Analisis Sentimen Masyarakat terhadap Isu Korupsi Dana Bencana di Indonesia Menggunakan Metode Bidirectional Long Short-Term Memory (Bi-LSTM)

Abstract

Corruption of disaster relief funds and social assistance is a critical issue that undermines social justice and public trust in government integrity in Indonesia. This phenomenon has triggered a massive wave of opinions on social media, necessitating deep computational analysis to objectively understand public perception dynamics. This study aims to implement and evaluate the performance of a Deep Learning algorithm, specifically Bidirectional Long Short-Term Memory (Bi-LSTM), in classifying public sentiment related to the issue of disaster fund corruption. The dataset comprises 1,358 textual data points categorized into negative, neutral, and positive sentiments, with a significant dominance of the negative class (926 entries). The proposed model architecture integrates a 300-dimensional embedding layer, a Bi-LSTM layer to capture bidirectional context, and a combination of Global Max Pooling and Global Average Pooling for optimal feature extraction. The experimental results demonstrate that the model achieved an accuracy of 0.75, with a Weighted F1-score of 0.76 and a Macro F1-score of 0.65. Confusion Matrix analysis reveals that the model is highly effective in identifying negative sentiments but faces challenges in distinguishing minority classes due to data imbalance and linguistic ambiguities such as sarcasm. These findings provide deep insights for policymakers regarding public sentiment and demonstrate both the potential and limitations of the Bi-LSTM method in processing informal and sarcastic Indonesian text within the context of political and corruption discourse.

Keywords: Sentiment Analysis, Bi-LSTM, Disaster Fund Corruption, Deep Learning, Natural Language Processing