Word Sense Disambiguation Using Active Learning with Pseudo Examples

摘要

In recent years, there have been attempts to apply active learning for Word sense disambiguation (WSD). This active learning technique selects the most informative unlabeled examples that were most difficult to disambiguate. The most commonly addressed problem has been the extraction of relevant information, where the system constructs a better classification model to identify the appropriate sense of the target word. Previous research reported that it is effective to create negative examples artificially (i.e., pseudo negative examples). However, this method works only for words that appear in a small number of topics (e.g., technical terms) because the evaluation set is strongly biased. For common noun or verb words, it is hard to apply this system so that problems still remain in the active learning with pseudo negative examples for WSD. In this paper, to solve this problem, we propose a novel WSD system based on active learning with pseudo examples for any words. This proposed method is to learn WSD models constructed from training corpus by adding pseudo examples during the active learning process. To evaluate the effectiveness of the proposed method, we perform some experiments to compare it with the result of the previous methods. The results of the experiments show that the proposed method achieves the highest precision of all systems and can extract more effective pseudo examples for WSD.