Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The influence of coordinated behavior on toxicity

Edoardo Loru, Matteo Cinelli, Maurizio Tesconi, Walter Quattrociocchi

TL;DR

This study investigates how Coordinated Behavior (CB) relates to toxic discourse on X around the 2019 UK general election using a large-scale dataset (11,264,280 tweets from 1,179,659 users) and toxicity scores from Perspective API. CB is identified via a superspreader framework built on retweet-based similarity and clustered with Louvain methods, while political leaning is inferred through a seed-driven hashtag propagation approach. The findings show that strongly coordinated users tend to share less toxic content overall, with cluster-specific patterns and a nuanced effect depending on whether content is original or retweeted; in particular, original CB content can be more toxic, but CB interactions generally do not markedly increase non-coordinated toxicity. Temporal analyses reveal higher toxicity peaks aligned with campaign events, suggesting that CB primarily serves amplification and influence, not straightforward toxicity dissemination, and that content nature and political framing can outweigh coordination in shaping online toxicity.

Abstract

In the intricate landscape of social media, genuine content dissemination may be altered by a number of threats. Coordinated Behavior (CB), defined as orchestrated efforts by entities to deceive or mislead users about their identity and intentions, emerges as a tactic to exploit or manipulate online discourse. This study delves into the relationship between CB and toxic conversation on X (formerly known as Twitter). Using a dataset of 11 million tweets from 1 million users preceding the 2019 UK general election, we show that users displaying CB typically disseminate less harmful content, irrespective of political affiliation. However, distinct toxicity patterns emerge among different coordinated cohorts. Compared to their non-CB counterparts, CB participants show marginally higher toxicity levels only when considering their original posts. We further show the effects of CB-driven toxic content on non-CB users, gauging its impact based on political leanings. Our findings suggest that CB only has a limited impact on the toxicity of digital discourse.

The influence of coordinated behavior on toxicity

TL;DR

This study investigates how Coordinated Behavior (CB) relates to toxic discourse on X around the 2019 UK general election using a large-scale dataset (11,264,280 tweets from 1,179,659 users) and toxicity scores from Perspective API. CB is identified via a superspreader framework built on retweet-based similarity and clustered with Louvain methods, while political leaning is inferred through a seed-driven hashtag propagation approach. The findings show that strongly coordinated users tend to share less toxic content overall, with cluster-specific patterns and a nuanced effect depending on whether content is original or retweeted; in particular, original CB content can be more toxic, but CB interactions generally do not markedly increase non-coordinated toxicity. Temporal analyses reveal higher toxicity peaks aligned with campaign events, suggesting that CB primarily serves amplification and influence, not straightforward toxicity dissemination, and that content nature and political framing can outweigh coordination in shaping online toxicity.

Abstract

In the intricate landscape of social media, genuine content dissemination may be altered by a number of threats. Coordinated Behavior (CB), defined as orchestrated efforts by entities to deceive or mislead users about their identity and intentions, emerges as a tactic to exploit or manipulate online discourse. This study delves into the relationship between CB and toxic conversation on X (formerly known as Twitter). Using a dataset of 11 million tweets from 1 million users preceding the 2019 UK general election, we show that users displaying CB typically disseminate less harmful content, irrespective of political affiliation. However, distinct toxicity patterns emerge among different coordinated cohorts. Compared to their non-CB counterparts, CB participants show marginally higher toxicity levels only when considering their original posts. We further show the effects of CB-driven toxic content on non-CB users, gauging its impact based on political leanings. Our findings suggest that CB only has a limited impact on the toxicity of digital discourse.
Paper Structure (11 sections, 5 figures, 3 tables)

This paper contains 11 sections, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Data
  4. Coordinated communities
  5. Political leaning inference
  6. Results and Discussion
  7. Toxicity across coordinated communities
  8. Toxicity in coordinated and non-coordinated users
  9. Interactions with coordinated users and toxic production
  10. Limitations
  11. Conclusions

Figures (5)

  • Figure 1: Empirical Cumulative Distribution Function (ECDF) of the number of tweets posted by superspreaders compared to other users, taking into account original tweets (left panel) and retweets (right panel).
  • Figure 2: Joint distributions of different metrics for the superspreaders in the three largest communities of the similarity network. Color intensity indicates the number of users in a bin, with red regions highlighting peaks. (a) Coordination score and toxicity of users, with the dashed line (median coordination) indicating the threshold used to label a superspreader as "coordinated"; (b) toxicity of users and weighted average of their neighborhood's toxicity; (c) cluster-normalized (least to most extreme) political leaning of users and their toxicity; (d) political leaning of users and weighted average of their neighborhood's leaning.
  • Figure 3: Comparison between coordinated and non-coordinated users in terms of expressed toxicity, defined as the average of the top 10% most toxic original tweets or retweets. (a) Tweeting activity and user toxicity smoothed via a GAM fit (the shaded region indicates the corresponding 95% CI), with the observed user toxicity distribution for both groups in miniature; (b) bootstrap distribution of the average user toxicity; (c) bootstrap distribution of the average user toxicity obtained by ignoring retweets.
  • Figure 4: Average toxicity of tweets produced by non-coordinated users, obtained with bootstrap resampling. (a) Distributions of the average toxicity produced following exclusive interactions with non-coordinated users or coordinated users; (b) estimates with their 95% CI obtained by factoring in the toxicity of the interactions, using a score of 0.6 as the threshold to label a tweet as 'toxic', and their political leaning.
  • Figure 5: Hourly average toxicity of the tweets produced by non-coordinated users compared with that of the tweets by coordinated users they have interacted with, overlaid with LOESS curves (the shaded regions indicate their respective 95% CI) for easier visualization. The size and visibility of a point are proportional to the number of tweets observed within the corresponding hour. On the right side, empirical distributions of the two metrics across the entire time period.