Metrics of datasets?

Question

It is quite common to discuss what metrics to be used for a model/algorithm/method. But when it comes to metrics of datasets, it is much less discussed.

I wonder, what are the general metrics for datasets? That is to say, how does one quantitively measure if a dataset is good or not for its purpose?

I understand, there may not be too many general metrics, given the purpose and nature of datasets differ vastly. So, more specifically, I am interested in the follow few cases:

what are the metrics for 1) text datasets, 2) classification datasets, 3) unbalanced classification datasets?

Any knowledge, insights, interesting ideas are welcome and appreciated.

Answer 1

For unbalanced datasets, please refer to http://www.cs.gmu.edu/~hrangwal/kd-hcm/proc/papers/3-Perez-Baranauskas.pdf

Answer 2

Datasets are usually considered as input upon which we wish to perform some task (e.g. classification, clustering, ...). A 'purpose', is therefore not generally considered for a dataset, but instead on the model which is applied to the dataset.