python re match, findall or search and then NLP (what to do with it?)

Question

I am starting to write code that would capture part of sentence "types" and if they match a criteria, start a specific python script that deals with the "type." I am "finding":) that findall kind of works better for what i am doing hence:

m = re.compile(r'([0-9] days from now)')
m.match("i think maybe 7 days from now i hope")
print m.match("i think maybe 7 days from now i hope")
None
f= m.findall("i think maybe 7 days from now i hope")
print f[0]
7 days from now

This seems to give me the part of sentence that i was looking for. I can then give this to for example - the pyparsing module using its example datetime conversion script that returns a datetime from a similar NL statement (I know there are other modules but they are rigid in input statements they can handle) .
Then I could do a db insert into my online diary for example or on a hosted web app if other parts of the sentence matched another "type" ie. appointments, deadlines etc.
I am just tinkering here but slowly i am building something useful. Is this structure /process logical or are there better methods/ ways: that is what i am asking myself now. Any feedback is appreciated

Answer 1

The reason why m.match() fails is that it expects the match to start at the beginning of the string.

findall() makes sense if you expect more than one (non-overlapping) match in your string. Otherwise, use the search() method (which will return the first match it finds).

This is all well covered in the docs .

Answer 2

From my knowledge of search interfaces, it seems like you'd need an awful lot of regular expressions to capture the great variety of ways in which people express themselves. For a feeling for just how many, see this writeup on "the vocabulary problem".

So, if you're just doing date/time stuff, and you're tying very specific actions to them that it would suck to get wrong, then RE's seem like a good way to go. On the other hand, if you're just trying to detect a "date" expression vs. eg an "email" expression or a "note" expression, then perhaps it might be worth a try to POS-tag the sentences using NLTK and match patterns on the part of speech level.