Paper: Active Learning of Extractive Reference Summaries for Lecture Speech Summarization

ACL ID W09-3105
Title Active Learning of Extractive Reference Summaries for Lecture Speech Summarization
Venue Building and Using Comparable Corpora
Session
Year 2009
Authors

We propose using active learning for tag- ging extractive reference summary of lec- ture speech. The training process of feature-based summarization model usu- ally requires a large amount of train- ing data with high-quality reference sum- maries. Human production of such sum- maries is tedious, and since inter-labeler agreement is low, very unreliable. Ac- tivelearninghelpsassuagethisproblemby automatically selecting a small amount of unlabeled documents for humans to hand correct. Our method chooses the unla- beled documents according to the similar- ity score between the document and the comparable resource—PowerPoint slides. After manual correction, the selected doc- uments are returned to the training pool. Summarization results show an increasing learning curve of ROUGE-L F-measur...