13 Nov 2004
WIDM 04: Lee et al. Co-training Web Block Classification
18
Conclusion
•Co-training model for web block classification
•Achieves 28.5% reduction in error in main task
•However, fails in
–Detecting fine grained classes
•→ Exploit templates, IE methods, path similarity and context
»
–Likely needs enough unlabeled data
•→ Re-run using more experimental data
»
–Dependent on learning model
•→ Looking to change learning package