A New Similarity Measure Based on Gravitational Attraction for Improving the Accuracy of Collaborative Recommendations
Vijay Kumar Verma, Rajesh Kumar Aggarwal · International Journal of Intelligent Systems and Applications · 2020
Recommender Systems (RSs) work as a personal agent for individuals who are not able to make decisions from the potentially overwhelming number of alternatives available on the World Wide Web (or simply Web).Neighborhood-based algorithms are traditional approaches for collaborative recommendations and are very popular due to their simplicity and efficiency.Neighborhood-based recommender systems use numerous kinds of similarity measures between users or items in order to achieve diverse goals for designing an RS such as accuracy, novelty, diversity etc.However, the existing similarity measures cannot manage well the data sparsity problems, which results in either very few corated items or absolutely no co-rated items.Furthermore, there are also situations where only the associations between users and items, such as buying/browsing behaviors, exist in form of unary ratings, a special case of ratings.In such situations, the existing similarity measures are either undefined or provide extreme values such as either 0 or 1.Thus, there is a compelling need to define a similarity measure that can deal with data sparsity problem and/or unary rating data.This article proposes a new similarity measure for neighborhoodbased collaborative recommender systems based on Newton's law of universal gravitation.In order to achieve this, a new way of interpreting the relative mass as well as the relative distance has been taken into consideration by using the rating data from the user-item matrix.Finally, for evaluating the proposed approach against baseline approaches, several experiments have been conducted using standardized benchmark datasets such as MovieLens-100K and MovieLens-1M.Results obtained demonstrate that the proposed method provides better predictive accuracy in terms of RMSE and significantly improves the classification accuracy in terms of precision-recall.