Cluster Analysis and Classification of Named Entities

Proceedings of the Fourth International Conference on Language Resources and Evaluation (LREC 2004)

Abstract

This paper presents a statistics-based and language independent unsupervised approach for clustering possible named entities. We describe and motivate the features and statistical filters used by our clustering process. Using the Model-Based Clustering Analysis software we obtained different clusters of named entities. The method was applied to Bulgarian and English. For some clusters, precision is close to 100%; this helps human validation and saves time. Other clusters still need further refinement. Based on the obtained clusters, it is possible to classify new named entities.