Mining scientific literature to predict new relationships

  • Authors:

  • Wei Huang;Yoshiteru Nakamori;Shouyang Wang;Tieju Ma

  • Affiliations:

  • School of Knowledge Science, Japan Advanced Institute of Science and Technology, Asahidai 1-1, Nomi, Ishikawa, 923-1292, Japan and Institute of Systems Science, Academy of Mathematics and Systems ...;School of Knowledge Science, Japan Advanced Institute of Science and Technology, Asahidai 1-1, Nomi, Ishikawa, 923-1292, Japan;Institute of Systems Science, Academy of Mathematics and Systems Sciences, Chinese Academy of Sciences, Beijing 100080, China;School of Knowledge Science, Japan Advanced Institute of Science and Technology, Asahidai 1-1, Nomi, Ishikawa, 923-1292, Japan

  • Venue:
  • Intelligent Data Analysis
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

It is compelling to process scientific literature to support the development of new science and technology. We propose a method to predict new relationships between a starting concept of interest and other concepts by mining scientific literature. In contrast to previous research, we measure the relationship between two concepts not only by their co-occurrence in scientific literature, but also by their sibling relationship in a hierarchical structure of concepts. Therefore, the predicted relationships of concepts obtained with our method are more pertinent to existing relationships within current scientific literature. By introducing a parent set, we propose a measure to evaluate the closeness of two concepts in a hierarchical structure of concepts. In order to deal with the combinatorial problems, we present two ways to limit the number of new relationships, which can be interactively enforced by the user. As in most of the previous research on literature-based discoveries, we choose biomedicine as the field in which to demonstrate our method. A comparison with related research shows that our method exhibits better performance, except in term of Recall. The new relationships predicted by this method can serve as candidates for new research themes, as impetus for inspiration, or as hypotheses to be tested in future.