Evaluating Text Structure Models for Cancer Risk Assessment
Author Information
Author(s): Guo Yufan, Korhonen Anna, Liakata Maria, Silins Ilona, Hogberg Johan, Stenius Ulla
Primary Institution: Computer Laboratory, University of Cambridge
Hypothesis
Can existing schemes for capturing information structure in scientific documents be effectively applied to biomedical abstracts for cancer risk assessment?
Conclusion
The study demonstrates that information structure schemes can be effectively applied to biomedical abstracts, improving the speed of information retrieval for cancer risk assessors.
Supporting Evidence
- The study found that all three information structure schemes could be identified in abstracts with high accuracy.
- Cancer risk assessors were able to find relevant information significantly faster in annotated abstracts compared to unannotated ones.
- The user test indicated that both manually and automatically annotated abstracts saved time for experts.
Takeaway
This study shows that using special labels on scientific abstracts helps experts find important information faster, which is really helpful when assessing cancer risks.
Methodology
The study involved annotating a corpus of cancer risk assessment abstracts using three different information structure schemes and evaluating their effectiveness through user tests.
Potential Biases
Potential bias may arise from the expertise level of annotators and the selection of abstracts.
Limitations
The study's findings may not generalize to all types of biomedical literature, and the inter-annotator agreement was lower for the most detailed scheme.
Participant Demographics
Three cancer risk assessors with varying levels of experience participated in the user test.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website