Applying Sentiment Analysis Techniques in Social Media Data About Threat of Armed Conflicts Using Two Times Series Models

Introduction

The evolution of the internet has enabled the advent of the social media as one of the main means of circulation of personal, political, and public information. As such, there is a need for the use of modern techniques such as machine learning and sentiment analysis to assist in accurate verification of specific information among these enormous volumes of data. One such areas of the society that calls for a more profound analysis of its causes and consequences is social extreme events, such as armed conflicts. An extreme event is a sequence of small events generated by human emotions or some reactions of nature that can evolve into a major event and even a catastrophic event. Armed conflicts are social extreme events that are part of the history of the human development (Sornette, 2006). Armed conflicts, within this context, are sequences of threats followed by attacks until they reach their climax with an armed conflict or war. All the problems generated by armed conflicts, call the attention to a solution that helps in the analysis, prediction, and possible alert of the population for a reduction of the damages that such events usually cause (Ibañez, et al., 2022). In this project three models of armed conflicts are used as case studies of social extreme events: the armed conflict between Syria and the USA, the armed conflict between Iran and the USA, and the armed conflict between Russia and Ukraine.

The study of the dynamics of the process of triggering an armed conflict is an area that has been analyzed for decades in many ways. One of the great scholars in this area, Lewis Richardson, has several approaches to the analysis of armed conflicts that cover different models: game-theoretic models, evolutionary games and agent-based models (ABMs), differential equations (DEQ) models, and statistical analyses of time-structured data (Richardson, 1960; Gleditsch, 2020). Thus, this chapter applies an approach using the concepts of sentiment analysis and machine learning to perform an analysis that considers the emotions contained in social media texts as a possible source of the beginning of armed conflicts. In this case, it analyzes more specifically the emotion of threat between heads of states involved in the conflicts addressed.

For information collection, the web search engine Google (Google LLC, 1998) and a chatbot (Lateral GmbH, 2019) are used to performs a search for news related to the topics addressed in this study, which are social and political threats. Each news is collected, stored, and grouped in ascending order according to its publication date.

As for the context, development to an approach of analysis and prediction of threat variation of endogenous social extreme events will be using information collected from news website and social media, such as (Reuters, 2019) (CNN, 2020) (The Guardian, 2020). In processing the news collected from social media, the technique of sentiment analysis is employed to enable the identification of human emotions present in the texts. Sentiment analysis makes it possible to identify similarities of a text to its given context by using a base text with words or vocabularies referring to and representing a domain (Bird et al., 2009). In this case, the domain is the threat of these extreme events (Ibañez, Rosa, & Guimarães, Sentiment Analysis Applied to Analyze Society's Emotion in Two Different Context of Social Media Data, 2020). Thus, each news piece is collected, analyzed, and identified for the percentage of threat existing in its text (Ibañez et al., 2022).