esl

NLP update

Apr 21, 2015 12:36

I made some progress understanding how to do feature extraction from text ( Read more... )

Leave a comment

Comments 17

dennisgorelik April 21 2015, 20:21:52 UTC
What problem would it solve?

Reply

esl April 21 2015, 21:31:58 UTC
Problem: computers do not understand text. However, they do understand images (or at least they understand images better than they understand text). So I'm trying to apply image understanding methods to text.
p.s. I have no idea if this is going to work at all, but it's worth a shot!

Reply

dennisgorelik April 21 2015, 23:01:07 UTC
1) Actually computers understand text much better, than images.
E.g. Google text search is way more efficient than Google Image search.

2) If you do not understand how this is going to work at all, then it probably does not worth a shot.

Reply

esl April 22 2015, 02:43:38 UTC
I'm talking about text summarizing, or text recognition, similar to image recognition, which is done through identifying common patterns (features), in a hierarchical manner. For example, if you read an article, or a short story, you can then tell me what is it about. Computers can do this already for images, but not yet for text (there are attempts to do text summarizing, but they are pretty weak).

I didn't say I don't know how to do it - my post outlines the approach. I said I don't know if this is going to work.
In machine learning, people often don't know how/why something works. Usually a lot of trial and error, and brute force search is involved before working parameters are found. It's similar to neuroscience - we know a lot about brain, but at the same time we don't "really" know how it works.

Reply


Leave a comment

Up