Inferring Undiscovered Public Knowledge by Using Text Mining-driven Graph Model

Abstract

Due to the recent development of Information Technology, the number of publications is increasing exponentially. In response to the increasing number of publications, there has been a sharp surge in the demand for replacing the existing manual text data processing by an automatic text data processing. Swanson proposed ABC model [1] on the top of text mining as a part of literature-based knowledge discovery for finding new possible biomedical hypotheses about three decades ago. The following clinical scholars proved the effectiveness of the possible hypotheses found by ABC model [2]. Such effectiveness let scholars try various literature-based knowledge discovery approaches [3, 4, 5]. However, their trials are not fully automated but hybrids of automatic and manual processes. The manual process requires the intervention of experts. In addition, their trials consider a single perspective. Even trials involving network theory have difficulties in mal-understanding the entire network structure of the relationships among concepts and the systematic interpretation on the structure [6, 7]. Thus, this study proposes a novel approach to discover various relationships by extending the intermediate concept B to a multi-leveled concept. By applying a graph-based path finding method based on co-occurrence and the relational entities among concepts, we attempt to systematically analyze and investigate the relationships between two concepts of a source node and a target node in the total paths. For the analysis of our study, we set our baseline as the result of Swanson [8]'s work. This work suggested the intermediate concept or terms between Raynaud's disease and fish oils as blood viscosity, platelet aggregability, and vasconstriction. We compared our results of intermediate concepts with these intermediate concepts of Swanson's. This study provides distinct perspectives for literature-based discovery by not only discovering the meaningful relationship among concepts in biomedical literature through graph-based path interference but also being able to generate feasible new hypotheses.