Abstract:
There is a rising trend in using conversational agents as virtual assistants and voice in terfaces in many business domains and applications. These agents aided users in getting the
information they required by responding to their queries through text input, audio input,
or both without human intervention. The conventional chatbots could not produce context based responses to the user inputs as they could not read the text data from both directions.
This work mainly focuses on designing and developing the AI-bot using Sequence to Sequence
Long Short Term Memory (Seq2Seq LSTM) for the English language on Cornell movie dia logue corpus and a manually created dictionary dataset. The AI-bot using Seq2Seq LSTM,
detected acronyms and generated complete relevant answers to the users’ queries. Sentiment
analysis retrieved meaningful data from text documents, and it helped the agents/bots to
communicate with the users based on the context of the conversation and their emotions.
Finally, the performance of AI-Bot using Seq2Seq LSTM is evaluated in terms of accuracy,
precision, and recall. The model has received an accuracy of 97% and outperformed the
RNN model on both datasets. The model has utilized long-range memory dependencies of
data. Also, the performance of BERT is found satisfactory in terms of accuracy
