TWeddit : A Dataset of Triggering Stories Predominantly Shared by Women on Reddit

TL;DR

This work tackles the lack of labeled trigger-warning data in real-world online discourse by introducing TWeddit, a dataset of 5000 Reddit stories from ~22 subreddits annotated with seven trigger categories related to pregnancy, abortion, miscarriage, and sexual violence. The authors deploy an active-learning pipeline that combines few-shot prompting with LLMs (e.g., GPT-4o-mini) and cosine-similarity retrieval to label data efficiently, achieving a macro-F1 improvement over a fanfiction-based baseline (XLNet achieving ~0.763). They perform extensive exploratory analyses, including BERTopic-based topic modeling, DAMF moral foundations, and NRC emotion profiling, revealing distinct linguistic and emotional signatures across trigger categories. TWeddit enables future direction in automatic trigger-warning generation, cross-domain transfer, and safer moderation in online communities, while addressing ethical considerations and data accessibility.

Abstract

Warning: This paper may contain examples and topics that may be disturbing to some readers, especially survivors of miscarriage and sexual violence. People affected by abortion, miscarriage, or sexual violence often share their experiences on social media to express emotions and seek support. On public platforms like Reddit, where users can post long, detailed narratives (up to 40,000 characters), readers may be exposed to distressing content. Although Reddit allows manual trigger warnings, many users omit them due to limited awareness or uncertainty about which categories apply. There is scarcity of datasets on Reddit stories labeled for triggering experiences. We propose a curated Reddit dataset, TWeddit, covering triggering experiences related to issues majorly faced by women. Our linguistic analyses show that annotated stories in TWeddit express distinct topics and moral foundations, making the dataset useful for a wide range of future research.

Figures (7)

• Figure 1: Active learning loop where model predictions on unlabeled data are filtered using perplexity-based uncertainty, reviewed, corrected, and added back to the labeled set for iterative k-shot learning.
• Figure 2: Pearson correlation matrix across trigger warning categories, showing positive associations (Medical-Pregnancy and Abuse-Sexual) and negative associations (Pregnancy-Abuse).
• Figure 3: Topic word clouds for three dominant trigger categories : Mental Health, Pregnancy, and Abuse, highlighting the most salient terms in each category and illustrating distinct thematic language patterns across triggers.
• Figure 4: With respect to harm and degradation foundations, moral differences between triggering and non-triggering stories. Here, TS stands for Triggering Stories, whereas NTS stands for Non Triggering Stories.
• Figure 5: Emotion distribution across trigger warning categories based on mean NRC emotion density (anger, fear, sadness, disgust, anticipation, trust, joy, and surprise).