Liuliu writeup of Wang 2008

This is a review of Wang_2008_iterative_set_expansion_of_named_entities_using_the_web by user:Liuliu.

The iterative SEAL is an extension to SEAL system. It is designed to handle more seeds by feeding some number of seeds at each iteration to the system and accumulating results from each iteration. I think this paper is more like a method comparison, where authors compared two different iterative processes(supervised and bootstrapping), two seeding strategies(fixed seed size and increasing seed size) and five ranking methods(random walk with restart, pagerank, bayesian sets, wrapper length and wrapper frequency) in the context of Iterative SEAL system.

Increasing seed strategy(ISS) is more conservative compared with Fixed seed strategy(FSS), hence it works better with bootstrapping while FSS works better with supervised iterative process. Bayesian nets is the best ranking method for supervised iterative processes, while random walk is the best for bootstrapping, since random walk is more tolerate for noisy entities.