Exploring Linguistic Features for Named Entity Disambiguation

Resolving named entities is important for a number of natural language processing applications. However, a named entity has multiple name variations while different entities could share the same surface. State-of-the-art systems are based on a global resolution method and mostly adopt link-based features that leverage relationships of co-occurring entities in the knowledge. We found that linguistic features can also significantly affect disambiguation. In this work, we try to explore important linguistic features from context, which could be the fundamental part of the combination of global resolution method and effective features. Therefore, we study and compare the effects of linguistic features in a comprehensive way. Moreover, we found effective linguistic features according to the experiment results.