AbstractAs customers enjoy the convenience of online shopping today, they face the problem of selecting from hundreds of thousands of products. Recommender systems, which make recommendations by matching products to customers based on the features of the products and the purchasing history of customers, are increasingly being incorporated into e-commerce websites. Collaborative filtering is a major approach to design algorithms for these systems. Much research has been directed toward enhancing the performance of recommender systems by considering various psychological and behavioural factors affecting the behaviour of users, e.g. trust and emotion.
While e-commerce firms are keen to exploit information on social trust available on social networks to improve their services, conventional trust-aware collaborative filtering does not consider the multi-facets of social trust. In this research, we assume that a consumer tends to trust different people for recommendations on different types of product. For example, a user trusts a certain reviewer on popular items but may not place as much trust on the same reviewer on unpopular items. Furthermore, this thesis postulates that if we, as online shoppers, choose to establish trust on an individual while we ourselves are reviewing certain products, we value this individual’s opinions on these products and we most likely will value his/her opinions on similar products in future.
Based on the above assumptions, this thesis proposes a new collaborative filtering algorithm for deriving multi-faceted trust based on trust establishment time. Experimental results based on historical data from Epinions show that the new algorithm can perform better in terms of accuracy when compared with conventional algorithms.
|Date of Award||2016|
|Supervisor||Wing Lok YEUNG (Supervisor)|