Personalized recommendation based on desktop context

Abstract

Aimed to the inherent detects in present information retrieval service, this paper proposed an approach to exploit desktop context to provide personalized recommendation service. The files restored on the local disk and the documents opened in a work scenario were regarded as two separate parts serving for personalizing. The algorithm for extracting to desktop resources to build the long-term document model was introduced in detail, which further provides information to build a user's interest model. And the method to establish the short-term model in a work scenario to predict the user's current information need was also introduced. The experiments were conducted to offer recommended items in a message window and analyzed the implicit information of each user's corresponding behaviors. The results showed that users were interested in recommended items and the performance was stable. © 2012 Springer-Verlag GmbH.