BERT-based Ensembles for Modeling Disclosure and Support in Conversational Social Media Text

There is a growing interest in understanding how humans initiate and hold conversations. The affective understanding of conversations focuses on the problem of how speakers use emotions to react to a situation and to each other. In the CL-Aff Shared Task, the organizers released Get it #OffMyChest dataset, which contains Reddit comments from casual and confessional conversations, labeled for their disclosure and supportiveness characteristics. In this paper, we introduce a predictive ensemble model exploiting the finetuned contextualized word embeddings, RoBERTa and ALBERT. We show that our model outperforms the base models in all considered metrics, achieving an improvement of in the F1 score. We further conduct statistical analysis and outline deeper insights into the given dataset while providing a new characterization of impact for the dataset.
View on arXiv