UC Riverside

\begin{abstract}How can we recognize users' cognition and identify behaviors in an online discussion forum? We see that online discussion forums constitute an untapped opportunity for understanding cognitional and behavioral information. Mining this publicly and freely available information can significantly benefit analysts as it helps reveal trends, behavior, and even bad actors. This thesis aims to answer the following problems. First, we identify and characterize thread-centric behaviors where the key novelty lies in an unsupervised model to recognize behaviors without requiring prior forum knowledge. The model reveals some fascinating abusive behaviors appearing in the forum. Second, we develop an aspect-based sentiment analysis model, a powerful state-of-the-art transformer-based model to detect sentiment toward specific aspects in posts. The model also helps quantify the effect of the real-world event on users' sentiment in the online forum. Third, we develop a stance detection model to recognize the user's position toward topics of interest and quantify the correlation of sentiment and stance conditioning to the events. Our finding on the relationship between sentiment and stance redefines how an analyst perceives this cognitional information. The contribution of our work can be summarized in threefold: (a) collect, analyze, and profile thread-based behaviors, (b) detect sentiment toward specific topics in response to real-world events, and (c) infer cognitional information and understand the relationship of sentiment and stance at the events. We see our systematic approaches and tools as a significant step towards cognitional and behavioral understanding in online discussion platforms.