Combining Named Entity Recognition and Emotion Analysis of Tweets for Early Warning of Violent Actions

Natural Language Processing techniques have gained popularity for the analysis of social media content. A number of techniques have been proposed to analyze various aspects such as public opinion, sentiments, and emotions expressed, opinion leaders, and extreme views. However, existing approaches take a retrospective approach that studies opinions after the occurrence of events. With the buildup of negative sentiments and extreme public opinion potentially leading to violent actions and civil disobedience, there is a need for a proactive and predictive approach that can offer early warning signs to government officials to intervene. In this work, we propose such an approach by combining two natural language processing techniques: Named entity recognition (NER) and emotions analysis. By tagging important entities within posts, such as prominent figures and important locations, and analyzing whether the tweets mentioning important entities carry negative emotions such as anger or violence, we are able to give insights about potential violent actions. Our framework was built and tested using 1290 tweets related to the 2020 US presidential election and the related US Capitol attack. The results obtained are promising and open the door for early intervention and appropriate preparedness for violent actions that may ensue from the buildup of negative sentiments and public views.