3.1 Subjectivity Lexicons

A subjectivity lexicon is a predefined list of words associated with emotional context such as positive/negative. Subjectivity lexicons are typically short (a few thousand words), but work because of Zipf’s law which holds that the nth-ranked item in a frequency table has a frequency count equal to 1/n of the top-ranked item. So infrequently used words are used very infrequently.

There are three common sentiment lexicons. Bing is common for polarity scoring, AFINN for emotion classification. NRC is a less common option for emotion classification.

Bing classifies words as positive or negative.

bing <- tidytext::get_sentiments("bing") %>%
  # remove dups
  filter(!word %in% c("envious", "enviously", "enviousness"))

bing %>% 
  count(sentiment) %>% 
  adorn_totals() %>%
  flextable::flextable() %>% 
  flextable::autofit()

sentiment	n
negative	4,778
positive	2,002
Total	6,780

AFINN, by Finn Arup Nielsen, associates words with a manually rated valence integer between -5 (negative) and +5 (positive).

afinn <- tidytext::get_sentiments("afinn")

afinn %>%
  count(value) %>% 
  adorn_totals() %>%
  flextable::flextable() %>% 
  flextable::autofit()

value	n
-5	16
-4	43
-3	264
-2	966
-1	309
0	1
1	208
2	448
3	172
4	45
5	5
Total	2,477

NRC lexicon associates words with eight emotions corresponding to the second level of Plutchik’s Wheel of Emotions and two sentiments (negative and positive). NRC was created by manual annotation on a crowd sourcing platform (see this).

nrc <- tidytext::get_sentiments("nrc")

nrc %>%
  count(sentiment) %>% 
  adorn_totals() %>%
  flextable::flextable() %>% 
  flextable::autofit()

sentiment	n
anger	1,245
anticipation	837
disgust	1,056
fear	1,474
joy	687
negative	3,316
positive	2,308
sadness	1,187
surprise	532
trust	1,230
Total	13,872