Can Smartphone Co-locations Detect Friendship? It Depends How You Model It

TL;DR

This study explores using smartphone co-location data to detect friendships and their strength, demonstrating a 30% improvement over random chance with robust models across different settings.

Contribution

It introduces a comprehensive feature set and validation schema for detecting friendships from smartphone data, highlighting the importance of modeling choices.

Findings

Co-location features can predict friendships with 30% accuracy above chance.

The model remains robust across different dyads and over time.

A novel validation schema improves assessment of model performance.

Abstract

We present a study to detect friendship, its strength, and its change from smartphone location data collectedamong members of a fraternity. We extract a rich set of co-location features and build classifiers that detectfriendships and close friendship at 30% above a random baseline. We design cross-validation schema to testour model performance in specific application settings, finding it robust to seeing new dyads and to temporalvariance.