questions which might need to be reasarched

Currently I am looking at various Text Classification methods and later on will
be delving into the area of Information Extraction.

Now there are many classification algorithms which weight word/text in
various manners leading to different accuracy results.
Now my question is what are the possible parameters/dimensions of the
input text which wud in a non-statistical way give you the most
appropriate classification method to use.
So one wud know exactly which classification method wud serve him best
for his particular domain or problem.