Understanding MOA framework to Evaluate Recommender Systems

In the recent past recommender systems have become very popular with wide variety of e-commerce based applications. Recommendations are used to discover items, the users might not have found by themselves. In an online scenario of a recommender system, the new ratings from users to items arrive constantly. The system has to make predictions of unrated items. The system in turn has to decide which items to recommend for the user’s search. For example recommending new electronic product range based on the five star rating of a company. Such recommendations are based on either customer interests or the recommendations made on their past history of purchases or product searches. The paper aims at understanding and evaluating the recommender systems by developing a model for recommendation systems using both massive online analysis framework and R programming environment. Evaluation is carried out mainly by predicting the user responses or ratings or preferences. The two important parameters considered for prediction are rating predictor and a data set. The present work uses publicly available movielens data set of varying sizes. A comparative study of results obtained from massive online analysis framework and R programming environment is presented and results are found to be correct using movie lense data set.