Related Experiment Video
Updated: Sep 2, 2025

Integrating Computerized Linguistic and Social Network Analyses to Capture Addiction Recovery Capital in an Online Community
Published on: May 31, 2019
Increased Online Aggression During COVID-19 Lockdowns: Two-Stage Study of Deep Text Mining and
Jerome Tze-Hou Hsu1,2, Richard Tzong-Han Tsai1,3
1Center for Geographic Information Science, Research Center for Humanities and Social Sciences, Academia Sinica, Taipei, Taiwan.
Lockdowns during the COVID-19 pandemic significantly increased aggressive online behaviors, including anger, offensive language, and hate speech. This study used natural language processing to confirm a causal link between lockdown measures and heightened online aggression.
Area of Science:
- Social Psychology
- Computational Linguistics
- Public Health
Background:
- The COVID-19 pandemic necessitated global lockdowns, impacting public health and social stability.
- Lockdowns may negatively affect mental health, potentially increasing aggressive emotions and behaviors.
- Understanding the link between lockdowns and aggression is vital for policy development.
Purpose of the Study:
- To investigate the relationship between lockdown measures and increased aggression.
- To analyze anger, offensive language, and hate speech in US-based tweets using natural language processing (NLP).
- To determine the spatiotemporal impact of lockdowns on online aggressive emotions.
Main Methods:
- A longitudinal study analyzed 1,281,362 tweets from 11,455 users between 2019 and 2020.
- A Bidirectional Encoder Representations from Transformers (BERT) model was trained to detect anger, offensive language, and hate speech.
- Difference-in-differences estimation was employed to assess the causal impact of lockdown status on aggressive tweets, controlling for seasonal and regional factors.
Main Results:
- Aggression levels increased in the first six months of 2020 compared to 2019, particularly among users under lockdown.
- A statistically significant positive correlation was found between lockdown status and increased anger (P=.002), offensive language (P<.001), and hate speech (P=.005).
- The analysis suggests a causal relationship between lockdowns and heightened online aggression.
Conclusions:
- Findings highlight a causal link between lockdown measures and increased online aggression.
- NLP and big data analysis of social media provide timely insights for policymakers.
- Understanding these impacts is crucial for mitigating the adverse societal effects of public health interventions.
Related Concept Videos
Bullying
Aggression
Group Polarization
Deindividuation
Relationship Formation
Social Facilitation

