Student Response Behavior Modeling Methods And Application Research Based On Cognitive Diagnosis

With the rapid development of education informatization,the booming intelligent educational systems have accumulated massive data of teaching and learning.Thus a crucial task is mining valuable information from the educational data with data analysis&mining techniques as well as educational domain knowledge.The task aims to reveal natural laws of education and guides teachers and students to improve teaching and learning.And as such,educational data mining,as a new research field,becomes an important implementation technique of intelligent education and plays an increasingly dominant role in education informatization.Educational data mining(EDM)is to utilize the massive accumulated information and build an available educational database,which serves to apply proven data mining techniques into exploring valuable information based on educational theories.EDM has significant implication on guiding learning,teaching,testing and professional educat-ing.In the cycle of teaching,learning and testing,the effectiveness of student learning improvement lies on the comprehensive and accurate understanding of the students.Therefore how to obtain the knowledge proficiency of each student becomes one of the most vital research problem and technical difficulty.That is,we need to model student to comprehensively analyze students' characteristics like learning ability and cognitive level.This procedure is called Cognitive Diagnosis in educational psychology,and the corresponding models of student knowledge is named as Cognitive Diagnoisis Mod-els.This thesis is based on cognitive diagnosis,focuses on student and investigates the following three aspects.Firstly,to jointly model the objective and subjective problems,we propose a fuzzy cognitive diagnosis model based student modeling framework.Educational assessment systems design test items related to specific knowledge components which can mea-sure the knowledge proficiency of each student.Traditional methods build cognitive diagnosis models based on students' responses to objective problems.However,most existing methods are unable to exploit information from students' response of subjec-tive problems,such as open mathematical problem and English writing.Different from the objective problem whose score lies on the input answer,the scoring manner of a subjective problem depends on both the detail of solving process and the final answer.Besides.traditional methods tend to diagnose student with qualitative and binary profi-ciency of specific knowledge skill instead of more accurate quantitative analysis.To this end,considering the fuzziness of responses to subjective problems,we propose a fuzzy cognitive diagnosis model based student modeling framework,which can jointly model both objective and subjective problems with quantitative analysis of students' knowl-edge proficiency.To be specific,inspired by the fuzzy system,we represent students'knowledge proficiency into membership of fuzzy set.Then two fuzzy logic operations,fuzzy intersection and fuzzy union,are adopted to model two educational hypotheses conjunctive and compensatory for responses to objective and subjective problems,re-spectively.At last,we simulate the generation of student response according to the two different distribution of responses to the two types of problems.Secondly,to deal with the outliers of student response,we develop a slip&guess detection method based on cognitive diagnosis for off-line learning scenario.In the off-line educational assessment,the evaluation for students' knowledge proficiency cannot avoid two outliers:slip&guess.Traditional methods treat slip&guess as fixed or problem-related parameters rather than detect slip&guess factor of a specific response.It is still largely under-explored in educational data mining that detecting slip&guess from students' response from a cognitive diagnosis based respective.To this end,we develop a slip&guess detection method based on cognitive diagnosis by combining outlier detection and Bayesian posterior inference.Specifically,along with traditional outlier detection methods,we investigate the data characteristics and cognitive level of slip&guess response.Then we propose a p-value based methods for discovering students of abnormal cognitive level,which can help to estimate slip&guess factor.On the other side,we construct a Bayesian net for students5 responses and compute the posterior probability of slip&guess of each response based on pre-trained cognitive diagnosis results.Furthermore,we also figure out a hybrid method which can fuse the outlier detection and Bayesian posterior strategies.Last,to recognize the motivation of student response,we devise a cognitive di-agnosis based gaming-the-system behavior modeling method.One of the key research questions in intelligent tutoring platforms is to mine the motivation of student response.For example,gaming-the-system means some students exploit the defects of learning system instead of learned knowledge to complete tests,which badly threatens the eval-uation quality of educational assessment system.Nevertheless,the existing methods rarely handle the gaming behavior based on cognitive diagnosis.Traditional educa-tional psychology models treat gaming as guess while data mining methods adopt a supervised fashion to train a classifier via feature engineering.To address this issue,we devise a cognitive diagnosis based gaming-the-system behavior modeling method.Considering gaming factor cognitively,we adopt some data mining methods to measure gaming factor from different aspects of multi-attempt response data.An unsupervised method is proposed to aggregate these measures.Then,we utilize collaborative filtering algorithm to infer gaming factor of one-attempt response data.Last,inspired by a signal detection model,we adapt the classical item response theory into a bi-variate student response model,which can jointly model students' knowledge and gaming factor.