Measures of Importance: a Network of UNGA Speech Mentions

United Nations General Assembly (UNGA) speeches are an under-explored source of information on government policy preferences over time. Unlike UNGA votes, each member state is able to address the issues that it deems important, irrespective of an exogenously set agenda. In this paper, I measure the salience that States and other actors have in the international system over time by identifying the number of times they are mentioned by UN member states in any given year between 1970 and 2020, as well as the tone in which they were mentioned. To this effect, I rely on a combination of natural language processing and social network analysis. This results in a ranking of the most mentioned countries in a given year and a list of edges which enumerates the number of times a country i mentioned a country j in a year t. Furthermore, I classify each of these mentions into three categories: neutral, positive, and negative, using Naive Bayes algorithm in order to exclude non-meaningful mentions. Subsequently, I perform a fixed-effects regression analysis where the dependent variable is the number of, positive and negative, mentions of an actor in a given year to determine which are the correlates that predict changes its salience. This approach can capture a higher degree of granularity on State preferences and stances than comparable existing data sets, as mentions data shows higher levels of variation over time than alliances or diplomatic relations.