Journals Proceedings

Abstract

The main strengths of k-means clustering, the most widely used clustering technique for recommender systems, is its simplicity and ease of applicability to practical problems. However, k-means clustering suffers from the drawbacks of falling in local optima and the quality of clusters is largely dependent on the initial cluster centers. An important contribution of this paper is a hybrid k-means clustering approach to recommender systems that combines \'outside the box\' recommendation ability of collaborative filtering with kmeans clustering and Genetic Algorithms. In this approach, genetic algorithm operators are used to pick up appropriate initial seeds for k-means clustering. This helps in improving cluster quality, thereby suggesting a new approach to recommender systems. The model focuses on identifying a set of users with similar liking for movies and accordingly making recommendations.