Klasifikasi Kategori Feedback EDOM Primakara University dengan Algoritma RNN LSTM

Abstract

This research aims to apply the Long Short-Term Memory (LSTM) algorithm in sentiment analysis of the Student Evaluation of Teaching (EDOM) at Primakara University, with the goal of improving higher education quality through the evaluation of faculty performance. Primary data was collected through interviews with the Quality Assurance Agency, while secondary data was obtained from EDOM for the academic years 2020/2021 to 2022/2023. A sentiment classification model was constructed using LSTM, initially dividing the data into 20 categories. To balance the data distribution, these categories were then merged into 6 main categories. The model was trained and tested using cross-validation, achieving an accuracy of 97%. However, when the model was tested with 100 new EDOM data points, the accuracy decreased to 74%, which is suspected to be caused by the emergence of new vocabulary that was not recognized or stored in the trained machine learning model. This decline in accuracy indicates the limitations of the model in handling new EDOM data that differs from the training data, and highlights the importance of periodic model updates or the use of out-of-vocabulary techniques to improve model performance in the future.