I'm using K-means for unsupervised learning, using data vectors with an IP address and a language. I need to represent them in an abstract way, so that I can use this algorithm.
If your data doesn't map in an obvious way onto vectors of continuous real numbers, don't use K-means.
To use K-means clustering you need to have a well defined distance between any two points in the space. To do that you need to have a notion of distance for each axis.
Typically discrete classifications like "language" don't have any natural mapping onto real numbers.