2
$\begingroup$

I'm interning in an auditing company and have audit reports that require categorising. I have been using simple text mining + classification techniques in R (DocumentTermMatrix in tm package, SVM via e1071 package, etc.) to help.

Unfortunately it is laborious to manually categorise the issues to create the train data, but as of now I have about 50+ samples categorised into about 7 categories. A shockingly small number, I know.

What then is the best way to continue? Is 50 too small a size for training data? I have about 150 reports that requires categorising. The reports are each roughly about 900 - 2000 chars (1-2 pages) long (hence laborious to manually categorize).

$\endgroup$
2
  • $\begingroup$ Honestly, I'd stick to categorizing them by-hand, or preparing some heuristic rules to categorize them (e.g. existence of some particular terms in documents etc). $\endgroup$
    – Tim
    Commented Jul 21, 2017 at 7:33
  • $\begingroup$ yes, i've thought about heuristics (e.g. certain terms coming up) but it is hard to come up with rigid rules because everyone writes differently and the presence of a term (e.g. "inaccurate") does not determine the report's category (e.g. "inaccurate report") $\endgroup$
    – hongsy
    Commented Jul 21, 2017 at 7:37

1 Answer 1

1
$\begingroup$

Did you try latent semantic analysis? Maybe you will be able to find meaningful patterns when you visualize the data in 2 or 3 dimensions.

If you have mostly unlabeled data then maybe you can leverage it using semi-supervised approach, for example label propagation.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.