Related Experiment Videos
Improving the performance of dictionary-based approaches in protein name recognition
Yoshimasa Tsuruoka1, Jun'ichi Tsujii
1CREST, Japan Science and Technology (JST) Agency, Honcho 4-1-8, Kawaguchi-shi, Saitama 332-0012, Japan. tsuruoka@is.s.u-tokyo.ac.jp
Journal of Biomedical Informatics
|November 16, 2004
Summary
This study improves protein name recognition in biomedical texts by using machine learning to reduce false positives and novel methods to handle spelling variations, boosting overall accuracy.
Area of Science:
- Biomedical Informatics
- Natural Language Processing
- Computational Biology
Background:
- Dictionary-based methods are crucial for protein name recognition in biomedical literature.
- These methods face challenges with false positives from short names and low recall due to spelling variations.
Purpose of the Study:
- To enhance protein name recognition accuracy in biomedical documents.
- To address false positives and low recall issues inherent in dictionary-based approaches.
Main Methods:
- Utilized machine learning, specifically a naive Bayes classifier, to filter false positive protein name recognitions.
- Developed two methods to improve recall for spelling variations: approximate string searching and a probabilistic variant generator for dictionary expansion.
Main Results:
- Naive Bayes filtering significantly improved precision with minimal recall loss, yielding a 10.8% F-measure increase.
- Dictionary expansion using the proposed variant generator further enhanced performance, achieving an F-measure of 66.6%.
Conclusions:
- Machine learning effectively reduces false positives in dictionary-based protein name recognition.
- Probabilistic variant generation offers a promising approach to overcome spelling variations and improve recall in biomedical text mining.