Incremental clustering of search history in personalized search

Long-term search history containing rich information about individual preference plays an important role in personalized search. Although various methods have proposed to explore long-term search history, most are assuming that the whole long-term history is available and every piece of history is helpful to the current search. This paper takes into account the individual history as streaming on-line sources and presents a personalized search model that can be updated by means of incremental clustering algorithm as new data arrive. Two strategies for selecting related history cluster are investigated, i.e. picking the latest cluster to represent user recent interests, and choosing the cluster with highest content relevance to current query. Tested on a large-scale search log of a commercial search engine, this approach outperforms the baseline model, especially when a user submits a fresh query with a new search need that has not occurred before.