MScanner: A Tool for Classifying Medline Citations
Author Information
Author(s): Graham L. Poulter, Daniel L. Rubin, Russ B. Altman, Cathal Seoighe
Primary Institution: University of Cape Town
Hypothesis
Can a non-domain-specific classifier effectively retrieve relevant Medline citations using supervised learning?
Conclusion
MScanner is an effective classifier for retrieving relevant Medline citations across various topics.
Supporting Evidence
- MScanner achieved ROC areas between 0.97 and 0.99 in cross-validation.
- The classifier returned results in approximately 90 seconds for 16 million Medline records.
- Averaged precision ranged from 0.69 to 0.92 across different topics.
Takeaway
MScanner helps find important medical articles quickly by learning from examples, making it easier to search through a lot of information.
Methodology
MScanner uses a Bayesian classifier trained on relevant PubMed IDs to rank Medline articles by relevance.
Potential Biases
Potential bias due to the reliance on the training set and the nature of the selected irrelevant articles.
Limitations
Performance varies by topic and may require a sufficient number of training examples for accurate classification.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website