Narrowing and Recovery of Aspect-sentiment Diversity in Online Discussions
Abstract Online discussions can keep growing while the range of evaluations they express contracts. We study this tunnel vision, the progressive concentration of a discussion on a smaller set of aspect-sentiment positions, and ask whether narrowing is transient or lasting and whether receding positions return. We develop a framework that links frozen, thread-specific aspect schemas and aspect-conditioned sentiment labels to whole-thread diversity trends, local contraction episodes and the identities of the positions expressed, and apply it to 2,967 event-centred Reddit threads from April 2019 drawn from a processed corpus of 6,128 threads. Observed tunnel vision occurs in 45.9% of threads, but only 9.3% narrow more than expected under a null that preserves labelled mention density. Local contractions are common (73.8% of threads) and usually reversible (83.7% of first episodes recover), whereas this density-robust minority recovers less often (60.9% versus 86.6%) and, after recovering, more often contracts again. Onset-aligned trajectories show that the groups diverge after onset rather than in their initial decline. Recovery brings back previously expressed positions and a more even distribution among continued positions, with small changes in first-observed pairs. Richer openings are associated with higher final diversity despite slightly steeper subsequent declines, separating relative narrowing from a narrow endpoint. An independent April 2018 sample reproduces these patterns. The results establish recovery as a central dimension of conversational diversity and provide empirical targets for explaining how online discussions sustain evaluative breadth.
Authors
- Quan Bai (ORCID: https://orcid.org/0000-0003-1214-6317)
- Shiqing Wu (ORCID: https://orcid.org/0000-0001-6785-1203)
- Minh Nguyen
- Xiaodan Wang (ORCID: https://orcid.org/0009-0008-2159-2339)
- Yanbin Liu
- Weihua Li
- Jianing Xia
- Yining Wang
Institutions
- University of Technology Sydney (AU)
- University of Tasmania (AU)
- The University of Melbourne (AU)
- Auckland University of Technology (NZ)
- City University of Macau (MO)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-16
- DOI
- https://doi.org/10.21203/rs.3.rs-11486267/v1
- Primary Topic
- Sentiment Analysis and Opinion Mining
- Type
- preprint