An Enhanced Neural Word Embedding Model for Transfer Learning

Due to the expansion of data generation, more and natural language processing (NLP) tasks are needing be solved. For this, word representation plays a vital role. Computation-based embedding in various high languages is very useful. However, until now, low-resource such as Bangla have had limited resources available terms models, toolkits, datasets. Considering this fact, paper, an enhanced BanglaFastText model developed using Python two large pre-trained models FastText (Skip-gram cbow). These were trained on collected corpus (around 20 million points text data, which every paragraph considered point). outperformed Facebook’s by significant margin. To evaluate analyze performance these proposed work accomplished classification based three popular textual datasets, machine learning classical approaches, well deep neural network. The evaluations showed superior over existing techniques Facebook for NLP. In addition, original concerning datasets provides excellent results. A toolkit proposed, convenient accessing embedding, obtaining semantic relationships word-by-word or sentence-by-sentence; sentence approaches; also unsupervised finetuning any linguistic dataset.