This repository is to supplement the paper "SABER : A Multi-task Framework Combining Sentiment and Behavioral Cues".
As the severity of climate change intensifies, understanding public attitudes toward this pressing issue is increasingly crucial for developing effective mitigation strategies. Social media platforms have emerged as influential tools for shaping and reflecting public opinion on climate change. However, the discourse on these platforms is deeply polarized, divided between those who acknowledge the urgency of the situation and those who maintain a skeptical stance. This stark divide underscores the necessity for disseminating accurate information and developing effective climate change mitigation strategies, making stance detection a vital task in this domain. In response to this need, this study introduces a novel multi-task framework, SABER, which jointly performs stance detection and sentiment analysis on climate change-related tweets. By employing various embedding techniques and attention frameworks, SABER leverages sentiment aspects learned from the data to create a comprehensive representation of the features relevant to the stance expressed in a given tweet. Extensive experiments conducted on a curated climate change dataset demonstrate the effectiveness of our approach, highlighting the importance of considering user interaction patterns and sentiment information in recognizing stance. The dataset and code are publicly available, providing a valuable resource for further research in this critical area.
Figure 1 : The overview architecture of our proposed multi-task framework SABER
TBD
Our dataset comprises 61,247 climate change-related tweets collected from Twitter between January 1, 2022, and December 31, 2022, using specific denier and believer hashtags. Each tweet is labeled for stance detection as either "believe" or "deny" based on the used hashtags, with 49,006 tweets identified as denying climate change and 12,241 as believing. In addition, we assign sentiment labels in four categories: positive, negative, and neutral, using the VADER.
TBD