Predicting Social Dynamics: Understanding Human Relationships through Conversations

In a groundbreaking study, researchers used natural language processing tools to analyze conversational text from Reddit users, emails from the Enron corpus, and dialogue from movie scripts to explore how fundamental dimensions of human relationships are expressed in everyday language. By annotating conversational text through crowdsourcing, the team trained their tools to detect the presence of ten fundamental relationship dimensions, and found that they could be predicted purely from conversations with up to 98% accuracy.

Their results also suggest that the combination of these dimensions can reveal the types of relationships people have (conflict vs. support) and the types of communities they shape (wealthy vs. deprived). This research has significant implications for understanding how humans interact, and could be used to inform fields such as psychology, sociology, and linguistics.

Read the research: Ten social dimensions of conversations and relationships