-1

I want to find the sentiment for

@SouthwestAir Fastest response all day. Hour on the phone: never got off hold. Hour in line: never got to the Flight Booking Problems desk.

When I use NLP sentiment class, I get

Neutral    @SouthwestAir Fastest response all day.
Negative    Hour on the phone: never got off hold.
Negative    Hour in line: never got to the Flight Booking Problems desk.

But when I use the Monkey Learn API for same text, I get a single sentiment 

label : "negative"

How can I determine that the tweet text is negative using NLP sentiment class?

Cœur
  • 37,241
  • 25
  • 195
  • 267
bunny sunny
  • 301
  • 6
  • 15
  • To clarify, is the concern here that CoreNLP splits the tweet into sentences and assigns sentiment on a sentence-by-sentence basis? – Gabor Angeli Jun 16 '16 at 16:16

1 Answers1

0

CoreNLP scores each sentence. If you replace each internal period (all but the last) with a semicolon or dash, I believe you'll get the effect you want.

@SouthwestAir Fastest response all day -- 
Hour on the phone: never got off hold -- 
Hour in line: never got to the Flight Booking Problems desk.
Prune
  • 76,765
  • 14
  • 60
  • 81
  • 1
    Disclaimers, however: this will (1) be rather slow (parsing is O(n^3)), and (2) will likely confuse the parser, and therefore the RNN sentiment analyzer. Hopefully not by much, but that's an empirical question. – Gabor Angeli Jun 17 '16 at 01:23