Cascaded filtering for topic-driven multi-document summarization
Katja Filippova, Margot Mieskes, Vivi Năstase, Simone Paolo Ponzetto, Michael Strube · MADOC (University of Mannheim) · 2007
This paper presents EMLR's NLP group's rst participation in the DUC summarization com- petitions. Our system combines document l- tering, ranking sentences using lexical chains and graph matching algorithms with the topic, on top of several annotation layers in the MMAX2 annotation tool. The system ranked 14 out of 30 participating teams in manual annotation, and had particularly good ranking from the linguistic quality point of view.