ECNU at TREC 2015: Microblog Track

This paper describes our participation in TREC 2015 Microblog track, which includes two tasks related to Scenario A and Scenario B. For Scenario A, we build a real-time tweet push system, which is mainly composed by three parts: feature extraction, relevance prediction and redundancy detection. Only the highly relevant and nonredundant tweets are sent to users based on the interest profiles. For Scenario B, we apply three query expansion methods, namely the web search based, the TFIDF-PRF based and the Terrier embedded PRF based. In addition, three state-of-the-art information retrieval models as the language model, BM25 model and DFRee model are utilized. The retrieval results are combined for final delivery. The experimental results in both scenarios demonstrate that our system obtains convincing performance.