|
Nowadays, techniques for automatically extracting viewpoints from electronic documents become crucial. In this paper, we propose a novel algorithm called KeyFlow to extract viewpoints from free text documents. KeyFlow algorithm builds a weighted directed graph for each document, where nodes represent words or phrases of high frequency in the document, and edges represent dependency relations between them. Our algorithm extends the KeyGraph algorithm for automatic keyword extraction in two steps: Firstly, we introduce semantic dependency relations among words so that viewpoint instead of individual keywords can be extracted. Secondly, we extract subgraphs as snippets of information that represent the viewpoint of document. Experiments show our algorithm can achieve higher performance for viewpoint extraction.
|
