Frequency-sensitive Diversification in Collaborative Filtering

Abstract

Due to the rapid development of data analysis technologies, recommender systems become indispensible in online markets and results in the shrinking of traditional offline competitors. As a representative leader, collaborative filtering has been widely adopted in online markets like Amazon.com, and, unlike content-based method, which suggests unexperienced items preferred by similar neighbors of a target customer. While collaborative filtering has been developed towards the recommendation precision improvement in the past, recently the diversity of recommendation becomes another important criteria for evaluating its performance. As related works, the dissimilarity measurement between recommended items in properties, and longtail item recommendation have been suggested. For the diversity improvement, this paper argues frequency-sensitive recommendation method, which aims to extend the global diversity of items rather than local diversity suggested by the past studies. As well, it introduces a comprehensive evaluation on both the diversity and the precision of recommendation, and shows the experimental results of comparing with existing methods using the actual Movielens dataset.