The Spectrum of Sentiment: Optimistic, Pessimistic, and Neutral Voices in Online Depression Discourse

Abstract

The relationship between depression and the concepts of optimism and pessimism has been extensively researched by psychologists. In this paper, we use computational approaches to study how optimism and pessimism are expressed in the online discourse of people with a depression diagnosis. Publicly available datasets are used for the development of an optimism/pessimism detection model, as well as for the analyses performed on social media posts of individuals with depression, as measured by BDI-II, a validated depression questionnaire. To analyze the optimistic and pessimistic posts by individuals with depression, we use LIWC features and perform topic modeling. We also investigate specific words driving mislabeling using SHAP. Our results show that while there may not be significant differences in the number of optimistic versus pessimistic posts between individuals in the depression and control groups, the content of the posts differs meaningfully, both in terms of linguistic features and approached topics.

Resources

Details