Modeling Precursors for Event Forecasting via Nested Multi-Instance Learning

Abstract: Forecasting events like civil unrest movements, disease outbreaks, financial market movements and government elections from open source indicators such as news feeds and social media streams is an important and challenging problem. From the perspective of human analysts and policy makers, forecasting algorithms need to provide supporting evidence and identify the causes related to the event of interest. We develop a novel multiple instance learning based approach that jointly tackles the problem of identifying evidence-based precursors and forecasts events into the future. Specifically, given a collection of streaming news articles from multiple sources we develop a nested multiple instance learning approach to forecast significant societal events across three countries in Latin America. Our algorithm is able to identify news articles considered as precursors for a protest. Our empirical evaluation shows the strengths of our proposed approaches in filtering candidate precursors, forecasting the occurrence of events with a lead time and predicting the characteristics of different events in comparison to several other formulations. We demonstrate through case studies the effectiveness of our proposed model in filtering the candidate precursors for inspection by a human analyst.