An Improved Graph-based Recommender System for Finding Novel Recommendations among Relevant Items
Ranran Liu, Zhengping Jin · 2015
Recommender system has been extensively studied to provide the most relevant data to users in this era of information explosion.Among all kinds of recommendation algorithms, collaborative filtering (CF) algorithm is one of the most famous ones because of its high accuracy and simple implementation.Recently, scholars have proposed a new approach to find fresh and novel items, but the relevance of some novel items may be far from good which reduced system's precision accordingly.In this paper, we propose an improved recommender system to increase the relevance when finding out novel items.This approach is motivated by the fact that social relationships could reflect the similar interests between users in a recommender system.Thus, social relationship is taken into consideration when we build the profile graph of each user.We test the system on Last.fm data and the result shows that the improved graph-based recommender system could indeed provide fresh recommendations while the accuracy have increased by 0.7% on average at the same time.