Inferring friendship network structure by using mobile phone data
This article is about a study involving 94 subjects that had software installed on their mobile phones that allowed researchers to obtain data about call logs, bluetooth devices, cell tower IDs, application usage, and phone status. All the participants in the study were students and faculty at a major research institution. This article was published in 2009 so I’m sure our phones are having its data collected by a number of companies. The data from the phones was studied to try to find patterns of friendship. One of the challenges of this is that there are many factors that go into what type of relationship you have with a person. For example, spending a large amount of time with someone on a Wednesday afternoon is very different than spending a large amount of time with someone on a Friday night. Additionally, 2 people can be friends even if they do not interact with each other for a certain period of time. The interactions were broken down into on/off campus and daytime/nighttime. These 2 factors do a good job of breaking up interactions. By just using the daytime/evening factor, the researchers were able to predict 96% of symmetric reports of non-friendship and 95% of symmetric reports of friendship. This means that the time of day we interact with people is very strongly correlated with how much we view them as a friend. The people we spend time with at night are generally people we would consider friends which makes sense.
I found this article interesting and it reminded me of the 6 degrees of separation we talked about in class and the friendship networks. 6 Degrees of Separation is based off the facebook social network but it would be interesting to see if all the text and call history ever was put into a graph. I think we might see that the degrees of separation between 2 people may be much lower than 6. Additionally, while the article did not mention it, it might be interesting to see the effects of robo calls in the context of friendship networks. I think this would be very similar to spam accounts on facebook. The study from the article is a few years old now and only involved a few participants so the data is not quite as interesting as a very large scale network. However, I think it would be very interesting similar to this one was done at a very large scale. I think some very interesting trends would show. Furthermore, I found it interesting how easy it was to predict friendships. I think think that if more participants with much different backgrounds were used, predicting friendships would be much more difficult. Also, it would be interesting to see the role that something like machine learning would have in predicting friendships. I think that a neural net would find some very interesting trends in human friendships.
https://www.pnas.org/content/106/36/15274