Predictability of social interactions

TL;DR

This paper investigates the predictability of social interactions, revealing that while individual interaction patterns are highly predictable, the identities of interaction partners are less so, and simple models can effectively predict future interactions.

Contribution

It demonstrates that social interaction patterns are predictable but the identities of interaction partners are less predictable, and a simple Markov model performs near the upper bound.

Findings

Interaction locations and times are highly predictable.

The identity of interaction partners is less predictable.

A simple Markov model effectively predicts future interaction partners.

Abstract

The ability to predict social interactions between people has profound applications including targeted marketing and prediction of information diffusion and disease propagation. Previous work has shown that the location of an individual at any given time is highly predictable. This study examines the predictability of social interactions between people to determine whether interaction patterns are similarly predictable. I find that the locations and times of interactions for an individual are highly predictable; however, the other person the individual interacts with is less predictable. Furthermore, I show that knowledge of the locations and times of interactions has almost no effect on the predictability of the other person. Finally I demonstrate that a simple Markov chain model is able to achieve close to the upper bound in terms of predicting the next person with whom a given individual will interact.