Extracting Aspect-Based Economic Sentiments From Thai Social Media Text

The pervasive impact of economics can be observed in online discussions on social media. Studies have shown that the ability to analyze social media text's overall sentiment or sentiment of a specific aspect is crucial for monitoring and predicting economic phenomena such as inflation and GDP. However, sentiment and aspect-based sentiment analyses in social media text composed in Thai are understudied mainly due to the lack of large-scale labeled datasets. This research proposes using the pairing encoding sentence for pre-trained language models (PLM) that utilize the Aspect Companion technique and transformer models to predict social media text's overall sentiment and aspect-category sentiments related to economics. Specifically, the proposed approach can train a language model that learns all aspect categories altogether. The results show that the Aspect Companion technique combined with WangchanBERTa outperformed other models in identifying polarity in overall economic sentiment and aspect-category sentiment analyses.