Recently learning to rank has been widely used in real-time Twitter Search by integrating various of evidence of relevance and recency features into together. In real-time Twitter search, whereby the information need of a user is represented by a query at a specific time, users are interested in fresh messages. In this paper, we introduce a new ranking strategy to rerank the tweets by incorporating multiple features. Besides, an empirical study of learning to rank for real-time Twitter search is conducted by adopting the state-of-the-art learning to rank approaches. Experiments on the standard TREC Tweetsll collection show that both the listwise and pairwise learning to rank methods outperform baselines, namely the content-based retrieval models.
展开▼
