A Content boosted hybrid recommendatıon system

Çapraz, Seval
Nowadays, most of e-commerce and social media sites use recommendation systems to help users find more relevant products easily. The key feature of recommendation is personalization which means different products are being offered for different users according to each user s interests. In literature, there are a lot of algorithms and tools which implement recommendation systems. The most common techniques for recommendation systems include Collaborative Filtering (CF) and Content-Based Filtering (CBF). To increase efficiency and accuracy, these methods can be combined in a hybrid recommendation system. Apache Mahout is one of the tools which focuses primarily on algorithms in the areas of CF, clustering and classification. In this study, we used Apache Mahout for blending item-based and user-based methods of CF with switching approach. The Pearson Correlation Similarity and Nearest N-User Algorithm is used in user-based CF, while Tanimoto Coefficient Similarity and Generic Boolean Preference is used in item-based CF. Moreover,we added genre-based average ratings as content-based filtering so that the final recommendation list becomes more relevant to user. The proposed hybrid algorithm is tested on MovieLens dataset and validated with k-fold cross validation. This new hybrid recommendation system that is used to find patterns in data and develop a model for the purpose of making accurate and efficient recommender systems is proposed and detailed in this thesis study.