Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Farshad Noravesh, Reza Haffari, Ong Huey Fang, Layki Soon, Sailaja Rajalana, Arghya Pal

TL;DR

Biomedical event extraction (BEE) is posed as a structured prediction task where nested events require modeling complex trigger-argument relationships. The authors propose a hybrid architecture combining BioBERT-based contextual embeddings with a graph neural network (GCN) operating on the full dependency parsing graph, augmented by dedicated head/tail networks and MLP-based distance encoding to mitigate oversmoothing. They present two training variants (with and without dependency graphs) and conduct extensive ablation studies on Genia11, Genia13, and BioNLP13 pc/cg datasets, demonstrating that leveraging the full dependency graph yields performance gains over baselines and SDP-only approaches. The work highlights the importance of syntactic structure in BEE, offering a robust framework for handling nested events with improved expressivity and targeted architectural design to counter GNN-related issues.

Abstract

Many models are proposed in the literature on biomedical event extraction(BEE). Some of them use the shortest dependency path(SDP) information to represent the argument classification task. There is an issue with this representation since even missing one word from the dependency parsing graph may totally change the final prediction. To this end, the full adjacency matrix of the dependency graph is used to embed individual tokens using a graph convolutional network(GCN). An ablation study is also done to show the effect of the dependency graph on the overall performance. The results show a significant improvement when dependency graph information is used. The proposed model slightly outperforms state-of-the-art models on BEE over different datasets.

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

TL;DR

Biomedical event extraction (BEE) is posed as a structured prediction task where nested events require modeling complex trigger-argument relationships. The authors propose a hybrid architecture combining BioBERT-based contextual embeddings with a graph neural network (GCN) operating on the full dependency parsing graph, augmented by dedicated head/tail networks and MLP-based distance encoding to mitigate oversmoothing. They present two training variants (with and without dependency graphs) and conduct extensive ablation studies on Genia11, Genia13, and BioNLP13 pc/cg datasets, demonstrating that leveraging the full dependency graph yields performance gains over baselines and SDP-only approaches. The work highlights the importance of syntactic structure in BEE, offering a robust framework for handling nested events with improved expressivity and targeted architectural design to counter GNN-related issues.

Abstract

Many models are proposed in the literature on biomedical event extraction(BEE). Some of them use the shortest dependency path(SDP) information to represent the argument classification task. There is an issue with this representation since even missing one word from the dependency parsing graph may totally change the final prediction. To this end, the full adjacency matrix of the dependency graph is used to embed individual tokens using a graph convolutional network(GCN). An ablation study is also done to show the effect of the dependency graph on the overall performance. The results show a significant improvement when dependency graph information is used. The proposed model slightly outperforms state-of-the-art models on BEE over different datasets.
Paper Structure (12 sections, 3 equations, 6 figures, 4 tables, 2 algorithms)

This paper contains 12 sections, 3 equations, 6 figures, 4 tables, 2 algorithms.

Table of Contents

  1. introduction
  2. Modeling
  3. Data Preprocessing
  4. BioBertBEE And GNNBEE
  5. Entity Or Trigger Classification
  6. Argument Role Classification
  7. Tackling Oversmoothing With MLP
  8. Training
  9. Experiments
  10. Evaluation
  11. Ablation Study
  12. Conclusion

Figures (6)

  • Figure 1: An example of pathway curation sample from BioNLP13-pc
  • Figure 2: an example of a dependency parse from BioNLP2011
  • Figure 3: The architecture of tree LSTMRNN model proposed in MakotoMiwa2016
  • Figure 4: taken from YuhaoZhang2018 that shows SDP between two entities in the bold.
  • Figure 5: architecture of bioBert based biomedical event extraction model
  • ...and 1 more figures