Serendipity Identification Using Distance-Based Approach

Abstract

The recommendation system is a method for helping consumers to find products that fit their preferences. However, recommendations that are merely based on user preference are no longer satisfactory. Consumers expect recommendations that are novel, unexpected, and relevant. It requires the development of a serendipity recommendation system that matches the serendipity data character. However, there are still debates among researchers about the available common definition of serendipity. Therefore, our study proposes a work to identify serendipity data's character by directly using serendipity data ground truth from the famous Movielens dataset. The serendipity data identification is based on a distance-based approach using collaborative filtering and k-means clustering algorithms. Collaborative filtering is used to calculate the similarity value between data, while k-means is used to cluster the collaborative filtering data. The resulting clusters are used to determine the position of the serendipity cluster. The result of this study shows that the average distance between the recommended movie cluster and the serendipity movie cluster is 0.85 units, which is neither the closest cluster nor the farthest cluster from the recommended movie cluster.