부산대학교 도서관

In recent years, there has been a demand for natural language processing techniques that accurately predict the following word in a phrase. Existing Bangla next word prediction systems have significant draw- backs, including the inability to predict outside of their library and lengthy training cycles. To address these limitations, we present a Bi-directional Long Short-Term Memory (Bi-LSTM) with a Self-Attention method for developing language models that can anticipate the next word from a given input. Our models beat existing state-of-the-art systems. Our models obtain high testing accuracy for several n-gram language models, including 97.98% for the 7-gram, 97.97% for the 6- gram, and 97.91% for the 5-gram. Furthermore, the training time and number of epochs required to reach the desired accuracy are greatly reduced by our suggested system. While other works utilized 1000s of epochs to reach desired precision, we employed the Self-Attention method to achieve the same degree of accuracy in only 200 epochs. Our goal is to contribute to the advancement of natural language processing technology for use in a variety of industries such as healthcare, education, and finance. We hope to assist industry in developing sustainable technology by delivering accurate and efficient language models