A new similarity coefficient for a collaborative filtering algorithm
SİNCAN Özge Mercanoğlu YILDIRIM · Communications Faculty of Sciences University of Ankara Series A2-A3 Physical Sciences and Engineering · 2017
Recommender systems give theopportunity to present automatically personalized content across many digitalmarketing channels to visitors depending on visitor movements on the site. Inrecent years, there has been a lot of interest in e-commerce companies in orderto offer personalized content. So, recommender systems become very popular andmany studies have been done in this regard. New works are being done day by dayto improve the results. In this paper, we propose a new memory-basedcollaborative filtering algorithm. Calculation of similarities between items orusers is a critical step in memory-based CF algorithms. Therefore, we proposeda new function for calculation of similarities based on user ratings. In thisstudy the more similar the user's pleasures are, the more similar it is to theproducts the users choose, is adopted. The adopted idea in this study is thatthe more similar the user's pleasures are, the more similar products arechosen. We estimate the degree which a user is interested in X product. To dothis, we find other users who are interested in product X and calculate thesimilarity ratios of those users to the user. We tested our algorithm inMovieLens 100K dataset and compared to other similarity functions. We used MAEand RMSE measures in our experiments.