Mine important contacts from an address book

I'm designing an algorithm to mine around 200 important contacts from a list of say 2000 or more. First factor I am considering is call history and second is commitment to call (or sales targets). Data for both is available from the client's current system.

Now we are trying to think of things, data of which if available, that would make this mining of important contacts even better. Make the system more "int开发者_如何学Goelligent" perhaps. (I wasn't able to find any papers related to this, maybe I'm using non-standard terminology?)

Update: To clarify, I'm looking for ideas which can be used alongside call history (frequency) and call targets. (Open to a different approach altogether too.)

---

It is pretty simple, you assign values to different factors that you consider important. For example 10 for call history, 20 for commitment, 30 for something else. Now, you involve this value to your count, if a person is contact 100 times, his value would be 100 * 10 and similar with others values. And then you have two item object, (name, value) and sort it with value and take top 200.

---

May be you can look at how Google arranges most frequently, most often, contacted friends on the top on GMail's chat friends list. I think it is called the default contacts group. Not that I am advertising for Google here, it just came on top of the mind.

---

So you need ranking of you contacts. There several approaches

• one is building function f(contact)

  • most simple case is linear combination f(contact) = a*feature1(contqact) + b*feature2(contact) i.e. logistic regression
  • nonlinear function f(contact) = func(feature1, feature2, etc) i.e SVM, neural network // both approaches can be trained using learning to rank optimisation

• domination i.e you can't convert data to double but you can define (a <= b) and then find top 200 by partial ordering sorting algorithms