Burst analysis of text document for concept map creation

Abstract

Concept maps are graphical representation showing the relationships among ideas, images, or words. It is proven that automatically generated concept maps help people understanding a document before reading. The network of concept map consists of nodes which represent concepts and links which represent relation between concepts. When automatically creating concept maps, a traditional approach captures the association relation between concepts using co-occurrence of words. This study is motivated from the assumption that the relation between two words are not fully captured using the co-occurrence method. Instead, the relation can be inferred from the occurrence pattern of words. To observe the occurrence pattern of words, the burst analysis of words is adopted in this study. There is often an intensive coverage of some topic within a certain period which we refer to as a bursty interval of a word. Instead of co-occurrence method, this study captures the relation between concepts by pairing two words using bursty intervals. The case study shows that the proposed method outperforms the co-occurrence method in illustrating flow of a story in a document. Since the co-occurrence method can still provide additional information about a document, the combination of the proposed method and co-occurrence method is suitable for illustrating a summary of a document.