点击次数：

论文类型：期刊论文

发表时间：2016-01-01

发表刊物：INTERNATIONAL JOURNAL OF DATA MINING AND BIOINFORMATICS

收录刊物：SCIE、Scopus

卷号：15

期号：3

页面范围：195-213

ISSN号：1748-5673

关键字：biomedical event extraction; trigger detection; convolutional neural network; word representation

摘要：Event trigger detection, which plays a key role in biomedical event extraction, has attracted significant attention recently. However, most approaches are based on statistical models, much relying on complex hand-designed features. In this paper, we utilise the ability of Convolutional Neural Network (CNN) for addressing higher-level features automatically to explore correlations between a trigger and an event type. We only keep one candidate trigger along with N-words around it and entity mention features as a raw input, giving up complex input with hand-designed features that derived from currently existed Natural Language Processing (NLP) tools. Our experiments on Multi-Level Event Extraction (MLEE) corpus showed that the method achieved a higher F-score of 78.67% compared to the state-of-the-art approaches. The results demonstrate that the proposed method is effective for event trigger detection.