Academic research has always been important to Idiro – that’s how we became world leaders in the application of Social Network Analysis (SNA) techniques to telecommunications business problems.  It is also why our CEO is on the board of CeADAR, the Irish Centre for Applied Data Analytics Research.

So we are happy to report that our colleague Davide Cellai, who has been working on advanced uses of SNA in solving telecommunications business problems, this week gave a seminar at the University of Aalto, Finland on his advanced SNA churn research.  Davide says:

“Last Thursday I was invited to give a seminar at the University of Aalto, near Helsinki. This week, I have been hosted by the group of Jari Saramaki and Kimmo Kaski, who were possibly the first researchers to focus a research group in social network analysis applied to telecommunications, several years ago. I presented our work on port-out churn, plus some percolation models of robustness of infrastructures I have been involved with in recent years.  Here are some examples:

Social Network Analysis in telecommunications - research by Idiro Analytics
Distribution of fraction of time two subscribers spend speaking to each others in an interval of 12 weeks

In order to profile the type of connection between pairs of subscribers, we calculate the amount of time two individuals spend at peak (i.e. during working hours) and off-peak time of the day. For each pair, we can evaluate if conversations occur mainly at peak or off-peak time and give a score accordingly. In the first figure, we show the distribution of this score over all the phone calls that occurred in a period of 12 weeks. We can see two strong peaks, representing calls occurring exclusively off-peak and at peak times, respectively. We also show an enlargement of the central zone, with a broad hill representing pairs of subscribers who speak both at peak and off-peak time. This polarization suggested us to identify three layers of acquaintances: peak, off-peak, and mixed peak/off-peak.

Social Network Analysis in telecommunications - research by Idiro Analytics
Probability that a subscriber churns as a function of churning subscribers in her network of social acquaintances
This plot shows the probability that a subscriber churns after m_out of her friends have churned in a recent time interval. Generally speaking, we can see that a subscriber with more churning friends is more likely to churn, as the red crosses tend to grow with m_out. Moreover, we can see that the probability of churning is higher for the mixed peak/off-peak layer (purple boxes), meaning that at this level of relationships, churning propagates more easily.
The seminar participants were quite interested in the way we could identify types of social acquaintance based on the time of calls. They also suggested that exposure to churn in terms of duration, instead of number of churners, would improve the sensitivity of the method. In particular, Jari Saramaki, who has also experience in the data analytics industry, envisions that machine learning methods should be fed with this kind of insightful social network information to produce best lift.

The week I have spent here has been very useful. For example, researchers have shown me a method to identify families that doesn’t use any community finding algorithm, and a way to map a temporal network into a network of events, that can be treated with a known formalism. Another post-doc is working with psychiatrists to detect the onset of a mental disorder in the pattern of social activities of a person. Finally, it has been also interesting to hear that a few students or post-docs are starting a company. Best of luck, and thank you Aalto!”

Idiro gratefully acknowledges the support of Science Foundation Ireland in this work.  Idiro works with telecommunications companies across the world, helping them with customer retention, customer acquisition etc.  To learn more about Idiro’s work on Social Network Analysis in telecommunications, or to find out how Idiro helps telcos to get better marketing results through our SNA models, contact our experts.

Recommended Posts