A New Document Representation Using a Unified Graph to Document Similarity Search

Abstract

Document similarity search is to retrieve a ranked list of similar documents and find documents similar to a query document in a text corpus or a web page on the web. But most of the previous researches regarding searching for similar documents are focused on classifying documents based on the contents of documents. To solve this problem, we propose a novel retrieval approach based on undirected graphs to represent each document in corpus. In addition, this study also considers unified graph in conjunction with multiple graphs to improve the quality of searching for similar documents. Experimental results on the Reuters-21578 data demonstrate that the proposed system has better performance and success than the traditional approach.