Effects of Binary Similarity Metrics on Energy Consumption During Recommendation by Using Collaborative Filtering Approach

Abstract

The rapid growth of online shopping and web-based services has increased the importance of recommender systems in supporting users' decision-making processes. Among recommendation techniques, collaborative filtering is widely used to predict an active user's preference for a target item or to generate a top-N list of items based on the preferences of users with similar tastes. Since similarity measurement is a core component of collaborative filtering, the choice of similarity metric can significantly influence recommendation accuracy, computational efficiency, and, indirectly, energy consumption. In this context, this study examines the effects of eleven similarity metrics on the performance of a collaborative filtering-based recommender system. A naïve Bayes algorithm was employed for prediction, while the F-measure was used to evaluate recommendation accuracy. In addition, the computational time required by each similarity metric was analyzed to assess efficiency. The empirical results indicate that the Yule similarity metric achieved the highest overall performance, whereas Dice, Hamann, and Kulczynski demonstrated the weakest performance. These findings suggest that the selection of an appropriate similarity metric can enhance recommender system accuracy, reduce computational time, and contribute to lower energy consumption and environmental impact.