This talk describes work I have been doing with Cyndy Parr at EOL to annotate text with DBpedia URIs and to generate a species associations network. It was presented to the Boston Python User Group in April 2013.
4. Performance Metrics
• 1631 URIs assigned to 487 text objects from 21 test
species
• 83% were correct
• 20% of the text objects were not assigned a URI
• 239 keys in the dictionary
• Precision 0.89, Recall 1, F1 Score 0.942
5. Challenges and Errors
• Many ways to say the same thing
– Uterine cannibalism = oophagy
• Negation (9%)
• Describing related taxa (30%)
• Word/phrase part (27%)
• Generalities (15%)
• Homonym (13%)