Document Expansion Based on WordNet for Robust IR

The use of semantic information to improve IR is a long-standing goal. This paper presents a novel Document Expansion method based on a WordNet-based system to find related concepts and words. Expansion words are indexed separately, and when combined with the regular index, they improve the results in three datasets over a state-of-the-art IR engine. Considering that many IR systems are not robust in the sense that they need careful finetuning and optimization of their parameters, we explored some parameter settings. The results show that our method is specially effective for realistic, non-optimal settings, adding robustness to the IR engine. We also explored the effect of document length, and show that our method is specially successful with shorter documents.

• Type: Journal
• Year: 2010
• Where: COLING
• Authors: Eneko Agirre, Xabier Arregi, Arantxa Otegi
• Keywords: COLING 2010, Computational Linguistics, Documents, IR Engine, State-of-the-art Ir Engine