TrustRank: A Cold-Start tolerant recommender system

Haitao Zou, Zhiguo Gong, Nan Zhang, Wei Zhao, Jingzhi Guo

Research output: Contribution to journalArticlepeer-review

16 Scopus citations

Abstract

The explosive growth of the World Wide Web leads to the fast advancing development of e-commerce techniques. Recommender systems, which use personalised information filtering techniques to generate a set of items suitable to a given user, have received considerable attention. User- and item-based algorithms are two popular techniques for the design of recommender systems. These two algorithms are known to have Cold-Start problems, i.e., they are unable to effectively handle Cold-Start users who have an extremely limited number of purchase records. In this paper, we develop TrustRank, a novel recommender system which handles the Cold-Start problem by leveraging the user-trust networks which are commonly available for e-commerce applications. A user-trust network is formed by friendships or trust relationships that users specify among them. While it is straightforward to conjecture that a user-trust network is helpful for improving the accuracy of recommendations, a key challenge for using user-trust network to facilitate Cold-Start users is that these users also tend to have a very limited number of trust relationships. To address this challenge, we propose a pre-processing propagation of the Cold-Start users’ trust network. In particular, by applying the personalised PageRank algorithm, we expand the friends of a given user to include others with similar purchase records to his/her original friends. To make this propagation algorithm scalable to a large amount of users, as required by real-world recommender systems, we devise an iterative computation algorithm of the original personalised TrustRank which can incrementally compute trust vectors for Cold-Start users. We conduct extensive experiments to demonstrate the consistently improvement provided by our proposed algorithm over the existing recommender algorithms on the accuracy of recommendations for Cold-Start users.

Original languageEnglish (US)
Pages (from-to)117-138
Number of pages22
JournalEnterprise Information Systems
Volume9
Issue number2
DOIs
StatePublished - Feb 17 2015

All Science Journal Classification (ASJC) codes

  • Computer Science Applications
  • Information Systems and Management

Fingerprint

Dive into the research topics of 'TrustRank: A Cold-Start tolerant recommender system'. Together they form a unique fingerprint.

Cite this