Place: Centre de Convencions Internacional de Barcelona, Barcelona, Spain
Abstract
Human events are events that directly involve individuals, communities, societies, or humanity as a whole. Human events are often influenced by factors such as the economy, public policies, and civil unrest. Recently, machine learning methods including graph neural networks as well as (large) language models have been developed to forecast and interpret temporal human events from online data sources such as social networks, news articles, and personal blogs. Advancing AI in this subject can enhance decision-making in social science, public health care, and governance. Workshops for predictive analytics have been organized to educate participants regarding various challenges in human event modeling across diverse sectors involving business, finance, healthcare, and government. Human event modeling focuses on forecasting that estimates future events based on historical data. Interpretation of events seeks to identify explainable factors for the predictions to understand the underlying mechanisms of events.
In this tutorial, we present recent and ongoing advances in human event modeling. We start by introducing how events are formulated into graphs with auxiliary features including text and time series. Next, we discuss graph learning models, a prominent research area in deep learning focusing on modeling data with structural information. Additionally, we explore the insights gained from using language models in human event modeling. Finally, we discuss challenges and identify future opportunities in the field.
We want to emphasize the timeliness and significance of this tutorial. Human event modeling has a long history, dating back to the 19th century. However, the recent rise of large language models has changed the space of the problem. We anticipate significant advancements in the near future, making this tutorial timely and relevant.
Outline (tentative)
Introduction and motivation [40]
Societal event prediction
Challenges in human event prediction
Explainable event prediction
Early approaches
Methodology
Part 1: Graph Neural Network (GNN)-based methods [55]
Vanilla graph learning
[Break - 15]
Graph learning with contextual information
Graph learning with causal reasoning
[Break - 15]
Part 2: Large Language Model (LLM)-based methods [50]