Too many false positives

[tshprecher]

The classifier yields too many false positives. Some ideas on improving this:

1. Prune results that occur more than once in a file. It's reasonable to assume that typos will not occur more than once in file.
2. Prune results that occur more than once in all the files. Same as The set of english words is poor (missing words like "called" and "having"!) #1, except slightly more strict.
3. Perhaps instead of skipping the symbols in code, store them and exclude any references to them in comments.