Learning word sense disambiguation in biomedical text with difference between training and test distributions

Word Sense Disambiguation methods based on machine learning techniques with lexical features suffer from the discordance between distributions of the training and test documents, due to the diversity of lexical space. To tackle this problem, this paper proposes Support Vector Machines with Example-wise Weights. In this method, the training distribution is matched with the test distribution by weighting training examples according to their similarity to all test data. The experimental results show the distribution change between the training and test data is actually recognised and the proposed method which considers this change in its training phase outperforms ordinary SVMs.