I have run text through a sentiment analysis model, and it gives me 3 probabilities for if the text is Negative, Neutral or Positive e.g.
[Negative = 0.38]
[Neutral = 0.42]
[Positive = 0.20]
I am wanting to convert this to a score so that each text string has a score with higher scores meaning the text is more positive. I was thinking of just doing [Positive - Negative] and ignoring the neutral case to generate a score, but wondered if there were any better ideas as this seems overly simple. I want to generate a score so I can run regressions on these scores for predictive purposes.
Doing that sum on my dataset gets me a distribution that looks like this: https://i.imgur.com/DewzBRM.jpeg (Sorry for not embedding, I don't have the 'reputation' required to do so).
This seems pretty decent, but I was curious if anyone had other alternative ideas.
Thanks in advance