Distributed Recommendation Considering Aggregation Diversity

Abstract

Recommender systems (RSs) are popular in e-commerce as they suggest different kinds of items for different users. Most existing research works focus on how to improve the accuracy of recommender systems. Recently, some recommendation ranking techniques have been proposed to obtain more diverse recommendations for all the users. In this paper, the authors propose design a distributed mechanism for improving the aggregated recommendation diversity and define three new metrics to evaluate the diversity of RSs. To avoid the disclosure of information to a central agency, a distributed mechanism is designed to collect user ratings. To increase the diversity of set recommendations, user-based and item-based weighted methods are proposed. The tasks of them are to deal with non-ratings by weighting the common ratings and calculating the weighted cosine similarities to predict the unknown ratings.