Biomedical Event Trigger Detection Based on BiLSTM Integrating Attention Mechanism and Sentence Vector

As the crucial and prerequisite step in biomedical event extraction, trigger detection has attracted much attention. Most of the existing trigger detection methods either rely on elaborately designed features or consider features only within a window. Another challenge is that the existing methods treat each word in sentence equally. Also, most methods ignore the sentence-level semantic information. Therefore, we propose a trigger detection method based on Bidirectional Long Short Term Memory (BiLSTM) neural network, which can skip manual complex feature extraction. Furthermore, to obtain more semantic and syntactic information, we train dependency-based word embeddings to represent words, and add sentence vector to enrich sentence-level features. Finally, we integrate attention mechanism to capture the most important semantic information in a sentence. The experimental results on the multi-level event extraction (MLEE) corpus show that the proposed method outperforms the state-of-the-art systems.

I. Introduction

With the development of Internet, the resources are expanding at an exponential speed, especially in the biomedical domain, which has made it harder than ever for scientists to research and extract knowledge from the vast unstructured biomedical scientific literature. Therefore, biomedical information extraction techniques are developing rapidly.