Events

Speaker - Nitin Kumar Research Scholar

ABSTRACT

We present a proof of concept implementation of the development of personal assistant chatbot that can accept input/output in both text and verbal form using google’s gTTS (Google Text-to-Speech) speech-synthesiser. We have used the machine learning (ML) library scikit-learn naive bayes classifier for question detection and natural language processing (NLP) library NLTK and nps_chat corpora for implementing the model at the backend and processing the data. Model is trained on over a small sample of full-text articles on which NLP based preprocessing steps & feature extraction are performed which is used down the line model training. The initial development of the bot is based on Keyword Recognition-Based method which is a command line based deployment. Similarly, we also introduce Transformer-based masked language models ScholarBERT, Which is pretrained on a large collection of scientific research articles (2.2B tokens) for future integration. We also propose an updated version with training across a larger range and number of articles and adding more features such as contextual and semantic interpretation and inclusion of Language models and more Robust NLU’s like RASA.