Ensembling based on Feature Space Restructuring with Application to WSD

We propose a new ensembling method of Support Vector Machines (SVMs) based on Feature Space Restructuring. In the proposed method, the weighted majority voting method is applied for several restructured feature spaces computed by Independent Component Analysis (ICA) or Latent Semantic Indexing (LSI). We evaluated the proposed method empirically by applying this method to Word Sense Disambiguation. This ensembling method is not specific to WSD and can be applied to various tasks, even when it is difficult to arrange different viewpoints at the task to construct multiple classifiers.

• Type: Conference
• Year: 2001
• Where: NLPRS
• Authors: Hiroya Takamura, Hiroyasu Yamada, Taku Kudo, Kaoru Yamamoto, Yuji Matsumoto
• Keywords: Feature Space Restructuring, Feature Spaces, Majority Voting Method, Natural Language Processing, NLPRS 2001