International Journal of Science, Engineering and TechnologyAuthors: Assistant Professor K.Petchiammal, M. Mathumathi, S. Ramesh, S.Karthikeyen, S.M. Nusra Fathima, K. Murugeshwari
Abstract: The growing demand for automated, human-like conversational systems has positioned Natural Language Processing (NLP) as a core enabler of intelligent chatbot design. This study presents a deep learning-based framework for building an intelligent chatbot capable of intent recognition, entity extraction, and context-aware response generation. Three NLP architectures are compared for the intent-classification and response-generation pipeline: a Bidirectional LSTM with attention, a fine-tuned BERT-based intent classifier, and a Transformer-based sequence-to-sequence model. A labelled conversational dataset is preprocessed through tokenization, stop-word handling, lemmatization, and word-embedding generation before model training. The models are evaluated using accuracy, precision, recall, F1-score, and BLEU score to assess both classification quality and generated-response fluency. Experimental results show that the BERT-based intent classifier achieves the highest intent-recognition accuracy, while the Transformer sequence-to-sequence model produces the most fluent and contextually relevant responses. The comparative study offers practical guidance for selecting NLP architectures suited to customer support, virtual assistance, and other conversational-AI applications, and contributes toward building chatbots that are more accurate, context-aware, and reliable.
