Combining Multiple Layers of Syntactic Information for Protein-Protein Interaction Extraction

Abstract

Protein-protein interaction extraction is a challenging information extraction task in the BioNLP field. Several kernels focusing on a part of syntactic information have been proposed for the task. In this paper, we pro-pose a method to combine multiple layers of syntactic information by using a combi-nation of multiple kernels based on several different parsers. We evaluated the method using support vector machine and achieved an F-score of 62.0 % on the AIMed corpus. Further, we analyzed the performance with or without including self-interaction pairs, and found that there is a danger of con-fusing classifiers and decreasing the perfor-mance when treating self-interaction pairs together with real pairs. 1