A new extraction algorithm for hierarchical keyword using text social network

Abstract

Keywords are the most basic and efficient means for document classification and information retrieval. Moreover, keywords connote the most abridged information about the document to readers; thus, they are the most important feature for understanding a document. As a starting point for understanding documents, we propose a keyword extraction algorithm based on graphs, and reveal the results in a manner that is easy to understand visually. Using the location of each word in the document, we estimate the influence of words and quantify the relationships of dominant words by computing the degree of overlap between each influence. On the basis of this relationship information, we analyze the elements of graph theory using a maximum spanning forest (MSF) graph; using this analysis, we can visualize the relationships between each pair of words and extract the core keywords. As a result of this extraction process, we confirmed that this method improves upon the performance of extraction methods based only on word frequency.