Undergraduate research on how the source platform of a text corpus can bias topic models, and how contrastive learning can remove that bias (published at EMNLP 2024 Findings). Related work applied KoBERT to utterance-based classification tasks in dialogue/prediction systems.