Enhancing Aspect-Based Sentiment Analysis in Student Reviews Using Bidirectional Autoencoder and Index Generator Algorithm

Abstrak

In the current digital era, student reviews of universities in Indonesia posted on social media often present significant challenges due to their non-standard sentence structures. These informal or fragmented constructions complicate the identification of aspects and sentiment analysis, resulting in reduced accuracy. To address these issues, an enhanced Aspect-Based Sentiment Analysis (ABSA) model that combines a Bi-directional Autoencoder, an index generator algorithm, and Indo-BERT, a language model fine-tuned specifically for Indonesian. The Bi-directional Autoencoder is crucial in improving the model's understanding of complex and non-standard sentence patterns, allowing it to capture the nuanced meanings in student reviews more effectively. The Index generator algorithm enhances the model by efficiently organizing and retrieving relevant data, streamlining the aspect and sentiment extraction process. Indo-BERT further strengthens the model's capability by accurately classifying reviews through better contextual understanding and sentiment polarity detection. The experimental results demonstrate that this hybrid model significantly improves accuracy and processing speed, outperforming traditional approaches. The model is particularly effective in extracting aspects, opinions, and sentiment polarities, providing deeper insights into student perspectives on universities in Indonesia. Despite some limitations, such as sensitivity to ambiguous or brief reviews, this model represents a substantial advancement in sentiment analysis techniques for the educational sector. Its application offers valuable potential for improving educational management and decision-making processes based on student feedback.

Identifier

• DOI: https://doi.org/10.18421/tem144-48
• OpenAlex: https://openalex.org/W4416793485