Clustering the information space using top-ranking sentences: a study of user interaction

By considering sentences selected by a query-biased sentence extraction model from the top-retrieved documents, we create a personalised information space which is characterised by the presence of search terms. We cluster this information space, and enable searchers to interact with the resulting clusters. In order to examine whether users can recognise, and benefit from, the clustered organisation, we compare user interaction and performance between an actual clustering and a pseudo-clustering of the information space for completing information seeking tasks. The results provide evidence for the utility and meaningfulness of the clustered organisation