Suryani, - (2024) GENERATIVE-BASED CHATBOT MENGGUNAKAN LONG SHORT-TERM MEMORY PADA WEBSITE DIREKTORI AKADEMIK FAKULTAS SAINS DAN TEKNOLOGI. Skripsi thesis, Universitas Islam Negeri Sultan Syarif Kasim Riau.

Preview

Text Laporan Tugas Akhir Suryani_12050320389.pdf - Published Version Download (6MB) | Preview

Abstract

Since 2018, the seminar FST website has been in operation and has served as the primary hub for administrative services for faculty of Science and Technology students. This website is a platform for handling requests for official correspondence from students, including requests for documents such as enrollment verification letters, good conduct certificates, and other administrative letters. Since its launch, the website has received a high volume of traffic. In 2023, the site averaged 27,753 clicks per day and 11,648 submissions were successfully processed. Unfortunately, this website still lacks information about administrative procedures. In addition, this website does not have a Frequently Asked Questions (FAQ) page. These issues, need to be resolved immediately. This study aims to develop a chatbot using sequence-to-sequence architecture integrated with the Bi-LSTM network to provide conversations about administrative services for Faculty of Science and Technology students. Several test scenarios were conducted with different parameters to obtain optimal model performance. The tests were conducted using Adam's optimizer with two different learning rates, namely 0.01 and 0.001. The training data was pre-divided with a ratio of 90:10, 80:20, and 70:30, and each training data was tested with different batch size values, namely 8, 16, and 32. The test results show that the best accuracy on training data is 0.9964 and on validation data is 0.8756, using training data with a ratio of 90:10, Adam optimizer with a learning rate of 0.01, and batch size 16. Then, the model performance is calculated using Bilingual Evaluation Understudy (BLEU) with a score of 0.7724 for some tested data. In addition to BLEU metrics, a user satisfaction survey was conducted using the Chatbot Usability Questionnaire (CUQ) to measure user acceptance of the developed chatbot. This survey obtained an average score of 62.2. Based on the SUS rating scale, it was found that the score of 62.2 falls within the "Good" category.

Actions (login required)

View Item