Dependency Graphs for Summarization and Keyphrase Extraction: We present a real-time long document summarization and key-phrase extraction algorithm that utilizes a unified dependency graph.

Abstract

We introduce a graph-based summarization and keyphrase extraction system that uses dependency trees as inputs for building a document graph. The document graph is built by connecting nodes containing lemmas and sentence identifiers after redirecting dependency links to emphasize semantically important entities. After applying a ranking algorithm to the document graph, we extract the highest ranked sentences as the summary. At the same time, the highest ranked lemmas are aggregated into keyphrases using their context in the dependency graph. Our algorithm specializes in handling long documents, including scientific, technical, legal, and medical documents.