An Emerging Area of Research: Building an Assamese AI Chatbot for Educational Institutions Using Bi-LSTM Model

With the evaluation of technological advancements and the increasing demand for educational resources in regional languages, AI chatbots have become one of the latest areas of research. Assamese is the language that the majority of people speak in the northeastern part of India, specially in the state of Assam. Compared to other languages such as English, building an AI chatbot in the Assamese language is a challenging task due to a lack of available resources. In this research work, we have developed our own Assamese dataset for chatbots by gathering data from various sources like student support services, FAQs, emails, Facebook, etc., from educational institutions and converted them into the Assamese language. The bag-of-words (BOW) method is used to obtain features from unprocessed text data. We then developed five deep learning models: FFDN, RNN, LSTM, GRU, and bi-LSTM, and tested the dataset with these models. This paper begins by highlighting the significance of AI chatbots in educational institutions and the need for regional language support. The next section presents the methodology, containing brief details of data collection, pre-processing of Assamese text data, and deep learning models and parameters used. Next, we have measured the performance of the deep learning model using metrics recall, precision, and F1 Score, and compared with other models to decide the best one. The bi-LSTM model has been able to achieve an accuracy of 89.87% compared to other models. Lastly, we present the limitations of the study and future prospects.