Social context-aware trust inference for trust enhancement in social network based recommendations on service providers

In Service-Oriented Computing environments, there is a large number of service providers providing a variety of services to service customers. Conventional recommender systems, which adopt the information filtering techniques, can be used to automatically generate recommendations of service providers to service customers who are also the system users. However, data sparsity and trust enhancement are the traditional problems in recommender systems. Targeting the data sparsity problem, recent studies on recommender systems have started to leverage information from online social networks to collect recommendations from more participants and derive the final recommendation. However, this requires the methods to infer the trust between participants without any direct interactions in online social networks, which should take into account both the social context of participants and the context of the target services to be recommended, for trust enhanced recommendations. In this paper, we first present a contextual social network model that takes into account both participants' personal characteristics (referred to as the independent social context, including preference and expertise in domains) and mutual relations (referred to as the dependent social context, including the trust, social intimacy, and interaction context between two participants). In addition, we propose a new probabilistic approach, SocialTrust, as the first solution in the literature, to social context-aware trust inference in social networks. The result delivered by this approach is particularly important in evaluating the trust from a source participant to an end recommender who recommends a target service or service provider, via the sub-network consisting of intermediate participants/recommenders between them and relevant contextual information. Moreover, we propose algorithms that consider cycles and information updates in social networks. Experiments demonstrate that our approach is effective and superior to existing trust inference methods, and can deliver more reasonable and trustworthy results. The proposed algorithms considering cycles and information updates in social networks are efficient and applicable to real social networks.