On the Behaviour of Deviant Communities in Online Social Networks

Mauro Coletto, Luca Maria Aiello, Claudio Lucchese, Fabrizio Silvestri

Abstract

Online social networks are complex ensembles of inter-linked communities. While "deviant" communities—those centered on behaviors like adult content consumption—are often assumed to be isolated niches, this study demonstrates they are deeply integrated into the broader social fabric. By analyzing Tumblr’s network, we find that content from a small core of producers reaches at least 450 times more users than the producers themselves, largely through "bridge" communities. This integration leads to significant involuntary exposure among the general population. We also identify core nodes whose removal could radically disrupt this diffusion process.

Introduction

The prevailing assumption in both sociology and computer science has been that deviant online communities operate as secluded islands, well-separated from mainstream social media. This "isolation hypothesis" suggests that users interested in topics like drug use or adult content stay within "dark" corners of the web. However, this study challenges that notion, arguing that because users belong to multiple communities simultaneously, deviant content naturally spills over into the general population.

Methodology and Data Collection

The researchers focused on adult content on Tumblr, employing a multi-step pipeline to map the deviant network. They developed a semi-supervised algorithm that began with a seed list of keywords and used a seven-month search engine query log (146 million entries) to identify relevant Tumblr URLs.

Convergence of the iterative expansion process for identifying deviant nodes, query log entries, and dictionary size.

As shown in Figure 5, this iterative process expanded the keyword dictionary and blog list until the network converged. The final dataset included over 130 million nodes and nearly 7 billion interactions. The researchers then mapped the "Follow" network (static interest) and the "Reblog" network (active diffusion), using the Louvain algorithm for community detection.

Network Structure and the 'Bridge' Phenomenon

The analysis revealed that the deviant network is not a monolith but a structured system of producers and intermediaries.

Visualization of the deviant network structure, highlighting core producer communities and the bridges that connect them.

As illustrated in Figure 6, the network consists of two massive "Producer" cores (primarily heterosexual and male-homosexual content) surrounded by "Bridge" communities. These bridges act as aggregators, reblogging content from the core to a much wider audience. These bridge communities are orders of magnitude more effective at spreading content than the average user, serving as the primary engines of "spillage" into the mainstream.

Key Insights and Impact

The reach of deviant content is exponential. While core producers represent only 0.7% of the network, their content reaches 450 times more users. This leads to several critical phenomena:

  • The Majority Illusion: In the reblog network, 40% of users see deviant material from more than 10% of their contacts, making the behavior seem more pervasive than it is globally.
  • Involuntary Exposure: An estimated 40 million users are potentially exposed to deviant content unintentionally by following active consumers who reblog that content into mainstream feeds.
  • Demographic Risks: While producers are typically older males (average age 38), consumers are younger and more gender-balanced, increasing the risk of exposure for underage users.

Conclusion and Targeted Intervention

The study highlights the fragility of this diffusion network. Because the process is top-heavy, targeted interventions are highly effective. Removing or marking posts from just the top 5,000 most influential nodes can curb total diffusion by over 50%. To protect underage users specifically, removing only the top 200 nodes would significantly clear their feeds. Future work should expand this analysis to other platforms and deviant behaviors to determine if these "bridge" structures are a universal feature of online social dynamics.

Summary updated

Abstract

On-line social networks are complex ensembles of inter-linked communities that interact on different topics. Some communities are characterized by what are usually referred to as deviant behaviours, conducts that are commonly considered inappropriate with respect to the society’s norms or moral standards. Eating disorders, drug use, and adult content consumption are just a few examples. We refer to such communities as deviant networks. It is commonly believed that such deviant networks are niche, isolated social groups, whose activity is well separated from the mainstream social media life. According to this assumption, research studies have mostly considered them in isolation. In this work we focused on adult content consumption networks, which are present in many on-line social media and in the Web in general. We found that few small and densely connected communities are responsible for most of the content production. Differently from previous work, we studied how such communities interact with the whole social network. We found that the produced content flows to the rest of the network mostly directly or through bridge-communities, reaching at least 450 times more users.We also show that a large fraction of the users can be inadvertently exposed to such content through indirect content resharing. We also discuss a demographic analysis of the producers and consumers networks. Finally, we show that it is easily possible to identify a few core users to radically uproot the diffusion process. We aim at setting the basis to study deviant communities in context.
BibTeX
@inproceedings{coletto2016behaviour,
  author = {Mauro Coletto and Luca Maria Aiello and Claudio Lucchese and Fabrizio Silvestri},
  title = {{On the Behaviour of Deviant Communities in Online Social Networks}},
  booktitle = {ICWSM},
  year = {2016},
  url = {https://openreview.net/forum?id=00uDt5emRk},
}
Figure