PKUICST at TREC 2014 Microblog Track: Feature Extraction for Effective Microblog Search and Adaptive Clustering Algorithms for TTG
Abstract
This paper describes our approaches to temporally-anchored ad hoc retrieval task and tweet timeline generation (TTG) task in the TREC 2014 Microblog track. In the ad hoc search, we apply a learning to rank framework which utilizes not only the various content relevance of a tweet, but also the quality of a tweet. External evidences are well incorporated in our approach with Web-based query expansion and document expansion techniques. In the TTG task, we apply star clustering and hierarchical clustering algorithm on the retrieved tweets from ad hoc retrieval task. Experimental results show that our learning to rank methods with many state-of-the-art features achieve good retrieval performance with respect to MAP and P@30 metrics. Besides, our systems for TTG task also obtain convincing recall and precision scores.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 2014
- Accession Number
- ADA624060
Entities
People
- Chao Lv
- Feifan Fan
- Jianwu Yang
- Runwei Qiang
- Yue Fei
Organizations
- Peking University