Consider using named fields for multi-key valence dictionaries

[kbenoit]

ANEW for instance has three categories, but if the dictionary is not sliced for a single key, then all three are averaged, which is probably not what a user wants.

library("quanteda")
#> Package version: 3.2.5
#> Unicode version: 14.0
#> ICU version: 70.1
#> Parallel computing: 10 of 10 threads used.
#> See https://quanteda.io for tutorials and examples.
library("quanteda.sentiment")
#> 
#> Attaching package: 'quanteda.sentiment'
#> The following object is masked from 'package:quanteda':
#> 
#>     data_dictionary_LSD2015

toks <- tokens(tail(data_corpus_inaugural))

tstat1 <- textstat_valence(toks, dictionary = data_dictionary_ANEW)
names(tstat1)[2] <- "sentiment_entire"

# (virtually) same as:
data_dictionary_ANEW["all"] <- data_dictionary_ANEW["arousal"]
valence(data_dictionary_ANEW)["all"] <- 
  list(all = apply(as.data.frame(valence(data_dictionary_ANEW)), 1, mean))
tstat2 <- textstat_valence(toks, dictionary = data_dictionary_ANEW["all"])

df <- cbind(tstat1, data.frame(sentiment_averaged = tstat2$sentiment))
df
#>       doc_id sentiment_entire sentiment_averaged
#> 1  2001-Bush         5.614141           5.616636
#> 2  2005-Bush         5.757551           5.758452
#> 3 2009-Obama         5.460327           5.459256
#> 4 2013-Obama         5.606507           5.606507
#> 5 2017-Trump         5.701053           5.701393
#> 6 2021-Biden         5.629749           5.630278
cor.test(df$sentiment_entire, df$sentiment_averaged)
#> 
#>  Pearson's product-moment correlation
#> 
#> data:  df$sentiment_entire and df$sentiment_averaged
#> t = 197.75, df = 4, p-value = 3.923e-09
#> alternative hypothesis: true correlation is not equal to 0
#> 95 percent confidence interval:
#>  0.9995085 0.9999947
#> sample estimates:
#>       cor 
#> 0.9999489

# different from (e.g.):
textstat_valence(toks, dictionary = data_dictionary_ANEW["pleasure"])
#>       doc_id sentiment
#> 1  2001-Bush  6.091330
#> 2  2005-Bush  6.308839
#> 3 2009-Obama  5.841437
#> 4 2013-Obama  6.045129
#> 5 2017-Trump  6.223944
#> 6 2021-Biden  6.018528

# what we probably want
rm(data_dictionary_ANEW)
structure(data.frame(
  textstat_valence(toks, dictionary = data_dictionary_ANEW["pleasure"]),
  textstat_valence(toks, dictionary = data_dictionary_ANEW["arousal"])["sentiment"],
  textstat_valence(toks, dictionary = data_dictionary_ANEW["dominance"])["sentiment"]),
  names = c("doc_id", paste("sentiment", names(data_dictionary_ANEW), sep = "_"))
)
#>       doc_id sentiment_pleasure sentiment_arousal sentiment_dominance
#> 1  2001-Bush           6.091330          5.252523            5.506055
#> 2  2005-Bush           6.308839          5.315839            5.650677
#> 3 2009-Obama           5.841437          5.207523            5.328807
#> 4 2013-Obama           6.045129          5.342878            5.431513
#> 5 2017-Trump           6.223944          5.334413            5.545822
#> 6 2021-Biden           6.018528          5.327389            5.544917

^{Created on 2023-02-13 with reprex v2.0.2}