Source-linked AI summary
Bots increase exposure to negative and inflammatory content in online social systems
Massimo Stella, Emilio Ferrara, Manlio De Domenico
TL;DR
The paper examines how social bots participate in polarization during the Catalan referendum. Using network, sentiment, stance, and semantic analyses of Twitter interactions, it finds that bots acted peripherally while targeting influential humans and promoting negative, inflammatory content.
Problem
Online social systems contain both humans and software-controlled bots, but the structural and emotional roles of bots in polarization during major events require clarification.
Method
The study analyzes Twitter interactions during the Catalan referendum using bot detection, network partitioning, sentiment analysis, and hashtag co-occurrence networks.
Results
The analysis identifies Independentists and Constitutionalists and finds that peripheral bots strategically target human hubs while promoting content matching each group’s emotional polarity.
Takeaways & Limitations
Bots can sustain polarized factions from the network periphery by increasing exposure to negative, hatred-inspiring, inflammatory content and potentially exacerbating online conflict.
Abstract
from arXiv · showhide
Societies are complex systems which tend to polarize into sub-groups of individuals with dramatically opposite perspectives. This phenomenon is reflected -- and often amplified -- in online social networks where, however, humans are no more the only players, and co-exist alongside with social bots, i.e., software-controlled accounts. Analyzing large-scale social data collected during the Catalan referendum for independence on October 1, 2017, consisting of nearly 4 millions Twitter posts generated by almost 1 million users, we identify the two polarized groups of Independentists and Constitutionalists and quantify the structural and emotional roles played by social bots. We show that bots act from peripheral areas of the social system to target influential humans of both groups, bombarding Independentists with violent contents, increasing their exposure to negative and inflammatory narratives and exacerbating social conflict online. Our findings stress the importance of developing countermeasures to unmask these forms of automated social manipulation.
Results
The analysis identifies two polarized referendum factions and shows that bots, despite peripheral network positions, strategically target human hubs and amplify negative content, especially among Independentists.
- Social activity: Humans and bots shared circadian activity patterns, with a dramatic increase in message volume on October 1.
- Social activity: Bots generated 23.6% of posts during the event, while Replies reached 38.8% bot-generated content.Retweets and Mentions showed comparable bot shares; the higher Reply share suggests targeted responses.
- Social activity: 19% of overall interactions were directed from bots to humans, mainly through Retweets (74%) and Mentions (25%).
- Emotional dynamics: Sentiment among human-to-human interactions fell from positive to negative after September 30, with negative content spreading on referendum day.The drop occurred after negative content began being reshared and reached a midday negative peak before moving toward neutrality.
- Polarized groups: A network-enhanced stance analysis combined social structure, emotional intensity, message semantics, exchange structure, and recipient type to identify Independentists and Constitutionalists.The method addresses the limitation that sentiment alone cannot identify whether users support or oppose an event.
- Network roles: Humans were 1.8 times more central than bots, yet bots strategically targeted highly connected human users.Bot-to-human in-degree correlated with human-to-human in-degree (Kendall Tau κ ≈0.62, p-value < 10^-4), and randomized networks lacked the observed correlations.
- Emotional dynamics: Human-to-human and bot-to-human interactions were negative in Group 1 and positive in Group 2, while bots promoted human-generated content with matching polarity.Group 1 preferentially endorsed negative content, linking bot activity to greater exposure to opposing parties’ negative material.
Discussion
The study finds that bots operate from the network periphery while targeting human influencers, directing negative and violent content toward Independentists. This automated activity increases exposure to inflammatory narratives and may exacerbate online social conflict.
- Bots sustain both factions from the network periphery by mainly targeting human influencers.
- Bots direct messages toward Independentists that evoke fight, violence, and shame against the government and police.
- Bots may accentuate exposure to negative, hatred-inspiring, inflammatory content, thereby exacerbating social conflict online.