Character based String Kernels for Bio-Entity Relation Detection

Extracting bio-entity relations has emerged as an important task due to the ever-growing number of bio-medical documents. In this paper, we present a simple and novel representation for extracting bio-entity relationships. The state-of-theart systems for such tasks rely on word based representations and variations of linguistic driven features. In contrast, we model bio-text by the most basic character based string representation with a family of string kernels. This eliminates time consuming parsing, issue of rare words and domain specific pre-processing. This simple representation makes our approach fast and flexible for any bio-NLP dataset. We demonstrate comparable performance and faster computation time of our approach versus previous state-of-the-art kernel methods.