Philgoo Han writeup of Sutton and McCallum
From Cohen Courses
Jump to navigationJump to searchThis is a review of Sutton_2004_collective_segmentation_and_labeling_of_distant_entities_in_information_extraction by user:Ironfoot.
- Optimizing joint prob. dist.(HMM) -> optimizing cond. dist.(CRF) -> long distance dependency(skip-chain CRF)
- Connect edges in probabilistic structure between similar tokens
- Now the structure contains loops, approximate inference
- How much would be reasonably accurate. This would have been an important factor for result analysis.
- As the result shows this algorithm seems to work good on locations and spearkers
- However low in stime, etime
- Might the algorithm try to make every time token consistent? (that is all into stime or all into etime)
