Biomedical event trigger detection with convolutional highway neural network and extreme learning machine

Abstract

Detecting biomedical events in text plays a critical role in building natural language processing applications, such as in medical search, disease prevention, and pharmacovigilance. Since an event trigger can signify the occurrence of the event, the detection of biomedical event triggers is a critical step in biomedical event extraction. Current methods usually extract rich features and then feed these features to a classifier. To enhance both automatic feature selection and classification, this paper presented an end-to-end convolutional highway neural network and extreme learning machine (CHNN–ELM) framework to detect biomedical event triggers. This structure has two stages. In the first stage, CHNN is used to efficiently select higher level semantic features based on four different dimensions: embedding, convolutional layer, pooling layer, and highway layer. In the second stage, the proposed model leverages ELM, which has great scalability and generalization performance, to identify various types of biomedical event triggers. Extensive experiments are conducted on the Multi-Level Event Extraction (MLEE) dataset. To the best of our knowledge, this paper is the first to introduce ELM into this task. The results demonstrated that with better feature selection and classification, our approach outperforms several current state-of-the-art methods.