Exploring External Knowledge Base for Personalized Search in Collaborative Tagging Systems.

Alongside the enormous volume of user-generated content posted to World Wide Web, there exists a thriving demand for search personalization services, especially those utilizing collaborative tagging data. To provide personalized services, a user model is usually required. We address the setting adopted by the majority of previous work, where a user model consists solely of the user's past information. We construct an augmented user model from a number of tags and documents. These resources are further processed according to the user's past information by exploring external knowledge base. A novel generative model is proposed for user model generation. This model leverages recent advances in neural language models such as Word Embeddings with latent semantic models such as Latent Dirichlet Allocation. We further present a new query expansion method to facilitate the desired personalized retrieval. Experiments conducted by utilizing real-world collaborative tagging data show that the methods proposed in the current paper outperform several non-personalized methods as well as existing personalized search methods by utilizing user models solely constructed from usage histories.