1003-1553, DOI: 10.13811/j.cnki.eer.2020.08.008 *
LI QUAN, LIU XINGHONG;XU XINHUA;LIN SONG: "Personalized test question recommendation method based on unified probalilistic matrix factorization", JOURNAL OF COMPUTER APPLICATIONS, vol. 38, no. 3, 10 March 2018 (2018-03-10), CN , pages 639 - 642+649, XP093309276, ISSN: 1001-9081, DOI: 10.11772/j.issn.1001-9081.2017082071 *
ZHOU YUQIANG; LIU QI; WU JINZE; WANG FEI; HUANG ZHENYA; TONG WEI; XIONG HUI; CHEN ENHONG; MA JIANHUI: "Modeling Context-aware Features for Cognitive Diagnosis in Student Learning", PROCEEDINGS OF THE 45TH INTERNATIONAL ACM SIGIR CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL, 14 August 2021 (2021-08-14), New York, NY, USA, pages 2420 - 2428, XP059239102, ISBN: 978-1-4503-8732-3, DOI: 10.1145/3447548.3467264 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN121029961A (zh) * 2025-11-03 2025-11-28 湘江实验室 一种法律知识智能问答方法、设备及介质

Also Published As

Publication number Publication date
JP2026511279A (ja) 2026-04-13
CN117370665A (zh) 2024-01-09

Similar Documents

Publication Publication Date Title
Swayamdipta et al. Dataset cartography: Mapping and diagnosing datasets with training dynamics
Blauch et al. Computational insights into human perceptual expertise for familiar and unfamiliar face recognition
CN109919810B (zh) 在线学习系统中的学生建模与个性化课程推荐方法
CN118822804B (zh) 基于多元行为数据的云端教学质量评价方法及系统
JP2026511279A (ja) 生徒の感情認識のための個人化されたオンラインテスト問題の推奨方法
CN115310520A (zh) 融合多特征的深度知识追踪方法及习题推荐方法
CN118333156B (zh) 一种基于辅助任务增强的多知识点融合知识追踪方法
Mi et al. Research on constructing online learning performance prediction model combining feature selection and neural network
Chu et al. Click-based student performance prediction: A clustering guided meta-learning approach
Ramesh et al. Interpretable engagement models for MOOCs using Hinge-loss markov random fields
CN117668360A (zh) 一种基于学习者在线学习行为分析的个性化习题推荐方法
CN118608334A (zh) 一种基于教育行业大数据的教育课程推荐系统、方法和应用
CN119088681A (zh) 一种智能软件测试充分性评价方法
Chen et al. LogCF: Deep Collaborative Filtering with Process Data for Enhanced Learning Outcome Modeling.
Wu et al. Programming knowledge tracing based on heterogeneous graph representation
Ma et al. Personalized exercise recommendation via knowledge enhancement and fuzzy cognitive fusion in large-scale e-learning environments
Gad et al. Temporal graph memory networks for knowledge tracing
CN101266793A (zh) 通过对话回合间上下文关系来减少辨识错误的装置与方法
Theophilus et al. Machine learning-based e-learners’ engagement level prediction using benchmark datasets
Madras Towards trustworthy machine learning in high-stakes decision-making systems
Zhang et al. Multiple Classification Models Based Student's Phobia Prediction Study
Zhang et al. Neural attentive knowledge tracing model for student performance prediction
Shou et al. All Element Selection Method in Classroom Social Networks and Analysis of Structural Characteristics.
Deepika et al. Analyze and predicting the student academic performance using data mining tools
Rosita et al. Lecturer Evaluation from the Perspective of Speech Emotion Recognition with Deep Learning

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2024571904

Country of ref document: JP

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: 2024571904

Country of ref document: JP

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23956592

Country of ref document: EP

Kind code of ref document: A1