Abstract
Due to the exponential growth of biomedical literature, event and relation extraction are important tasks in biomedical text mining. Most work only focus on relation extraction, and detect a single entity pair mention on a short span of text, which is not ideal due to long sentences that appear in biomedical contexts. We propose an approach to both relation and event extraction, for simultaneously predicting relationships between all mention pairs in a text. We also perform an empirical study to discuss different network setups for this purpose. The best performing model includes a set of multi-head attentions and convolutions, an adaptation of the transformer architecture, which offers self-attention the ability to strengthen dependencies among related elements, and models the interaction between features extracted by multiple attention heads. Experiment results demonstrate that our approach outperforms the state of the art on a set of benchmark biomedical corpora including BioNLP 2009, 2011, 2013 and BioCreative 2017 shared tasks.
Highlights
Event and relation extraction has become a key research topic in natural language processing with a variety of practical applications especially in the biomedical domain, where these methods are widely used to extract information from massive document sets, such as scientific literature and patient records
In addition to evaluating our main model (4MHA-4CNN), we have evaluated the performance of three variants of our proposed approach: (i) 4MHA: 4 parallel multi-head attentions apply self-attention multiple times over the input features; (ii) 1MHA: only 1 multi-head attention applies self-attention to the input features; (iii) 4CNN-4MHA: multiple self-attentions are applied to the input features via a set of 1D convolutions4
We have proposed a novel architecture based on multi-head attention and convolutions, which deals with the long dependencies typical of biomedical literature
Summary
Event and relation extraction has become a key research topic in natural language processing with a variety of practical applications especially in the biomedical domain, where these methods are widely used to extract information from massive document sets, such as scientific literature and patient records. This information contains the interactions between named entities such as proteinprotein, drug-drug, chemical-disease, and more complex events. RNNs are difficult to parallelize while they do not fully solve the long dependency problem (Verga et al, 2018) Such approaches are proposed for relation extraction, but not to extract nested events. We evaluate our model on data from the shared tasks for BioNLP 2009, 2011 and 2013, and BioCreative 2017
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
