Implementasi Chatbot menggunakan Long Short Term Memory (LSTM) dengan Pendekatan Retrieval-Based (Studi Kasus: Fakultas Ilmu Komputer)

Abstract

Education has been impacted by Industrial Revolution era Technology 4.0. Developing competent future generations is mostly dependent on education. To achieve this objective, higher education institutions and universities depend on websites as information hubs. However, challenges emerge when students attempt to access academic content on faculty web pages. This research focuses on creating a retrieval-based chatbot with the Long Short Term Memory (LSTM) model to enhance information access efficiency. This research discusses the implementation of tag prediction using the LSTM model and answer prediction within the chatbot. The LSTM model was selected due to its ability to remember long-term and sequential information. This research also explores the impact of hyperparameters such as the learning rate value, the number of epochs, and the number of hidden sizes on the LSTM model’s performance. The findings show that the LSTM model can provide high-quality and relevant responses. In tag prediction, the LSTM model with a hidden size of 64 and 100 epochs with a 0.01 learning rate yields an average accuracy value of 0.823. Meanwhile, in answer prediction, the combination of the LSTM model with the cosine similarity function and FastText embedding achieves an accuracy value of 0.907. This study produced successful results by effectively integrating the LSTM model into a retrieval-based chatbot. However, more modifications and improvements are required to improve the standard and effectiveness of this chatbot’s user support offerings. This study helps to enhance the effectiveness of chatbots and offers valuable information for further research on retrieval-based chatbots