Research On Learner Response Matrix Completion Based On Ensemble Learning

Educational Data Mining(EDM) is a cross topic of computer science, pedagogy and psychology. It analyzes the learner's feedback data in the intelligent learning system, comprehends the learner's knowledge of the status and the learning content of the knowledge points. On the basis of the Internet and educational big data put forward by the state, EDM will play a more important role in the information construction, and realize the goal of teaching the Internet education according to the individuality.However, in practical applications, the feedback data often occurs in the case of insufficient learner feedback. This paper mainly studies the problem of learners'feedback Matrix Completion(MC). This problem has profound theoretical and practical application prospects. On the one hand, the student reaction matrix is a natural low rank matrix, and its research is helpful to further strengthen the low rank matrix restoration theory understanding and in-depth, another response MC for personalized teaching also has a very practical significance.The work of this paper can be divided into two parts. In the first part, we use the method of integrated learning to improve the classical matrix complement method, and build two novel MC algorithm based on the Ensemble Learning(EL) of Bagging and AdaBoost, namely BaggingMC and AdaBoostMC algorithm. The second part combines the advantages and disadvantages of BaggingMC and AdaBoostMC and The Improved AdaBoostMC algorithm is proposed to solve the problem that BaggingMC due to the MC of the simple vote make the error remains high and AdaBoostMC randomly selected threshold lead to the waste caused by the resource.In this paper, the experiment is carried out on the simulated data and the real data respectively. The error distribution of the different data sets is completed by analyzing the matrix to judge the accuracy and completeness of the different algorithms. The experimental results show that BaggingMC is close to the three classical matrix completion algorithms, and the error of AdaBoostMC is relatively small. Improved AdaBoostMC has the least error under the same data set and the same acquisition rate.At the same time through the binary Lena diagram can be intuitively seen in the same data set, the higher the capture rate of the picture through the matrix to complete the recovery after the better.