Exploring Trust-Aware Neighbourhood in Trust-based Recommendation

TL;DR

This paper introduces a trust-aware neighborhood algorithm that integrates social trust signals into recommender systems, significantly improving coverage and accuracy, especially for cold-start users, by leveraging social network data.

Contribution

It proposes a novel trust-aware neighborhood method that enhances collaborative filtering with social trust information, addressing cold-start issues and improving recommendation performance.

Findings

Doubling of rating coverage compared to traditional CF

Significant accuracy improvements on real-world datasets

Near-complete coverage for cold-start users with trust networks

Abstract

Traditional Recommender Systems (RS) do not consider any personal user information beyond rating history. Such information, on the other hand, is widely available on social networking sites (Facebook, Twitter). As a result, social networks have recently been used in recommendation systems. In this paper, we propose an efficient method for incorporating social signals into the recommendation process by building a trust network which supplements the users' rating profiles. We first show the effect of different cold-start users types on the Collaborative Filtering (CF) technique in several real-world datasets. Later, we propose a "Trust-Aware Neighbourhood" algorithm which addresses a performance issue of the former by limiting the trusted neighbourhood. We show the doubling of the rating coverage compared to the traditional CF technique, and a significant improvement in the accuracy for some datasets. Focusing specifically on cold-start users, we propose a "Hybrid Trust-Aware Neighbourhood" algorithm which expands the neighbourhood by considering both trust and rating history of the users. We show a near complete coverage with a rich trust network dataset-- Flixster. We conclude by discussing the potential implementation of this algorithm in a budget-constrained cloud environment.