<![CDATA[Automated Essay Scoring is one of the challenges in the field of educational technology, particularly in subjects requiring language assessment, such as the English language. Manual assessment performed by teachers is time-consuming, subjective, and has the potential for inconsistency between assessors, which can give rise to unfairness in scoring. In order to overcome this issue, this study proposes an Automated Essay Scoring (AES) approach by combining FastText word embeddings and the Long Short-Term Memory (LSTM) algorithm to predict the scores of student essays. The innovation of this research lies in combining question and answer data into a single input variable, where the student's answer is positioned between the special tokens <startanswer> (prefix) and <endanswer> (suffix), allowing the model to learn the context of the question and answer simultaneously. The data 500 data points (100 students x 5 essay questions). The LSTM model was then trained with a combination of hyperparameters, namely the number of units in the hidden layer, learning rate, batch size, and dropout rate. Model performance was measured using Mean Absolute Percentage Error. Based on the experimental results, the Bi-LSTM model with the best hyperparameter settings achieved a MAPE value of 7.84%, which is better than the unidirectional LSTM model at 9.12%. This study has proven that the combination of FastText and Bi-LSTM is an effective approach for essay score prediction in the context of language learning for junior high school students]]>
Copyrights © 2026
