Cognitive Diagnosis And Deep Learning Based Personalized Quiz Recommendation In K12 Education

With the rapid development of the network and advent of the era of big data,traditional industry in K12 education has gradually transitioned to digital education system.More and more resources in K12 education are integrated into the internet,which can be easy research by students.Questions in kinds of quizzes for each subject are one of the most important resources in the field of K12 education,However,massive educational resources also greatly influence the efficiency of student to obtain useful information.Therefore,it is imminent to research the recommended approach to recommend suitable quizzes for students,which can help to practice themselves and consolidate the knowledge they learn in the school more effectively.The traditional quiz recommendation approach implemented by collaborative filtering is sim-ilar to product recommendation.It generate the recommendation quiz from those students with high similarity,resulting in simple quiz and popular quiz having a higher probability of being recommended.It can not recommendation according to the learning situation of students so as can not give strong interpretable recommendations.Meanwhile the cognitive diagnosis,which used in the field of K12 education,can diagnose the learning situation to find the weakness part of students,but it can not extract the commonality of students and its parameters estimation is sensitive to the collected data set which may have much deviation.And also The traditional K12 education quiz recommendation method ignores the influence of the question content text.This thesis proposed three new approaches in a progressive way.The first approach proposed to overcome those shortcomings existed in traditional recommendation approaches,which combines the cognitive diagnosis and question content information and model-based collaborative filtering to generate a recommendation quiz.However the first approach can not extract high order commonality of students and cannot to deal with the question information with long content text.Therefore the second approach proposed to overcome those shortcoming existed in previous one,it uses the deep factorization machine and long memory recurrent network network to extract the high order relationships of students' commonality and the information from long question content.However the second approach neglects the high order factors such as the general intelligence in cognitive diagnosis,and it may not fully use of the estimated learning situation by just simply treats it as the input feature.Therefore the third approach proposed to overcome those shortcoming existed in previous one.based on previous approach,the third approach uses high order cognitive diagnosis model to diagnose the learning situation for each student under the consideration of high order factors such as general intelligence.And it uses attention mechanism to extract important prior from the estimated learning situation for better utilizing.All of the there approaches can endow recommendations with highly interpretability,students' commonality and question content information.and have lots of improvement compared to traditional approaches.