Denominator for "proportionalPol" sentiment computation

[odelmarcelle]

It always tickles me that compute_sentiment can yield values outside the [-1;1] range when using the "proportionalPol" method.

library(sentometrics)
sample_text <- setNames(nm = c("C'est un abandon", "C'est un vaste abandon"))
compute_sentiment(
  sample_text,
  lexicons = sentometrics::sento_lexicons(list(LM_sample = head(sentometrics::list_lexicons$LM_fr_tr, 5)),
                                          list_valence_shifters$fr),
  how = "proportionalPol"
)
#>                        id word_count LM_sample
#> 1:       C'est un abandon          3      -1.0
#> 2: C'est un vaste abandon          4      -1.8

^{Created on 2021-12-03 by the reprex package (v2.0.0)}

With this change, I adjusted the denominator so that it takes into account sentiment words that have been amplified-deamplified. Thus, the sentiment will always lie down within the [-1;1] interval. The same example after the change:

library(sentometrics)
sample_text <- setNames(nm = c("C'est un abandon", "C'est un vaste abandon"))
compute_sentiment(
  sample_text,
  lexicons = sentometrics::sento_lexicons(list(LM_sample = head(sentometrics::list_lexicons$LM_fr_tr, 5)),
                                          list_valence_shifters$fr),
  how = "proportionalPol"
)
#>                        id word_count LM_sample
#> 1:       C'est un abandon          3        -1
#> 2: C'est un vaste abandon          4        -1

^{Created on 2021-12-03 by the reprex package (v2.0.0)}

[sborms]

I think it's better to add a new weighting option than to change one which has been around for many versions and is implemented correctly. In the end it's just a choice and a matter of preference.

I would thus add a new weighting scheme, for instance called proportionalPolNorm, and clearly document it. It should also be added to the get_hows() function.