Predicting Tie Strength with the Facebook API

Abstract

This paper presents a user study that employed a Facebook application to calculate the strength of Facebook users' friendships. Specifically, 18 variables were collected via the Facebook API for 1728 friendships and used to predict tie strength reported by 90 participants. The resulting model had an accuracy of 65.9% in differentiating between strong and weak ties, and 86.3% in differentiating between very strong and weaker ties. The tie-strength calculation was performed in real time by the application, conferring the key advantage that the result can be instantly used by the live application. We argue this functionality has the potential to enable many novel customization and recommendation scenarios. Furthermore, examining the effect of the use of different Facebook features and types of communication on the perceived tie strength gives a more comprehensive understanding of the concept of tie strength in social media and sheds light on people's use of social network sites.