Recommendation System Using Double Averaging-Based Collaborative Filtering
Recommendation System is filtering the enormous quantity of information to obtain useful information based on
the user’s preferences (ratings). Rating prediction can be formulated as a neighbourhood-based collaborative filtering
approach. It first explores the role of different item averages as the foundation of similarity weighting and then predicts the
missing ratings with the traditional neighbourhood-based method. Ensemble method is an effective tool to improve the
performance of simple neighbourhood-based prediction algorithm by iteratively applying the algorithm on the same dataset.
In all iterations, interpolation weights for all nearest neighbours are simultaneously derived by minimizing the
recommendation error. The present paper describes the AdaBoost framework with the double averaging-based collaborative
filtering for the prediction of missing ratings. Instances that are hard to predict are reinforced by updating sample weights in
the goal function that need to be minimized. The experimental evaluation demonstrates that the boosting double averaging
based collaborative filtering significantly improves the prediction performance with different levels of sparsity (25%, 30%
and 40%) on MovieLens dataset.
Keywords— AdaBoost, collaborative filtering, double averaging-based prediction, AdaBoost ensemble method,
recommendation system, weighted prediction.