Improving accuracy of recommender systems through triadic closure

Tselenti, Panagiota (2017) Improving accuracy of recommender systems through triadic closure. (PhD thesis), Kingston University, .

Full text available as:
[img]
Preview
Text
Tselenti-P.pdf - Submitted Version
Available under License Creative Commons Attribution.

Download (3MB) | Preview

Abstract

The exponential growth of social media services led to the information overload problem which information filtering and recommender systems deal by exploiting various techniques. One popular technique for making recommendations is based on trust statements between users in a social network. Yet explicit trust statements are usually very sparse leading to the need for expanding the trust networks by inferring new trust relationships. Existing methods exploit the propagation property of trust to expand the existing trust networks; however, their performance is strongly affected by the density of the trust network. Nevertheless, the utilisation of existing trust networks can model the users’ relationships, enabling the inference of new connections. The current study advances the existing methods and techniques on developing a trust-based recommender system proposing a novel method to infer trust relationships and to achieve a fully-expanded trust network. In other words, the current study proposes a novel, effective and efficient approach to deal with the information overload by expanding existing trust networks so as to increase accuracy in recommendation systems. More specifically, this study proposes a novel method to infer trust relationships, called TriadicClosure. The method is based on the homophily phenomenon of social networks and, more specifically, on the triadic closure mechanism, which is a fundamental mechanism of link formation in social networks via which communities emerge naturally, especially when the network is very sparse. Additionally, a method called JaccardCoefficient is proposed to calculate the trust weight of the inferred relationships based on the Jaccard Cofficient similarity measure. Both the proposed methods exploit structural information of the trust graph to infer and calculate the trust value. Experimental results on real-world datasets demonstrate that the TriadicClosure method outperforms the existing state-of-the-art methods by substantially improving prediction accuracy and coverage of recommendations. Moreover, the method improves the performance of the examined state-of-the-art methods in terms of accuracy and coverage when combined with them. On the other hand, the JaccardCoefficient method for calculating the weight of the inferred trust relationships did not produce stable results, with the majority showing negative impact on the performance, for both accuracy and coverage.

Item Type: Thesis (PhD)
Physical Location: This item is held in stock at Kingston University library.
Research Area: Computer science and informatics
Faculty, School or Research Centre: Faculty of Science, Engineering and Computing (until 2017)
Depositing User: Jennifer May
Date Deposited: 21 Dec 2017 15:00
Last Modified: 06 Nov 2018 10:17
URI: http://eprints.kingston.ac.uk/id/eprint/39871

Actions (Repository Editors)

Item Control Page Item Control Page