QU at TREC-2014: Online Clustering with Temporal and Topical Expansion for Tweet Timeline Generation

In this work, we present our participation in the microblog track in TREC-2014, building upon our first participation last year. We present our approaches for the two tasks of this year: temporally-anchored ad-hoc search and tweet timeline generation. For the ad-hoc search task, we used topical expansion in addition to temporal models to perform retrieval. Our results show that our run based on the typical pseudo relevance feedback query expansion outperformed all of our other runs with a relatively high mean average precision (MAP). As for the timeline generation task, we approached this problem using online incremental clustering of tweets retrieved for a given query. Our approach allows the dynamic creation of “semantic” clusters while providing a framework for detecting redundant tweets and selecting representative ones to be added to the final timeline. The results demonstrate that using incremental clustering of tweets retrieved through a temporal retrieval model produced the best effectiveness among the submitted runs.