A Link to the Past: Constructing Historical Social Networks

摘要

To assist in the research of social networks in history, we develop machine-learning-based tools for the identification and classification of personal relationships. Our case study focuses on the Dutch social movement between 1870 and 1940, and is based on biographical texts describing the lives of notable people in this movement. We treat the identification and the labeling of relations between two persons into positive, neutral, and negative both as a sequence of two tasks and as a single task. We observe that our machine-learning classifiers, support vector machines, produce better generalization performance on the single task. We show how a complete social network can be built from these classifications, and provide a qualitative analysis of the induced network using expert judgements on samples of the network.