Named Entity Disambiguation (NED) aims at disambiguating named entity mentions in a text to their corresponding entries in a knowledge base such as Wikipedia. Itis a fundamental task in Natural Language Processing (NLP)and has many applications such as information extraction, information retrieval, and knowledge acquisition. In the past decade, a number of methods have been proposed for the NED task. However, most of existing work focuses on exploring many more useful information to help tackle this problem. The effectiveness of different features proposed for the task are not well-studied in a same platform. In this paper, we extract various remarkable features by leveraging statistical, textual and semantic information, and evaluate various combinations of the multiaspect features for the disambiguation task in the same platform. Specifically, we utilize two learning to rank methods to combine different features, train and test the combined methods on several standard data sets. Through extensive experiments, we investigate the effects on the quality of the disambiguation of exploiting different features and show which combinations of features are the best choices for disambiguation.
展开▼
