<![CDATA[This study analyzes the polarization of public opinion regarding the removal of fuel subsidies in Indonesia using a graph-based text classification approch that integrates Pointwise Mutual Information (PMI) and Graph Convolution Networks (GCN). One of the main challenges in sentiment analysis of social media data lies in capturing contextual and semantic relationships within short and noisy texts, which often limits the performance of conventional classifications methods. A dataset consisting of 602 tweets was preprocessed and represented as a graph, where word-word relationships were weighted using PMI and document-word connections were weighted using TF-IDF. Two model configurations were evaluated: a baseline GCN without PMI and a PMI-enhanced GCN, both trained using a 70:15:15 train-validation-test split with indentical hyperparameter settings. Experimental results indicate that the baseline GCN achieved a maximum accuracy of 65%, while the integration of PMI improved performance, yielding the highest the accuracy of 73%. These results demonstrate that incorporating PMI enriches graph representations and enhances the effectiveness of GCN in sentiment classification of political opinion data from social media.]]>
Copyrights © 2026
