Search

Social media has become integral to societal discourse and play a role in shaping public engagement, particularly in democratic electoral processes. This paper addresses the pressing issue of hate speech on social media during the 2022 US midterm elections. Unlike previous research, which often relies on limited datasets and classic methodologies, we leverage Open Source Intelligence (OSINT) and Natural Language Processing (NLP) techniques to analyze Twitter data through advanced models of entity recognition, sentiment analysis, and community extraction, having persistence in Knowledge Graphs for consuming the intelligence efficiently. Results indicate that in the US midterm elections 2022, Arizona was the state that provided more content (507,551 tweets) related to a Chief Electoral Official, with 31.58% of them identified in the most aggressive cluster due to its mean attribute values of "attack on commenter" (0.7), "inflammatory" (∼ 0. 3), "attack on author" (∼ 0. 2), and "toxicity" (∼ 0. 2). The name entity recognition model also identified an association between those aggressive tweets and the previous 2020 US Presidential campaign, characterized by attacks on election officials based on conspiracy theories campaigns. Knowledge graphs contributed to understanding the concentration of attacks and connectivity between topics commonly mentioned in hate speech content. Thus, our results offer detailed insights into the actors and dynamics of online harassment in electoral contexts, illuminating the challenges posed by harassment and proposing preventive mechanisms applicable to diverse electoral processes worldwide. • Daily online violence affects vital processes like elections via hate speech. • This study analyzes 2022 US midterms' hate speech on Twitter using NLP. • Our methodology is modular and scalable for large-scale real scenarios. [ABSTRACT FROM AUTHOR]