BIOSMILE: A semantic role labeling system for biomedical verbs
Author Information
Author(s): Tsai Richard Tzong-Han, Chou Wen-Chi, Su Ying-Shan, Lin Yu-Chun, Sung Cheng-Lung, Dai Hong-Jie, Yeh Irene Tzu-Hsuan, Ku Wei, Sung Ting-Yi, Hsu Wen-Lian
Primary Institution: Institute of Information Science, Academia Sinica
Hypothesis
Can a maximum entropy model with automatically generated template features improve semantic role labeling for biomedical verbs?
Conclusion
The study demonstrates that using a biomedical proposition bank significantly enhances the performance of semantic role labeling systems in the biomedical domain.
Supporting Evidence
- Using BioProp improved the F-score of the SRL system by 21.45% over a newswire-trained system.
- Adding automatically generated template features further improved the F-score by 0.52%.
- Statistically significant performance improvements were observed for specific argument types.
Takeaway
The researchers created a system that helps computers understand biomedical texts better by identifying important words and their roles in sentences.
Methodology
The study involved training a semantic role labeling system on a biomedical proposition bank and evaluating its performance against a newswire-trained system.
Limitations
The effectiveness of named entity features was limited in the biomedical domain, and the system's performance may vary with different biomedical verbs.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website