Hawkes-modeled telecommunication patterns reveal relationship dynamics and personality traits


Abstract in English

It is not news that our mobile phones contain a wealth of private information about us, and that is why we try to keep them secure. But even the traces of how we communicate can also tell quite a bit about us. In this work, we start from the calling and texting history of 200 students enrolled in the Netsense study, and we link it to the type of relationships that students have with their peers, and even with their personality profiles. First, we show that a Hawkes point process with a power-law decaying kernel can accurately model the calling activity between peers. Second, we show that the fitted parameters of the Hawkes model are predictive of the type of relationship and that the generalization error of the Hawkes process can be leveraged to detect changes in the relation types as they are happening. Last, we build descriptors for the students in the study by jointly modeling the communication series initiated by them. We find that Hawkes-modeled telecommunication patterns can predict the students Big5 psychometric traits almost as accurate as the user-filled surveys pertaining to hobbies, activities, well-being, grades obtained, health condition and the number of books they read. These results are significant, as they indicate that information that usually resides outside the control of individuals (such as call and text logs) reveal information about the relationship they have, and even their personality traits.

Download