Thought Leaders in Big Data: Interview with Matti Aksela, VP of Analytics at Comptel (Part 4)

Sramana Mitra: I fully understand the information of who called whom, for example, is also proprietary information for the carrier and the carrier can do whatever they want with it. That is not exactly social network behavior. It is more about information about transactions happening within the customer base. What does that tell you? What correlations are you drawing there?

Matti Aksela: We look at a huge stream of events. From the calling behavior, you can isolate what point in time somebody makes calls, how many calls are being made between subscribers, etc. So you can derive an intensive communication between particular subscribers. Some variables that are of much interest are how many calls are being made to the operator’s own network and how many calls have been made to other networks, or how many subscribers are moving from one network to another. We are really focused on the machine learning side here. We would calculate 1,000 variables – just to give a number – that we derive from things we see on the network. Then we look at how deeply connected subscribers are and how many connections they have, how dense the network is around them, etc.

SM: There is one thing that is confusing me and as a result it is confusing the audience. When you talk about social network, are you talking about the social network or social graph map you are creating based on the customer base and their actions, or are you talking about social networks like Facebook?

MA: My apologies for that. I should have been more careful about that from the start. It is about the network we calculate from the call detail records, not using social media.

SM: That is very different. Then there is no issue about data privacy, because the carrier owns all that data. So there is no problem using and drawing from that graph.

MA: Especially when it comes to providing a better service and helping the subscribers. There have been studies that most subscribers of mobile operators are almost expecting to be contacted if they have problems, for example. This increases customer satisfaction and experience.

SM: Let’s say you have all information about who is calling whom and who is sending text messages to whom, etc. Give me one or two examples of the heuristics you are able to derive from this behavior.

MA: They are twofold. On the one hand we aim to provide better understanding of subscribers to mobile operators – how they are behaving, how they are using certain services. From my background and way of thinking, that is actually more of a side product of what we are doing. What we really want to do is being able to access the very large amount of information and extract meaningful features for predictive analytics. From that we would extract all kinds of information derived from the even stream of data that is happening on the network.

We would then look to build variables from this information to be fed into what we consider one of our core assets – machine-based predictive analytics modules – where we look at a certain use case. We would build a predictive model to see that the target value is true and then we would want to see how we can best predict phenomena happening in the future. Essentially, that is the definition of predictive analytics.

This segment is part 4 in the series : Thought Leaders in Big Data: Interview with Matti Aksela, VP of Analytics at Comptel
1 2 3 4 5