How does Apple find dates and times in emails?

stackoverflow.com

108 points·elchief··9 comments

I created a (now defunct) product called Inbox Assistant which would parse plain text emails and create events automatically in your calendar.

I had to figure out how to do essentially what Apple does, and I'm by no means a computer scientist. Here's what I learned: http://blog.myinboxassistant.com/post/9613650483/experimenti...

Link should point here: http://www.miramontes.com/writing/add-cacm/

Note this is from around 1998. And it seems heavily grammar based.

Things might have changed, since the hardware has become more powerful, and some statistical techniques have proven pretty good at Named-Entity recognition.

It's possible that the current ones are still based on the work of Apple's Advanced Technology Group. Apple used HTC over its patents and this article about the suit mentioned the technology being added back in 10.5:

http://forums.appleinsider.com/showthread.php?t=138975

Apple's Advanced Technology Group had a few projects that dealt with processing text to discover information structure. Jim Miller who was part of ATG has written a few articles about the work they did:

Data Detectors: http://www.miramontes.com/writing/add-cacm/

LiveDoc: http://www.miramontes.com/writing/livedoc/

DropZones: http://www.miramontes.com/writing/dropzones/

Oops ... didn't notice zaroth's earlier link http://news.ycombinator.com/item?id=3633077 to the Miramontes articles, sorry for the dupe.

The timex module in NLTK contrib does a fairly good job at that but uses regular expressions

https://github.com/nltk/nltk_contrib/blob/master/nltk_contri...

This is called Timex tagging. See http://www.timeml.org/ etc.

I believe DARPA runs some sort of competition every year or so where different algorithms are submitted to extract temporal data from text.

from the stackoverflow link, helpful video on machine learning: http://videolectures.net/mlas06_nigam_tie/

The 14 min mark alludes to the question above.

lsm(1) - Latent Semantic Mapping tool

Reply on news.ycombinator.com