Not all friendships are equal, and many applications — friend recommendation, feed ranking, viral marketing — need to know which ties are close. Most existing measures are ad hoc: interaction counts, common neighbors, embedding similarity. This work instead starts from social identity theory, the sociological account of how people's group memberships shape their relationships.

The measure

The key idea is that closeness between two users is reflected in the alignment of their social identities — the overlapping communities and roles they inhabit — combined with the intensity of direct interaction. The paper operationalizes this into a computable closeness score over large graphs and shows how to evaluate it efficiently.

Validation

On real social platforms (including Tencent data with ground-truth intimacy signals), the identity-based measure identifies close friends more accurately than interaction-only and structure-only baselines, and improves downstream tasks that consume tie strength.