In this paper we aim at filtering documents containing timely relevant information about an entity (e.g., a person, a place, an organization) from a document stream. These documents that we call vital documents provide relevant and fresh information about the entity. The approach we propose leverages the temporal information reflected by the temporal expressions in the document in order to infer its vitality. Experiments carried out on the 2013 TREC Knowledge Base Acceleration (KBA) collection show the effectiveness of our approach compared to state-of-the-art ones.
展开▼
