Browse Teach Curate

Full Dependency-Graph GCN Token Embeddings for Biomedical Event Extraction

Students learn to represent each token in a sentence by running a graph convolutional network over the complete adjacency matrix of the sentence's dependency parse, instead of collapsing the syntax to the shortest dependency path between a trigger and its argument. This preserves off-path words (such as negations) that can decide argument classification, and shows how syntactic graph structure supplies signal for structured prediction tasks like biomedical event extraction.

R2-D2 Text

Attending To Syntactic Information In Biomedical Event Extraction Via Graph Neural Networks Farshad

This paper improves biomedical event extraction (BEE) by embedding each sentence token with a graph convolutional network that operates over the full adjacency matrix of the sentence's dependency par…