Event

ClusterRank: A Graph-Based Method for Meeting Summarization

Nikhil Garg

ICSI

Tuesday, January 13, 2009
12:30

In this talk, I will present an unsupervised, graph based approach for extractive summarization of meeting transcripts. Graph based methods such as TextRank have been used for sentence extraction in structured text like news articles. Text was modeled as a graph with sentences as nodes and edges based on word overlap. A sentence node was then ranked according to its similarity with the rest of the nodes. The spontaneous speech in meetings leads to incomplete, ill-formed sentences and high redundancy which calls for additional measures to extract relevant sentences. I will describe an extension of the TextRank algorithm that segments the meeting transcript into clusters and uses these clusters to construct the graph. The evaluation is on the AMI meeting corpus and the results show a significant improvement over TextRank and some other baseline methods.