A Hybrid Approach to Recommend Long Tail Items

Abstract

Techniques in recommendation systems generally focuses on recommending the most important items for a user. The purpose of this work is to generate recommendations focusing on long tail items, and then to conduct the user to less popular items. However, such items are of great relevance to the user. Two techniques from the literature were applied in this study in a hybrid way. The first technique is through markov chains to calculate node similarity of a user item graph. The second technique applies clustering, where items are separated into distinct clusters: popular items (short tail) and non-popular items (long tail). Using the Movielens 100k database, we conducted an experiment to calculate the accuracy, diversity, and popularity of the recommended items. With our hybrid approach we were able to improve the recall by up to 27.97 % when compared to the markov chain-based algorithm, which indicates greater targeting to long tail products. At the same time the recommended items were more diversified and less popular, which indicates greater targeting to long tail products.