sentiment analysis - wordNet , sentiWordNet lexicon

Question

I need a list of positive and negative words with the weights assigned to words according to how strong and week they are. I have got :

1.) WordNet - It gives a + or - score for every word.

2.) SentiWordNet - Giving positive and negative values in the range [0,1].

I checked these on few words,

love - wordNet is giving 0.0 for both noun and verb, I dont know why i think it should be positive by at least some factor.

repress - wordNet gives -9.93
- SentiWordNet gives - 0.0 for both pos and neg. (should be negative)

repose - wordNet - 2.488
- SentiWordNet - { pos - 0.125, neg - 0.5 } (should be positive)

I need some help to decide which one to use.

Thanks.

Answer 1

Quite often the degree and/or polarity may depend on the domain and/or the context, so the word alone isn't really enough to make a decision.

If you have some annotated data, I suggest training a classifier on that using the scores provided by the two resources as features. If you don't, one option is to use one of the available sentiment-annotated corpora that matches the domain in question. Without any data at all the whole task becomes somewhat tricky, although there is a substantial body of work on unsupervised approaches to sentiment classification, I believe, see, e.g. Unsupervised Sentiment Analysis

Answer 2

There is an interface to give different opinions for SentiWordNet, if you think they are wrong: http://sentiwordnet.isti.cnr.it/search.php?q=repose

I downloaded latest Wordnet 3.1, and checked the file format documentation, and don't see any mention of the sentiment numbers you mention. It is also not shown in the online search.

So, for both those reasons I'd suggest going with SentiWordNet!

(I see your question is a year old, so perhaps you can tell us what you did go with, and why?)

Answer 3

The degree of the polarity depends not only on the words alone but also on the context of the sentece or the phrase. SO if there are different results regarding the same word then it is because of the difference in the context.