Stance Detection on Tweets with Multi-task Aspect-based Sentiment: A Case Study of COVID-19 Vaccination

ABSTRACT

Public opinion analyses on Twitter conducted based on sentiment analysis cannot identify the author’s stance regarding agreement or disagreement with a given target. Stance detection determines whether the author of a text is in favor, against, or neutral towards a target. However, stance detection based on text-only is less representative opinion, especially on a tweet, which is a short text with slightly contextual information. Therefore, more information is needed to represent the author's stance better. In previous research, most research on stance detection was carried out using simple sentiment information to measure the support to target. This study addresses multi-task aspect-based sentiment analysis (ABSA) and social features for stance detection based on deep learning models of BiGRU-BERT on tweets. Our contribution combines aspect-based sentiment information with features based on textual and contextual information that does not emerge directly from Twitter texts. ABSA approach can provide more accurate sentiment information at aspect level on tweets, which is possible contains multiple issues discussed. Aspect information on tweets can reflect the issue that influences the author’s stance toward a target. Multi-task learning was applied to help improve the generalization performance of ABSA with simultaneous processes. We extracted social attributes and online behavioral features for contextual information. Since same community tends to have the same opinion towards a target, we applied a community detection task and combine with the Twitter social attributes. The proposed method has significantly improved evaluation metrics (>10%) than textual features only for stance detection on tweets © 2022. International Journal of Intelligent Engineering and Systems.All Rights Reserved