A Generative-Based Chatbot for Daily Conversation: A Preliminary Study
Fitra Abdurrachman Bachtiar, Alfirsa Damasyifa Fauzulhaq, Marvel Timothy Raphael Manullang, Fabiansyah Raam Pontoh, Kuncahyo Setyo Nugroho, Novanto Yudistira · 2022
Currently AI has been integrated in our daily life. One of basic example of AI in our daily life is chatbot. The common chatbot that is developed is a rule-based chatbot. Rule-based chatbot have several drawbacks that is easy to predict, repetitive, an unnatural conversation might happened. In addition rule-based chatbot would rely on fixed pair question and answer, developed in a closed domain conversation, and has limited self-learning. Due to this problem, there is a need to develop a chatbot that is able to dynamically response to the question. The chatbot model in this study is based on Simple Dialog and Daily Dialog dataset. The dataset are then merged into single dataset. In this research, we proposed the Seq2seq model architecture with LSTM and GRU cell to create a Generative-Based chatbot. The Seq2seq model consists of two parts, encoder and decoder. The proposed model is evaluated using Cross Entropy Loss and BLEU score. The result shows that the LSTM model has better performance than the GRU model. The LSTM model resulted in 0.7064 training loss, 8.9740 validation loss and 0.0588 BLEU. Meanwhile, the GRU model resulted in 2.1125 training loss, 7.1840 validation loss and 0,0028 BLEU. Compared to GRU, LTSM model have the ability to generate a more acceptable response to the given questions.