[Research] Data extracted from cell phone use

In The Feature, an interview of Tomi T. Ahonen, a mobile phone guru from Finland. It's marketing-oriented (know your client stuff/market segmentation and all this sort of stuff) but he deals with the large amount of data generated by cell phone uses, collected by phone operators.

in the mobile business, we have perfect information on what a person does with their device every minute. And because we get immediate indications from our systems to suggest when that behavior is changing, we can also study how customer segments evolve. (...) The problem for the mobile operator is that there is too much data. Theoretically, we could hire a staff of sociologists, psychologists and statisticians to go through everything and try to build some profiles, but you'd only get the tip of the iceberg. (...) Operators tried to solve this with data mining (...) Some types of profiles can be developed that way, but it's usually one-dimensional. We can find out things like which customers provide the biggest amount of money for us, or which ones are most likely to remain loyal, but the data being collected from a user has about 60 different parameters.

The he thinks that one solution could be to use neural network techniques like Self Organizing Maps (hehe...Teuvo Kohonen, another finnish). This is indeed a trend and some folks at France Telecom also works on it.

here is a type of mathematics called neural networking. And there's something called the "self-organizing map" (SOM). SOMs will find patterns in enormous amounts of data. The SOM itself is stupid: it has no idea why this part of the map is green, and that corner is yellow, and that part in the middle is red. You need analysts to ask: "Why did the SOM say that this part is green and this part is yellow? Ah! These people are now very heavy users between each other, and these people have only a few different contacts, but there's a massive amount of traffic between them. These people here receive traffic, but don't really originate much." Recognizing the differences between what makes one part of the map a different color to another is where we need the human input. We come up with an absolutely fantastic understanding of real-life usage patterns. (...) One of the early things that came out was the concept of the "Alpha User" — a person who teaches everyone else how to use communication technology. The Alpha User is that person who taught you how to send a text message or a picture message. Someone either sat with you and taught you how, or else you are an Alpha User. The Alpha User concept is about two years old. (...) is kind of data has only now become available. When we start to analyze this we notice a pattern that doesn't fit the conventional wisdom. All at once, we get a really deep understanding of what subscribers are doing, what we offer them and how we can then move them further along.

BTW, I hate this 'segment' notion, are people lines? :) But it's really interesting, it's the second time SOM spots on my radar in the mobile research community and I am pretty confident this is a trend (deriving a trend from two spots is an error but knowing the potential of SOM...). Let's have a look if I can use this for CatchBob analysis...