Abstract
This study investigates how platform architecture influences toxic behavior by tracking "common users" who maintain profiles on both Instagram and Ask.fm. By performing a within-subject comparison, the researchers isolate the effects of platform-specific features—such as anonymity and interaction style—from inherent user personality. The findings reveal an "anonymity paradox" where non-anonymous interactions often exhibit higher negativity due to defensive "upstanding" or affectionate profanity, while also confirming that Ask.fm generally hosts higher levels of negativity than Instagram.
Introduction
Cyberbullying remains a critical societal challenge, yet its study is often hampered by the "siloed" nature of social media research. Most existing literature focuses on single platforms, making it difficult to discern whether toxic behavior stems from a user’s personality or the environment provided by the platform. This paper addresses this gap by identifying a subset of ~8,000 "common users" active on both Instagram (a non-anonymous, media-centric site) and Ask.fm (a semi-anonymous, text-centric Q&A site).
Methodology
The researchers identified common users by finding Ask.fm profiles that explicitly linked to Instagram IDs. To analyze behavior across these platforms, the study employed:
- Sentiment Analysis: A dictionary-based approach utilizing 1,500 positive and 1,500 negative words, including internet slang.
- Linguistic Analysis: Using LIWC to categorize posts into psychological dimensions like anger and social orientation.
- Statistical Distribution: Complementary Cumulative Distribution Functions (CCDFs) were used to compare the frequency and intensity of behaviors (posting, liking, following) between "normal" and "common" user groups.
The Anonymity Paradox and Sentiment Trends
A primary finding of the research is that anonymity on Ask.fm does not inherently lead to higher negativity. In many cases, non-anonymous comments were more negative. This is attributed to "upstanding," where friends defend victims by attacking bullies non-anonymously, and "affectionate profanity," where close friends use aggressive language as a sign of intimacy.
Despite the higher baseline of negativity on Ask.fm, positivity remains the dominant sentiment on both platforms. As shown in the sentiment distributions, the probability of encountering high levels of positivity significantly outweighs that of negativity across all user groups.
Cross-Platform Dynamics
While underlying user traits remain consistent—positive users on one platform tend to be positive on the other—the platform architecture significantly shifts the baseline of behavior. Ask.fm users generally exhibit higher negativity (~6.5%) compared to their behavior on Instagram (~3.5%).
Furthermore, the interaction style plays a role: Ask.fm’s Q&A format creates a high correlation (0.7) between the sentiment of a question and its answer, suggesting a feedback loop. In contrast, Instagram’s one-to-many media-centric model shows much lower sentiment correlation in comment threads.
Key Results and Conclusion
- Activity Gaps: Common users are more active and popular on Ask.fm than normal users, but conversely less active on Instagram than the average user.
- Popularity Metrics: On Instagram, there is a strong correlation (0.89) between receiving likes and receiving comments, though the number of media objects posted does not strongly predict popularity.
- Conclusion: The study demonstrates that while individual character provides a baseline for behavior, platform features like the Q&A format and the social norms surrounding anonymity can amplify or mitigate toxic interactions. Future work should aim to distinguish between "affectionate profanity" and genuine bullying through manual labeling and ground-truth datasets.
Summary updated