Apriori Association Rule for Course Recommender system

Abstract

Until recently, recommender systems have been applied in learning, such as to recommend appropriate courses. They are based on users’ ratings, learning history, or curriculum that provide relationship between courses. The last approach, however, can’t be applied to Massive Open Online Courses (MOOCs) that don’t maintain such information. Hence, course recommender systems for MOOCs must be based on other learners’ experience. This paper discusses such recommender systems. We apply Apriori Association Rule and the case study used in this study is the Canvas Network dataset and the HarvardX-MITx dataset. The proposed recommender system consists of a pre-processing to normalize data and reduce anomalous data, data cleaning to handle empty data, K-Modes clustering to group users, grouping registration transactions for filtering user registration transaction, and finally, rule formation using the Apriori Association Rule. The performance of the association rules obtained, a lift ratio evaluation metric is used. The experiments results show the best parameters in this study are 0.01 for minimum support and 0.6 for minimum confidence. With these two parameters, the number of rules and the average lift ratio value on the Canvas Network dataset are 110 rules and 19.055, while the HarvardX-MITx dataset is 48 rules and 3.662.