Bangla Text Generation Using Bidirectional Optimized Gated Recurrent Unit Network

摘要

Natural language processing is a vital branch of Artificial Intelligence, a bridge to communicate between computer and human in their languages and is used to generate automatic text, summarize articles, machine translation, etc. In this paper, we proposed a model of Bangla sentence generation that uses the Optimized Gated Recurrent Unit network based Recurrent Neural network model. We used OGRU based model instead of the regular Gated Recurrent Unit network because OGRU generates better results than GRU. For this purpose, we have used three Bangla corpus datasets, which consist of more than 49 million words. Here we used Bangla Natural Language Toolkit to preprocess the datasets. We trained our model with those three different datasets and achieved an accuracy of 97% on average and it can generate a paragraph of text based on the context. With this accuracy and ability to generate text, it is proved that our model is much efficient in Generating Bengali Sentences.