Classification algorithm for estimating the score

1
votes

I want to detect some sort of sentiment orientation for text article. This problem seems related to classification problem, but instead of detecting probabilities of each class (negative, positive, neutral), I want to know some overall score rating, like 0.76 and then classify my article into category that covers the predefined ranges. (e.g. [0.75...1) is positive).

What ml algorithms are suitable for such problem?

1
algorithmmachine-learningclassification
So, you are basically looking for binary classifier that also gives the probability of the class being correct? Naive Bayes, and SVM supports that - and are pretty good for text classification (from my personal experiance) – amit
@amit AFAIK Naive Bayes output is list of probabilities for each class, instead of this I want just one number (e.g. [0..1]) which can represent the article rating. Smth like automatic classifier for movie review rating. – mishadoff
And what is your training set? You could try linear regression if your training set is numbers as well. – amit
did u tried Weka framework? – amrfaissal
@FGraviton Weka consist of a lot of various algorithms – mishadoff

1 Answers

1
votes

As far as I see it, you can do it with one of these two approaches:

  1. Use a classifcation algorithm, for binary classifier it gives you (p,1-p) - where p is the "chance" the binary classifier is giving it to be "true".
  2. Use linear regression (or other numerical ML algorithm), and give the score it returns you. You will label "pos" as 1 and "neg" as 0 when training your algorithm.

Personally, I'd go for the first approach with SVM, since I know it handles large feature space well - and it is likely to be the case in text problems.