Semi-supervised learning by Fuzzy clustering and Ensemble learning
Hiroyuki Shinnou, Minoru Sasaki
This paper proposes a semi-supervised learning method using Fuzzy clustering to solve word sense disambiguation problems. Furthermore, we reduce side effects of semi-supervised learning by ensemble learning. We set N classes for N labeled instances. The n-th labeled instance is used as the prototype of the n-th class. By using Fuzzy clustering for unlabeled instances, prototypes are moved to more suitable positions. We can classify a test instance by the k Nearest Neighbor (k-NN) with the moved prototypes. Moreover, to reduce side effects of semi-supervised learning, we use the ensemble learning combined the k-NN with initial labeled instances, which is initial prototype, and the k-NN with prototypes moved by Fuzzy clustering.
Fuzzy clustering, Semi-supervised learning, Ensemble learning, Word Sense Disambiguation