Visual Network tool: Individual belief networks from LLM-guided interviews and a visual canvas

Belief systems are theorized on individual and collective levels, but are typically studied and inferred only at the collective level. We introduce a new tool to measure individual-level belief networks from open-ended conversations, using a visual canvas. This ``Visual Network'' tool addresses several limitations of existing measures and enables fast and scalable measurement of the differences in content and structure of individual networks. Participants complete a semi-structured interview about their beliefs and behaviors on a given issue guided by an LLM, evaluate LLM-extracted statements that summarize the beliefs they have expressed, and construct personal belief networks by creating supporting and conflicting relations between these statements on a visual canvas. We validate the tool in a two-wave study on meat-eating. Participants evaluate the experience positively, and both belief network content and structure show satisfactory reliability across the two measurement points. We also find support for face, concurrent, and criterion validity of the measured individual belief networks. The new ``Visual Network'' tool opens up new empirical possibilities to test theories of belief dynamics and belief change.