JMIR Publications

ABSTRACT

Background:

Extracting events is essential in natural language processing. In the biomedical field, the nested event phenomenon (event A is a participating role of event B) makes extracting this event more difficult than extracting a single event. Therefore, the performance of nested biomedical events is always underwhelming. In addition, previous works rely on a pipeline to build an event extraction model, which ignores the dependence between trigger recognition and event argument detection tasks and produces significant cascading errors.

Objective:

We aim to design a unified framework to train biomedical event triggers and arguments jointly, and improve the performance of extracting nested biomedical events.

Methods:

We proposed an end-to-end joint extraction model that considers the probability distribution of triggers to alleviate the cascading errors. Moreover, we integrate the syntactic structure into an attention-based gate GCN to capture potential interrelations between triggers and related entities, which improves the performance of extracting nested biomedical events.

Results:

The experimental results demonstrate that our proposed method achieves the best F1-score on the MLEE biomedical event extraction corpus and achieves a favorable performance on the BioNLP-ST 2011 GE corpus.

Conclusions:

Our CPJE model is good at extracting nested biomedical events because of the joint extraction mechanism and the syntax graph structure. Moreover, because our model does not rely on external knowledge and specific feature engineering, it has a particular generalization performance.