Temporal Features as Measures of Tie Strength in Mobile Phone Networks
Perustieteiden korkeakoulu | Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Master’s Programme in Life Science Technologies
AbstractThe use of auto-recorded communication data, such as mobile phone call logs, has reshaped our capacity to model and understand of social systems. In such studies, the strength of a tie between two people has been of great value from both theoretical and sociological perspectives, yet it is not easy to quantify. Tie strengths are commonly measured in terms of communication intensity (number or duration of calls, etc) as a form of convenience rather than a justified choice, yet these intensity-based measures do not uncover the myriad of ways in which such intensity takes place, hindering information about the strength of ties. Here, we conceive tie strength as a latent variable we want to predict based on features of the time sequences of interactions. We assume that tie strength is expressed as the structural overlap in social networks, in a manner inspired by Granovetter's hypothesis, where strong ties are embedded in community structures, while weak ties serve as inter-community bridges. With this assumption, we use temporal and static features to predict overlap in lieu of the latent tie strength. We analyze a mobile phone dataset of ~6.5 million people for a period of 4 months, and measure overlap based on an extended network of ~77 million users, to ensure minimal sampling errors. We observe a strong relationship between local topology and tie-level behaviour, with some temporal features outperforming communication intensity in overlap prediction. Indeed, the number of bursty cascades, differences in daily behaviour and temporal stability play large roles in our models. We find that communication intensity is one of many characterizations of tie strength for which the Granovetter effect is observable.
Thesis advisorKivelä, Mikko
complex networks, social networks, tie strengths, temporal patterns, granovetter effect