Domain Information For Fine-Grained Person Name Categorization

Named Entity Recognition became the basis of many Natural Language Processing applications. However, the existing coarse-grained named entity recognizers are insufficient for complex applications such as Question Answering, Internet Search engines or Ontology population. In this paper, we propose a domain distribution approach according to which names which occur in the same domains belong to the same fine-grained category. For our study, we generate a relevant domain resource by mapping and ranking the words from the WordNet glosses to their WordNet-Domains. This approach allows us to capture the semantic information of the context around the named entity and thus to discover the corresponding fine-grained name category. The presented approach is evaluated with six different person names and it reaches 73% f-score. The obtained results are encouraging and perform significantly better than a majority baseline.