Unveiling Aspects in Environmental and Energy Discussions: A Confidence-Weighted Approach

This study presents a refined approach to Aspect-Based Sentiment Analysis (ABSA), leveraging a pretrained BERT model within the FASTATPC framework on posts and their associated comments sourced from environmental and energy-focused subreddits. Rather than limiting itself to the most frequent aspect-sentiment pairs, the method incorporates model confidence by computing a weighted score that multiplies the relative frequency of each sentiment by its average confidence. This ensures that both occurrence and certainty guide the selection of the most salient aspect-sentiment pairs. Results show a notable divergence between the sentiment profiles of posts and comments (with negative sentiment at 44.6% in posts vs. 72.2% in comments), indicating that user interaction on comments is significantly more critical or skeptical. Additionally, integrating confidence scores reorders the top aspects, revealing nuances in topics like pollution, climate policies, and renewable energy that might otherwise be overlooked when relying purely on frequency. Overall, these findings foster a more comprehensive understanding of environmental discourse on Reddit, highlighting how both sentiment frequency and model certainty shape public perception.