Four Types of Toxic People: Characterizing Online Users' Toxicity over Time

Raghvendra Mall, Mridul Nagpal, Joni Salminen, Hind Almerekhi, Soon Gyo Jung, Bernard J. Jansen

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Identifying types of online users' toxic behavior reveals important insights from social media interactions, including whether a user becomes "radicalized" (more toxic) or "pacified" (less toxic) over time. In this research, we design two metrics to identify toxic user types: F score that captures the changes in a user's toxicity, and G score that captures the direction of the shift taking place in the user's toxicity pattern. We apply these metrics to a dataset of 4M user comments from Reddit by defining four toxic user types based on the toxicity scores of a user's comments: (a) Steady Users whose toxicity scores are steady over time, (b) Fickle-Minded Users that switch between toxic and non-toxic commenting, (c) Pacified Users whose commenting becomes less toxic in time, and (d) Radicalized Users that become gradually toxic. Findings from the Reddit dataset indicate that fickle-minded users form the largest group (31.2%), followed by pacified (25.8%), radicalized (25.4%), and steadily toxic users (17.6%). The results suggest that the most typical behavior type of toxicity is switching between toxic and non-toxic commenting. This research has implications for preserving the user-friendliness of online communities by identifying continuously toxic users and users in danger of becoming radicalized (in terms of their toxic behavior), and designing interventions to mitigate these behavior types. Using the metrics we have defined, identifying these user types becomes possible. More research is needed to understand why these patterns take place and how they could be mitigated.

Original languageEnglish (US)
Title of host publicationNordiCHI 2020 - Proceedings of the 11th Nordic Conference on Human-Computer Interaction
Subtitle of host publicationShaping Experiences, Shaping Society
PublisherAssociation for Computing Machinery
ISBN (Electronic)9781450375795
DOIs
StatePublished - Oct 25 2020
Event11th Nordic Conference on Human-Computer Interaction: Shaping Experiences, Shaping Society, NordiCHI 2020 - Virtual, Online, Estonia
Duration: Oct 25 2020Oct 29 2020

Publication series

NameACM International Conference Proceeding Series

Conference

Conference11th Nordic Conference on Human-Computer Interaction: Shaping Experiences, Shaping Society, NordiCHI 2020
CountryEstonia
CityVirtual, Online
Period10/25/2010/29/20

All Science Journal Classification (ASJC) codes

  • Human-Computer Interaction
  • Computer Networks and Communications
  • Computer Vision and Pattern Recognition
  • Software

Fingerprint Dive into the research topics of 'Four Types of Toxic People: Characterizing Online Users' Toxicity over Time'. Together they form a unique fingerprint.

Cite this