Bengali Named Entity Recognition using Classifier Combination

This paper reports about the development of a Named Entity Recognition (NER) system for Bengali by combining the outputs of the classifiers like Maximum Entropy (ME), Conditional Random Field (CRF) and Support Vector Machine (SVM) using a majority voting approach. The training set consists of approximately 150K wordforms and has been manually annotated with the four major NE tags such as Person name, Location name, Organization name and Miscellaneous name tags. Lexical context patterns, generated from an unlabeled corpus of 3 million wordforms, have been used in order to improve the performance of the classifiers. Evaluation results of the voted system for the gold standard test set of 30K wordforms have demonstrated the overall recall, precision, and f-Score values of 87.11%, 83.61%, and 85.32%, respectively, which shows an improvement of 4.66% in f-Score over the best performing SVM based system and an improvement of 9.5% in f-score over the least performing ME based system.