Where academic tradition
meets the exciting future

You are here: TUCS > PUBLICATIONS > Publication Search > Hypothesis Generation in Large...

Hypothesis Generation in Large-Scale Event Networks

Kai Hakala, Farrokh Mehryary, Suwisa Kaewphan, Filip Ginter, Hypothesis Generation in Large-Scale Event Networks. In: Fabio Rinaldi, Kim Jin-Dong (Eds.), Proceedings of the 5th International Symposium on Language in Biology and Medicine (LBM 2013), 19–27, Database Center for Life Science, 2013.

Abstract:

Hypothesis generation from literature is
among the most prominent goals of the
BioNLP research community. The existence of EVEX, a large-scale event network mined from the entire available biomedical literature, opens the possibility to cast this task in a supervised machine learning setting, defining it as the prediction of edges in this network, based on features from their network context.

In this paper, we study the task from two
perspectives. First, we build a machine
learning system which predicts novel pairwise relationships in the EVEX network
and evaluate its performance using both
the standard measures as well as through
a manual inspection on a subset of the output. And second, we analyze and discuss
the issues in evaluation arising from cross-
validation in densely connected graphs
with uneven edge distribution.

We find that the task is learnable, achieving performance clearly above baseline.
Further, a manual inspection of predictions not found in the EVEX network showed several candidate pairs, whose interaction could be verified in the literature. These pairs hint at the possibility that true novel interacting pairs were identified by the system as well, even though further work is necessary to confirm whether that is indeed the case.

Files:

Full publication in PDF-format

BibTeX entry:

@INPROCEEDINGS{inpHaMeKaGi13a,
  title = {Hypothesis Generation in Large-Scale Event Networks},
  booktitle = {Proceedings of the 5th International Symposium on Language in Biology and Medicine (LBM 2013)},
  author = {Hakala, Kai and Mehryary, Farrokh and Kaewphan, Suwisa and Ginter, Filip},
  editor = {Rinaldi, Fabio and Jin-Dong, Kim},
  publisher = {Database Center for Life Science},
  pages = {19–27},
  year = {2013},
}

Belongs to TUCS Research Unit(s): Turku BioNLP Group

Edit publication