Beliefs are connected within systems that exist both in individuals and in collectives, but the structures that emerge at these two levels do not always align. The collective patterns seen in public discourse could not fully represent the diversity and change that occur within individual belief systems. This work-in-progress examines this tension by constructing individual networks of beliefs from comments across three politically diverse news providers (Motherjones, The Hill, Breitbart). Using LLM-based stance detection, we assume that when people discuss multiple topics in the same comment, a link between them in their belief space can be inferred. Early results reveal topological differences between individual networks and aggregate ones in terms of nodes, degree distributions, and clustering coefficients. Going beyond simple topic detection, this approach aims to reveal how opinions structurally connect and evolve for each individual through time.

Please note that this talk is intended for internal participants only. If you are an external guest and would like to join, kindly send an email to .