Detecting Topics in Biomedical Texts
Author Information
Author(s): Lee Minsuk, Wang Weiqing, Yu Hong
Primary Institution: Columbia University
Hypothesis
Can supervised and unsupervised methods effectively detect topics in biomedical literature?
Conclusion
Supervised topic spotting methods are more accurate than unsupervised topic clustering methods, but the latter are more robust in real-world applications.
Supporting Evidence
- The naïve Bayes classifier achieved an accuracy of 66.4% for topic detection.
- Incorporating MeSH terms and UMLS semantic types improved classification performance.
- Unsupervised methods showed robustness but lower accuracy compared to supervised methods.
Takeaway
This study shows how computers can help find topics in medical articles, making it easier for scientists to find the information they need.
Methodology
The study used supervised machine learning (naïve Bayes) and unsupervised hierarchical clustering to categorize over 15,000 biomedical articles.
Potential Biases
Potential bias in the selection of articles and the reliance on predefined topics.
Limitations
The study primarily focused on documents with a single topic, which may not reflect the reality of multi-topic documents.
Participant Demographics
The study analyzed articles from the Online Mendelian Inheritance in Man (OMIM) database.
Statistical Information
P-Value
0.0226
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website