The Research Of Adaptive Learning System Based On Data Analysis

Online education continues to thrive today and it becomes the main approach to learn for modern people.MOOC(massive open online courses)represents the most popular education platform not only for students but for people who have jobs.Online education platform powered by Internet technology is capable of digitalizing learning behaviors,which facilitates the analysis of learning behaviors automatically and scalably.And it will assist teachers in teaching and give students proper feedbacks.We have analyzed desensitized data of CNMOOC and provide proper advice based on the learning behaviors which is an approach to realize adaptive learning system.However,the industrial data is much sophisticated and sparse,which makes it hard for building the model.In the paper,we introduce the approach using machine learning algorithms to predict student's proficiency.We processed the data of basic information,tests information,course video watching logs and interactions in the course forum through feature engineering.We propose some methods tackling data sparsity and insufficiency.We only focus on predicting student learning proficiency on specific course and compare different models under the scenario of inadequate data.The result shows that random forest outperforms neural network in this case and compact data improve the accuracy of proficiency estimation.We compared the effect of different types of features on students' learning efficiency,and finally got the best model with test accuracy of 65.27%.We also tried to calibrate the timing feature of test data and we introduce word embedding to process text features to further improve the model accuracy.It is found that the test data has a greater impact than the course video watching data.The introduction of the word embedding improves the accuracy of TF-IDF by about 3%.