Data Science and Analytics Thrust Seminar | User-Centric Research in Social-Information Systems – "It’s Not Just What You Know, But Who You Know"

Language-related data have been extensively studied in Natural Language Processing (NLP) and Large Language Models (LLM). However, our recent studies demonstrate that people-related data are also valuable and, in some cases, critical for predicting and modeling social dynamics.

For example, the prediction of a person’s reaction to a received message has traditionally relied primarily on language-related data. Our research shows that incorporating people-related data—such as a person’s interests, activity patterns, and interactions with others—can significantly enhance the accuracy of these predictions. More importantly, when predicting a person’s reaction to a message on a new, unseen topic, it is essential to rely solely on people-related data and exclude language-related data (i.e., the content of messages) to avoid erroneous results.

We also find that it is possible to classify a person’s affiliation or opinion based exclusively on their social networking data. For instance, we can identify the political affiliation of politicians by analyzing their “following” relationships on X (formerly Twitter), without considering any user profile information or the content of their posts. This approach may lead to a promising new method for opinion measurement.

We call for increased study and attention to the networking aspects of social information systems. After all, a person’s decision or opinion is influenced not only by the content of the information they receive but also by the source of that information and their relationship with it.