The TREC 2004 genomics track categorization task: classifying full text biomedical documents
2006

Classifying Biomedical Documents in the TREC 2004 Genomics Track

Sample size: 11880 publication Evidence: low

Author Information

Author(s): Aaron M Cohen, William R Hersh

Primary Institution: Oregon Health & Science University

Hypothesis

Can automated classification improve the categorization of full-text biomedical documents for gene ontology annotation?

Conclusion

Automated classification of documents for gene ontology annotation is challenging, but necessary for improving biomedical curation.

Supporting Evidence

  • The triage subtask had a mean utility measure of 0.3303.
  • The best performing run achieved a Unorm of 0.6512 using the MeSH term 'Mice'.
  • Automating GO annotation tasks could significantly benefit biomedical curation.
  • Feature overlap between training and test sets was less than expected.
  • Gene name recognition improved classification performance.

Takeaway

This study tried to teach computers how to sort scientific papers about genes, but it was really hard and they didn't do much better than just looking for the word 'mouse'.

Methodology

The study involved classifying full-text documents from three journals over two years, using a triage task and annotation tasks based on gene ontology.

Potential Biases

Potential bias due to reliance on the MeSH term 'Mice' as a predictor for triage decisions.

Limitations

The study faced challenges due to sparse sampling of GO terms and the complexity of the GO triage task.

Participant Demographics

The study involved 33 participating groups from various institutions.

Digital Object Identifier (DOI)

10.1186/1747-5333-1-4

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication