Evaluating Biomedical Named Entity Recognition
Author Information
Author(s): Tsai Richard Tzong-Han, Wu Shih-Hung, Chou Wen-Chi, Lin Yu-Chun, He Ding, Hsiang Jieh, Sung Ting-Yi, Hsu Wen-Lian
Primary Institution: Institute of Information Science, Academia Sinica
Hypothesis
How do different matching criteria affect the performance evaluation of biomedical named entity recognition systems?
Conclusion
The study found that right match and left match can serve as effective alternatives to traditional matching criteria in biomedical named entity recognition.
Supporting Evidence
- The study compared the performance of systems using different matching criteria.
- Right match and left match were found to be effective alternatives to exact match.
- The relaxed-match criterion improved the evaluation of Bio-NER systems.
Takeaway
This study looks at how to better identify names of genes and proteins in medical texts by using different ways to check if the names are correct.
Methodology
The study conducted four experiments comparing different matching criteria on datasets from JNLPBA and BioCreAtIvE.
Potential Biases
The reliance on specific matching criteria may introduce bias in evaluating system performance.
Limitations
The evaluation may not fully recreate the original class definitions used in BioCreAtIvE, leading to potential discrepancies in performance assessment.
Statistical Information
P-Value
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