Multi-Engine Approach for Named Entity Recognition in Bengali

Multi-Engine Approach for Named Entity Recognition in Bengali* Asif Ekbal a, and Sivaji Bandyopadhyay a a Department of Computer Science and Enginneering, Jadavpur University, Kolkata-700032, India asif.ekbal@gmail.com and sivaji_cse_ju@yahoo.com Abstract. This paper reports about a multi-engine approach for the development of a NER system in Bengali by combining the classifiers such as Maximum Entropy (ME), Conditional Random Field (CRF) and Support Vector Machine (SVM) with the help of weighted voting approach. The training set consists of approximately 272K wordforms, out of which 150K wordforms have been manually annotated with the four major named entity (NE) tags such as Person, Location, Organization and Miscellaneous tags. An appropriate tag conversion routine has been defined in order to convert the 122K wordforms of the IJCNLP-08 NER shared task1, into the desired forms. The classifiers make use of the different contextual information of the words along with the variety of features that are helpful in predicting the various NE classes. Lexical context patterns, which are generated from an unlabeled corpus of 3 million wordforms in a semi-automatic way, have been used as the features of the classifiers in order to improve their performance. In addition, we have developed a number of techniques to post-process the output of each of the classifiers in order to reduce the errors and to improve the performance. Finally, we have applied weighted voting approach to combine the systems. Results show the effectiveness of the proposed approach with the overall average recall, precision, and f-score values of 93.98%, 90.63%, and 92.28%, respectively, which shows an improvement of 14.92% in f-score over the best performing baseline SVM based system and an improvement of 18.36% in f-score over the least performing baseline ME based system. The proposed system also outperforms the other existing Bengali NER system. Keywords: Named Entity Recognition, Maximum Entropy, Conditional Random Field, Support Vector Machine, Weighted Voting, Bengali 1. Introduction Named Entity Recognition (NER) is an important tool in almost all Natural Language Processing (NLP) application areas including machine translation, question answering, information retrieval, information extraction, automatic summarization etc. The current trend in NER is to use the machine-learning (ML) approach, which is more attractive in that it is trainable and adoptable and the maintenance of a ML based system is much cheaper than that of a rule-based one. The representative ML approaches used in NER are Hidden Markov Model (HMM) (BBN’s IdentiFinder in (Bikel, 1999)), ME (New York University’s MENE in (Borthwick, 1999)), CRFs (Lafferty et al., 2001) and SVM (Yamada et al., 2002). The process of stacking and voting method for combining strong classifiers like boosting, SVM and TBL, on NER task can be found in (Wu et al., 2003). Florian et al. (2003) tested different methods for combining the results of four systems and found that robust risk minimization worked best. The work reported in this paper differs from the existing works in the sense that here, we have conducted a number of experiments to improve the performance of the classifiers with the lexical context patterns, …

Content truncated to 500 words