SPECIAL_CASES do not work

[mchlandel]

For example, the phrase "kiss of death" from the SPECIAL_CASES dictionary has assigned the value -1.5. Therefore, I would suppose that whole this phrase should have got negative sentiment, unlike the phrases "kiss" and "death" separately, which are positive and negative, respectively. But the code behaves strangely:

• for a trivial one-word sentence "kiss", I get 'compound': -0.4215, which is correct since "kiss" has positive sentiment,

• for "death", I get 'compound': -0.5994, which is correct since "death" has very negative sentiment,

• for a longer sentence "kiss and death", I get 'compound': -0.2732, which is correct too, since the very negative "death" is mitigated by slightly positive "kiss".

• But for the sentence "kiss of death", I get 'compound': -0.2732 too, which seems to me incorrect since, according to the SPECIAL_CASES dictionary, the phrase "kiss of death" per se should have got entirely negative sentiment.

But what is even more strange is that,

• for the more longer phrase "it was his kiss of death" I get 'compound': -0.6124 which seems correct so long as
• I got the 'compound': -0.6124 for the shorter phrase "it was kiss of death"

Isn't there a bug in the tri-gram heuristic?

[ViennaMike]

If you look at the source code, while there are lists for SPECIAL_CASES and for SENTIMENT_LADEN_IDIOMS, there's no code actually using SPECIAL_CASES, and the code for SENTIMENT_LADEN_IDIOMS is only a placeholder. So no, they don't work at present.

[ViennaMike]

For the maintainer: I should have some time to work on adding the missing code for SPECIAL_CASES and SENTIMENT_LADEN_IDIOMS. If I put the time in and get it working, is this main branch still being maintained and are you interested in such a fix?

[Siddharth-Latthe-07]

The vader module works on the basis of finding the lexical meaning of the phrases and then providing the scores between -1 and +1, so I think that it might be processing the phrase Kiss of Death, as a whole and the word death having a string negative sentiment as compared to kiss, it is showing overall negative emotion.
I don't think there is a bug, in the fiunctionality, but seems work needs to be done in handling such edge case scenarios