Once More, With Feeling: Using Sentiment Analysis to Improve Models of Relationships Between Non-State Actors

We collect an original corpus of official documents released by Hezbollah and use text analysis to create a measure of their relationships with other non-state groups. Despite recent research that demonstrates these types of relationships affect the length and severity of conflict, data limitations hinder efforts by researchers to capture important variation in them over time. Our approach uses fine-grained text data to capture dynamic trends in these relationships. We demonstrate its effectiveness by showing our model is able to reproduce qualitative accounts of Hezbollah's known alliances and rivalries with other non-state actors with greater accuracy and precision than existing measures. We also compare the approach with event data and demonstrate its ability to provide more granular and complementary information. With further exploration, this technique could assist researchers in improving and developing measures of intrastate cooperation and competition for use in empirical analyses.