Dynamic Weighted Hybrid Recommender Systems

Abstract

Recommender Systems (RSs) have emerged since the mid-90s for dealing with the problem of information overload. They are commonly defined as software tools and techniques that serve as an assistant providing suggestions to the users. The two most familiar recommendation techniques are probably Collaborative filtering (CF) and Content-based filtering (CBF). Whereas CF computes recommendations based on past ratings of users with similar preferences, CBF assumes that each user operates independently, thus exploits only information derived from item features. Technically speaking, the performance of every single recommendation algorithm is limited and each has its own strengths and weaknesses, so recently more attentions are paid to the hybrid recommendation algorithms. In this work, we focus on the weighted hybridization and rather than using fixed weighted for the combination, we aim to propose a simple method to dynamic weight a combination of CF and CBF. Our experimental results on one of the most popular public datasets in the field of RSs - MovieLens have verified the effectiveness of our strategy against the traditional CF, and CBF approaches. It not only boost the prediction performance, but also alleviate the problem of new item cold start.