Deriving ratings through social network structures

A review of existing approaches to recommendation in e-commerce systems is provided. A recommendation system is primarily used to identify services which may be of interest to a user based on a similarity in purchasing (or browsing) patterns with another user, or to filter services that have been returned as a result of a search. Existing systems primarily make use of collaborative filtering approaches or a semantic-annotation approach which tries to find similarity by matching on the definition of a service. However, such systems suffer from "sparseness" of ratings - as it is difficult to find enough ratings to help make a recommendation for a user. We therefore propose the use of a social network as the basis for defining how ratings can be aggregated, based on the structure of the network. We also suggest the use of product categories as the basis for aggregating ratings - and define this as a "context" in which a particular service is used. A model for a recommendation system that combines context-based rating with the structure of a social network has been suggested, along with an architecture for a system that implements the model