0

I am working on a project in which I need to find out the semantic orientation of an adjective i.e. whether the adjective is positive, negative or neutral using NLTK in python.

For eg: If I have a sentence such as:

"This Camera is awesome."

So here 'awesome' is the adjective of 'camera' which has a positive orientation.

How do I go about doing this?

akhil
  • 71
  • 1
  • 4
  • Have you actually tried to have a go yourself? If so, what didn't work? – DavidG Apr 15 '16 at 08:36
  • Ya, I have tried doing it using wordnet in NLTK. But I was only able to figure out how to find synonyms, antonyms and similarity between words using Wordnet. Please tell a way to find orientations using this. – akhil Apr 15 '16 at 13:58

1 Answers1

0

WordNet focuses mostly on taxonomic relations rather than semantic ones. However, this is not necessarily a shortcoming of WordNet, since the qualia of "awesomeness" are very, very nuanced. Moreover, how do you know that This camera is awesome is in fact positive?-- what if the whole statement is:

This camera is awesome: First, the flash died on my wedding night --- the day after I bought it --- and then the battery caught fire and burnt my wedding dress, which was hand-sewn by my great-great grandmother.

However, one usage of WordNet for inferring semantic orientation was successful (at least in certain tasks): The work done by Kamps et al. in 2004. They based their orientation metric on the geodesic distance of a given word (e.g. awesome) from both good and bad:

EVA(w) = (d(w,bad)−d(w,good))/d(good,bad)

They claim that in certain tasks with certain data, this can work quite well, but its success is nevertheless very dependent on the task at hand; Looking into their work and work related to it may be a good place to start looking for answers.

errantlinguist
  • 3,658
  • 4
  • 18
  • 41
  • @akhil: If this answer helped you out, I'd be very grateful if you could [accept](https://stackoverflow.com/help/accepted-answer) it as I'm currently trawling for a few extra reputation points; Thanks a lot for the help. – errantlinguist May 11 '16 at 18:55