Development of FasTtext Pre-Trained Model for Bangla NLP Research
Lists
Only you see this
Abstract
Due to the expansion of data generation, many more NLP tasks need to be solved. To do so, word representation plays a vital role. Computational-based word embedding in various high languages is very resourceful. However, considering low resource languages like Bangla have very limited resources available like models, toolkits, and datasets. Here, in this paper, we have proposed a BanglaFasttext word embedding toolkit based on python and two big pre-trained Bangla models of fasttext (skipgram & cbow). These pre-trained models are trained on a large Bangla corpus (around 20 million observations). Besides BaglaFastText outperforms Facebook’s fasttext with a significant margin. To evaluate these pre-trained models, we have used three popular textual Bangla datasets, developed models on top of that various machine learning classical approaches as well as deep neural network and these have been implemented for performance analysis. The evaluations have been shown a pretty good performance rather than existing word embedding techniques and Facebook Bangla fasttext pre-trained for Bangla NLP. In addition, we also compared the performances of this work considering the original works of these textual datasets where our proposed work outperforms all these existing works.
BibTeX
@misc{kowsher2021development,
author = {Md. Kowsher and Md. Shohanur Islam Sobuj and Md. Fahim Shahriar and Nusrat Jahan Prottasha},
title = {{Development of FasTtext Pre-Trained Model for Bangla NLP Research}},
year = {2021},
howpublished = {SSRN},
}Loading related content...