Electoral violence is an important concept, but measuring its connection to elections has been challenging. This paper introduces a novel method using convolutional neural networks (CNNs) on social media data to measure electoral violence during three specific election events. Our approach demonstrates significantly improved accuracy compared to existing measurement models and aligns better with theoretical expectations about this political conflict than current event-based datasets like ACLED, ICEWS, or SCAD. The validity of our methodology relies on supporting qualitative coding.
### Key Findings:
Methodological improvement: CNN analysis achieves >30% greater accuracy in identifying electoral violence related to elections. Alignment with theory: Our measure better captures theoretical expectations for electoral violence than existing data sources.
| We Need to Go Deeper: Measuring Electoral Violence using Convolutional Neural Networks and Social Media was authored by David Muchlinski, Xiao Yang, Sarah Birch, Craig Macdonald and Iadh Ounis. It was published by Cambridge in PSR&M in 2021. |
