The class merging algorithm introduced in the paper is useful. Instead of finding the optimal classes, the author used class merging to merge two classes at a time, according to the minimum loss of average between-word mutual information. It is suboptimal but can be done in practice. But it still seems a little slow. Maybe guided clustering is helpful: only introduce classes to some of the words based on linguistic knowledge, leaving other words unclustered.
Reviewed by
zzb3886
as
- 2008-12-29 22:42:57
Search
Reviews
[Write a review of this article]
Find related articles from these CiteULike users
