Two-way collaborative filtering on semantically enhanced movie ratings

Abstract

A key step in recommendation systems is to estimate if a user would likely enjoy an item who has not considered yet. In this study, a new framework is defined to predict user ratings on new items from previously given ratings by other users. The systems has two major steps: (1) Enhancing available data based on semantic content to get a full item-user matrix, and (2) Predicting the unknown rating using an integrated feature set of “other ratings given by the same user” and “other ratings given to the same item”. This allows the classifier to consider both user similarities and item similarities simultaneously. The system is shown to outperform existing methods in terms of prediction accuracy on a benchmark movie dataset.