Optimizing recommender system for e learning using machine learning techniques

Abstract

E-learning has taken a giant leap due to the rise of digitalization in the past two decades. The e-learning courses are available for learners of all age-groups. European countries have implemented OpenEd e-learning platform. India has implemented SWAYAM platform for providing free and quality e-learning resource to students. National Digital Library of India has more than five crore e-learning resources in the form of e-books, videos, pdf etc. The increase in number of digital resources is a proof of large scale acceptance of e-learning methodology. But to search learner specific course material from the plethora of e-learning resources is a tedious and time consuming task. To solve the above said issue, recommender systems have been created to assist the learners. It is relevant to mention here that in the last two decades, Natural Language Processing (NLP) has also made considerable progress. The NLP based sentiment analysis is also used to know users sentiment about certain events or items. The use of sentiment analysis with recommender system can give more relevant recommendations. The prior knowledge of learner s inclination about a particular course using sentiment score can be very helpful in recommending the course that matches the learner s profile. The better recommendations will lead to higher course completion percentage. A content based recommender system using K-L divergence technique with sentiment score has been proposed. The course recommendations have been generated on two types of data. In Case I, the recommendations have been generated using simulated data. In case II, the recommendations have been generated on real data based on survey. Case I: The course recommendations have been generated using K-L Divergence with sentiment score technique by using 10000 simulated data. The best recommendation score was 65.8% using existing Cosine Similarity technique with sentiment score. However, the recommendations have been improved to 82.5% by using K-L Divergence with sentiment score. Case II: The