Related Experiment Video
Updated: Jul 15, 2025
08:45
Mapping Cortical Dynamics Using Simultaneous MEG/EEG and Anatomically-constrained Minimum-norm Estimates: an Auditory Attention Example
Published on: October 24, 2012
14.7K
End-to-end clinical temporal information extraction with multi-head attention
Timothy Miller1, Steven Bethard2, Dmitriy Dligach3
1Computational Health Informatics Program, Boston Children's Hospital, Harvard Medical School.
Proceedings of the Conference. Association for Computational Linguistics. Meeting
|October 2, 2023
Summary
This study introduces a novel multi-headed attention mechanism for temporal relation extraction in clinical text. The system achieves state-of-the-art results on the THYME corpus, improving clinical applications.
Area of Science:
- Clinical Informatics
- Natural Language Processing
- Artificial Intelligence
Background:
- Temporal relationship extraction from electronic health records (EHRs) is crucial for clinical applications.
- Limited progress has been made in end-to-end temporal relation extraction systems since Clinical TempEval 2017.
- Existing methods often rely on gold-standard annotations for events and time expressions.
Purpose of the Study:
- To develop an advanced end-to-end system for temporal relation extraction from clinical text.
- To improve the accuracy and applicability of temporal information in EHRs.
- To address the limitations of previous approaches by not requiring gold-standard annotations.
Main Methods:
- Utilized a novel multi-headed attention mechanism.
- Integrated this mechanism with a pre-trained transformer encoder.
- Enabled the model to attend to multiple facets of contextualized embeddings for enhanced learning.
Main Results:
- Achieved state-of-the-art performance on the THYME corpus.
- Demonstrated significant improvements in both in-domain and cross-domain settings.
- Outperformed previous methods by a considerable margin.
Conclusions:
- The proposed multi-headed attention mechanism significantly enhances temporal relation extraction.
- The developed system offers a robust solution for analyzing temporal dynamics in clinical text.
- This advancement holds promise for numerous downstream clinical applications.
