A Negative Similarity Based Hybrid Recommender System Using Apache Spark

Abstract

The rapid technological advances in digital business after the millennium has exponentially boosted the options of available items to the users. Customized lists of recommendations assist people by limiting the number of options to a very few items that are related to the specific user and thus supporting them to have the final decision with ease. In this study, we used Apache Spark to show an effective parallel implementation of our hybrid recommendation algorithm. Our hybrid algorithm is based on Alternating Least Squares and Negative Similarity Collaborative Filtering methods. Alternating Least Squares is a type of algorithm for matrix factorization, which can be parallelized during runtime. It performs relatively well when processing large scale datasets. Negative Similarity Collaborative Filtering (NSCF) technique exploits the tastes of people at the opposite poles. The cascade hybridization method was used in the study to create a hybrid algorithm. In the evaluations, we observed that hybridization achieved significant accuracy performance improvement when compared to the performance of NSCF algorithm alone. Moreover, our hybrid method performed slightly better than the ALS algorithm alone in the overall accuracy of recommendations.