Single and Multiple Document Summarization with Graph-based Ranking Algorithms

Опубликовано 6 сентября 2016, 5:04

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.