A Novel Approach of Multi-label classification for low resource Braj Language with integration of Synonyms and POS Tagging

Abstract

This research explores Braj language as one of the niche Indian low-resource languages for text analytics for the very first time. The research mainly focuses on entities synonym extraction, POS (part of speech tagging). Three feature extraction methods namely deep learning, transformer and count based are implemented for multi label classification tasks on the Holi Sagar Braj dataset. The best results are achieved with a combination of Linear SVM classification and TF-IDF with F1 score 0.88, precision 0.81, and recall 0.97. Results showed low efficiency in evaluating LSTM and IndicBERT models on the same dataset due to dialectal forms of the language challenges. The research on the Braj language for multi label classification advances in feature engineering and classification techniques for Indian regional language processing.