Media reporting during armed conflict

Media reports are a key source of information on armed conflicts. In fact, the main data providers on political contention rely on news articles when manually coding events. In addition, the relationship between news reporting and the formation of public opinion has long been acknowledged. Despite this dependency on media, there is little systematic knowledge on the validity of news reporting during armed conflict, going beyond common concerns of lacking data reliability in the field. News reporting on armed conflict might vary substantially across platforms, as most outlets are leaning towards a certain political position. The political stance of a newspaper is likely to effect in what matter reporting takes place, including which events are emphasized and what language is used. In addition to differences across media outlets, news reporting on armed conflict might vary over time due to certain key events on the ground or changes in the international community as a whole. In this paper, a sentiment analysis is conducted using news articles from varying outlets in the context of the 2023 Israel-Hamas war. The goal is to investigate differences in sentiment between news providers and as the conflict unfolds over time. The analysis distinguishes between varying underlying political positions, comprising BBC News, Al Jazeera, and the Times of Israel. For comparability, only text published as blog posts in the live reporting is considered. Using a bag of words approach, sentiment is estimated using the VADER classifier, which yields a sentiment score for each day in the time series of news reports. Analyzing how sentiment differs between news outlets, techniques from time series clustering are applied to obtain recurring patterns in subsequences of sentiment scores. Using the k-Means algorithm and Dynamic Time Warping, time series of sentiment are matched based on similarity, which reveals recurring shapes specific to each news provider. Comparing shapes and the average distance of sentiment sequences within and across news outlets allows uncovering differences in reporting. Investigating how sentiment changes over time, a break point detection analysis is conducted, which serves to validate whether certain key events in the armed conflict significantly alter the sentiment of the news reporting within and across outlets.