[Paper Review] Commenter Behavior Characterization on YouTube Channels

Shadi Shajari, Nitin Agarwal+1 more|arXiv (Cornell University)|Apr 16, 2023

Misinformation and Its Impacts10 citations

TL;DR

The paper presents a social network analysis-based method to detect commenter mobs on YouTube, characterize channels by suspicious commenting behavior, and identify coordination across channels, validated on a large YouTube dataset.

ABSTRACT

YouTube is the second most visited website in the world and receives comments from millions of commenters daily. The comments section acts as a space for discussions among commenters, but it could also be a breeding ground for problematic behavior. In particular, the presence of suspicious commenters who engage in activities that deviate from the norms of constructive and respectful discourse can negatively impact the community and the quality of the online experience. This paper presents a social network analysis-based methodology for detecting commenter mobs on YouTube. These mobs of commenters collaborate to boost engagement on certain videos. The method provides a way to characterize channels based on the level of suspicious commenter behavior and detect coordination among channels. To evaluate our model, we analyzed 20 YouTube channels, 7,782 videos, 294,199 commenters, and 596,982 comments that propagated false views about the U.S. Military. The analysis concluded with evidence of commenter mob activities, possible coordinated suspicious behavior on the channels, and an explanation of the behavior of co-commenter communities.

Motivation & Objective

• Motivate the study by the prevalence of suspicious commenting and its impact on discourse quality in YouTube comments.
• Develop a methodological framework to detect commenter mobs and quantify suspicious behavior at the channel level.
• Characterize channels based on the level of suspicious commenter activity.
• Detect coordination among channels through co-commenter networks.
• Provide empirical evidence of mob activities and co-commenter communities on a large YouTube dataset.

Proposed method

• Proposes a social network analysis-based methodology to identify commenter mobs.
• Characterizes channels by the level of suspicious commenter behavior.
• Detects coordination among channels via analysis of commenter interactions.
• Applies the model to a dataset comprising 20 channels, 7,782 videos, 294,199 commenters, and 596,982 comments.
• Evaluates whether mobs propagate false views about the U.S. Military and analyzes co-commenter communities.

Experimental results

Research questions

• RQ1 Can commenter mobs be detected on YouTube channels using social network analysis?
• RQ2 How can channels be characterized by the level of suspicious commenter activity?
• RQ3 Is there evidence of coordination among channels through co-commenter interactions?
• RQ4 What are the patterns and dynamics of co-commenter communities in relation to false-view propagation about specific topics?

Key findings

• Evidence of commenter mob activities across analyzed channels.
• Indications of possible coordinated suspicious behavior among channels.
• Insights into the behavior of co-commenter communities surrounding the discussed topic.
• Large-scale dataset analysis supports the methodological framework for detecting mobs and coordination.