Abstract:
Extracting chemical-protein relations between chemicals and proteins plays a key role in various biomedical tasks, such as drug discovery, precision medicine, as well as ...Show MoreMetadata
Abstract:
Extracting chemical-protein relations between chemicals and proteins plays a key role in various biomedical tasks, such as drug discovery, precision medicine, as well as clinical research. Most popular methods for the chemical-protein interaction (CHEMPROT) task are based on neural networks to avoid the complex hand-crafted features derived from linguistic analyses. However, their performances are usually limited due to long and complicated sentences. Therefore, we propose a novel hierarchical recurrent convolutional neural network (Hierarchical RCNN)-based approach to efficiently learn latent features from short context subsequences. The experimental results on the CHEMPROT corpus show that our method achieves an F-score of 65.56%, and outperforms the state-of-the-art systems.
Date of Conference: 03-06 December 2018
Date Added to IEEE Xplore: 24 January 2019
ISBN Information: