Single and Multiple Document Summarization with Graph-based Ranking Algorithms

September 10, 2004
Rada Mihalcea | University of North Texas

Graph-based ranking algorithms have been traditionally and successfully used in citation analysis, social networks, and the analysis of the link-structure of the World Wide Web. In short, these algorithms provide a way of deciding on the importance of a vertex within a graph, by taking into account global information recursively computed from the entire graph, rather than relying only on local vertex-specific information.

In this talk, I will present an innovative unsupervised method for extractive summarization using graph-based ranking algorithms. I will describe several ranking algorithms, and show how they can be successfully applied to the task of automatic sentence extraction. The method was evaluated in the context of both a single and multiple document summarization task, with results showing improvement over previously developed state-of-the-art systems.

I will also outline a number of other NLP applications that can be addressed with graph-based ranking algorithms, including word sense disambiguation, domain classification, and keyphrase extraction.

Speaker Details

Rada Mihalcea is an Assistant Professor of Computer Science at University of North Texas. Her research interests are in lexical semantics, minimally supervised natural language learning, and multilingual natural language processing. She is currently involved in a number of research projects, including word sense disambiguation, shallow semantic parsing, (non-traditional) methods for building annotated corpora with volunteer contributions over the Web, word alignment for language pairs with scarce resources, and graph-based ranking algorithms for language processing. Her research is supported by NSF and the state of Texas.