Creating a custom categorized corpus in NLTK and Python

Question

I'm experiencing a bit of a problem which has to do with regular expressions and CategorizedPlaintextCorpusReader in Python.

I want to create a custom categorized corpus and train a Naive-Bayes classifier on it. My issue is the following: I want to have two categories, "pos" and "neg". The positive files are all in one directory, main_dir/pos/*.txt, and the negative ones are in a separate directory, main_dir/neg/*.txt.

How can I use the CategorizedPlaintextCorpusReader to load and label all the positive files in the pos directory, and do the same for the negative ones?

NB: The setup is absolutely the same as the Movie_reviews corpus (~nltk_data\corpora\movie_reviews).

Litman

see #29276114 – Alameda

Answer 1

Here is the answer to my question. Since I was thinking about using two cases I think it's good to cover both in case someone needs the answer in the future. If you have the same setup as the movie_review corpus - several folders labeled in the same way you would like your labels to be called and containing the training data you can use this.

reader = CategorizedPlaintextCorpusReader('~/MainFolder/', r'.*\.txt', cat_pattern=r'(\w+)/*')

The other approach that I was considering is putting everything in a single folder and naming the files 0_neg.txt, 0_pos.txt, 1_neg.txt etc. The code for your reader should look something like:

reader = CategorizedPlaintextCorpusReader('~/MainFolder/', r'.*\.txt', cat_pattern=r'\d+_(\w+)\.txt')

I hope that this would help someone in the future.

Litman