Improving Supervised Keyphrase Indexer Classification of Keyphrases with Text Denoising

摘要

Text denoising is a text reduction method that extracts the content-rich parts from full-text research articles. These content-rich parts, known as the denoised texts, suffice information extraction tasks, such as automatic relation mining and keyphrase extraction. In this paper, we concentrate on the latter and show that two state-of-the-art supervised keyphrase indexers named KEA and KEA++, when paired with text denoising, induce improved keyphrase classifiers. The classifiers' performances are demonstrated on three standard full-text corpora collected from the food and agriculture, nuclear physics and biomedical domains. Using the denoised parts of the texts, the indexers induce keyphrase classifiers that are later used for full-text keyphrase extraction. Experimental results show that against a gold standard these classifiers perform better than those induced from full texts.