QCRI at TREC 2013 Microblog Track

We report our work in the real-time ad hoc search task of TREC-2013 Microblog track. Our system focuses on improving retrieval effectiveness of Microblog search through query expansion and reranking of search results. We apply web-based query expansion algorithm for enriching the microblog queries with additional terms from concurrent webpages related to the search topic. Later we apply results reranking through utilizing state-of-the-art learning to rank algorithms to train 12 different ranking models using relevance judgment of Tweets2011-12 queries, for which we conduct feature engineering, validation dataset selection, and the ensemble of these models. Our approach differs from salient approaches in the previous Microblog tracks that are based on document expansion utilizing embedded URLs and that leverage some single ranking model for tweets re-ranking. Our pipeline constructed using the hybrid of these two components showed promising retrieval results on Tweets2013 benchmark dataset