Vector Representation of Bengali Word Using Various Word Embedding Model

Abstract

To transfer human understanding of language to a machine we need word embedding. Skipgram, CBOW, and fastText is a model which generate word embedding. But finding pretrained word embedding model for the Bengali language is difficult for researchers. Also, training word embedding is time-consuming. In this paper, we discussed different word embedding models. To train those models, we have collected around 500000 Bengali articles from various sources on the internet. Among them, we randomly chose 105000 articles. Those articles have 32 million words. We trained them on SkipGram and CBOW model of Word2Vec, fastText. We also trained those words in Glove model. Among the all result fastText (Word2Vec) gave us a satisfactory result.