Please use this identifier to cite or link to this item: http://bura.brunel.ac.uk/handle/2438/15001
Title: B2B network analysis and communities-based recommender systems for the B2B electronic marketplace
Authors: Mohamad, Abdul Hadi
Advisors: Wang, F
Shepperd, M
Keywords: B2B recommender system;Communities-based recommender system;Social network analysis for B2B e-marketplace and B2B users;Recommender system with heterogeneous inputs
Issue Date: 2016
Publisher: Brunel University London
Abstract: Nowadays the amount of information available to human beings is extremely huge. With the evolvement of the Internet and mobile devices, users can access information from almost anywhere at any time. This enables great convenience to users, but at the same time causes a severe problem of information overload. That is, it becomes more and more difficult for online users to choose the correct information to suit their own needs. These days Recommender Systems (RSs) are often used to address this information overload problem by automatically providing pertinent information of users’ interests. However, the current RSs including content-based, collaborative filtering and social network-based recommendations are normally constrained by several limitations, such as overspecialised recommendations, cold-start, data sparsity and scalability problems. Moreover, even though RSs have been widely applied to B2C (Business-to-Customer) and C2C (Customer-to- Customer) E-Marketplaces such as EBay and Amazon, they have rarely been used to help business users in the B2B (Business-to-Business) E-Marketplace. In fact, there is a lack of studies on the fundamental structure of B2B, which makes recommendations for it even harder to achieve. This thesis presents a comprehensive study on B2B network analysis, and based on it, a novel approach of a communities-based recommender system is proposed. The social network analysis of these B2B networks has shown that they are small-world networks with a power-law degree distribution and short average path lengths between users. A clear modular structure has been found in B2B networks, suggesting business people tend to work more closely with others in their own business circles or communities. The mixed degree-degree correlations indicate that active sellers in B2B are associated more with small buyers than frequent buyers so there is a need to introduce or recommend good sellers and their products to buyers so as to improve business transactions. Based on the social relationships and communities discovered from the B2B network analysis, a communities-based recommender system is developed to provide quality recommendations for B2B users. The communities-based recommender system exploits the associations of business people in the same or neighbouring communities to find qualifying products, even though some of these business people may not have direct connections with the requesters. Various forms of the communities-based RSs have been designed and implemented by using different search strategies on communities and users. The experimental results on real data from EBay have demonstrated that communities-based RSs have outperformed non-communities recommenders such as random search and social network-based recommenders. In particular, the social network-based recommender has exhibited the worst performance because the search could not go beyond the limited friend circles. In contrast, communities-based RSs have successfully found solutions for all queries with high precision, recall and search efficiency. The findings presented in this thesis provide useful information about the underlying structures of the B2B networks and constructive mechanisms on how to make quality recommendations for B2B users by utilising business communities to overcome the data sparsity problem when prior information (e.g., users’ purchase history) is limited.
Description: This thesis was submitted for the award of Doctor of Philosophy and was awarded by Brunel University London
URI: http://bura.brunel.ac.uk/handle/2438/15001
Appears in Collections:Computer Science
Dept of Computer Science Theses

Files in This Item:
File Description SizeFormat 
FulltextThesis.pdfFile embargoed until 01/08/202036.44 MBAdobe PDFView/Open    Request a copy


Items in BURA are protected by copyright, with all rights reserved, unless otherwise indicated.