Ontology-Based Feature Transformations: A Data-Driven Approach

Abstract. We present a novel approach to incorporating semantic information to the problems of natural language processing, in particular to the document classification task. The approach builds on the intuition that semantic relatedness of words can be viewed as a non-static property of the words that depends on the particular task at hand. The semantic relatedness information is incorporated using feature transformations, where the transformations are based on a feature ontology and on the particular classification task and data. We demonstrate the approach on the problem of classifying MEDLINE-indexed documents using the MeSH ontology. The results suggest that the method is capable of improving the classification performance on most of the datasets.

• Type: Conference
• Year: 2004
• Where: TAL
• Authors: Filip Ginter, Sampo Pyysalo, Jorma Boberg, Jouni Järvinen, Tapio Salakoski
• Keywords: Classification Task, Natural Language Processing, Semantic Relatedness, Semantic Relatedness Information, TAL 2004