Dynamic extended tree conditioned LSTM-based biomedical event extraction

Extracting knowledge from unstructured text has become essential to the text mining and knowledge discovery tasks in biomedical field. In this paper, we propose a novel Long Short Term Memory (LSTM) networks framework DET-BLSTM to extract biomedical events among biotope and bacteria from biomedical literature. In our framework, a dynamic extended tree is introduced as the input instead of the original sentences, which utilises the syntactic information. Furthermore, the POS and distance embeddings are added to enrich input information. In final, considering that shallow machine learning methods can effectively take advantage of the domain expert experience, the predictions of SVM are used for post-processing. Our DET-BLSTM model with post-processing achieves 58.09% F-score in the test set, which is better than all official submissions to BioNLP-ST 2016 and 2.29% higher than the best system.