Philgoo Han writeup of Wang and Cohen
From Cohen Courses
Jump to navigationJump to searchThis is a review of Wang_2008_iterative_set_expansion_of_named_entities_using_the_web by user:Ironfoot.
- Iterative SEAL
- Limitation of SEAL on using larger seeds => iterative SEAL
- Fixed Seed Size
- Introduce two seeds at every iteration
- Increasing Seed Size
- Introduce one seed at every iteration and combine to previous seed set. (increase by one)
- Until maximum values reached
- Bootstrapping
- The new seed introduction in FSS and ISS can be done in unsupervised manner.
- Ranking Methods
- Random walk with restart
- PageRank
- Bayesian sets
- Wrapper length
- Wrapper frequency (baseline)
- Experximent result
- Supervised: FSS converge faster than ISS (RW = BS > ...)
- Bootstrapping: ISS converge to improve but FSS fail (RW > BS > ...)
- ISS used in this paper seems closer to a variation of FSS. The sead set doesn't increases only twice.
- Might it be more worth to improve the tool itself than finding a way to use the tool in a more clever way?