African Language Sentiments

library(tidyverse)
library(tidytext)
library(praise)

The Data

The data this week comes from AfriSenti: Sentiment Analysis dataset for 14 African languages via @shmuhammad2004 (the corresponding author on the associated paper, and an active member of the R4DS Online Learning Community Slack).

afrisenti <- read_csv("afrisenti.csv")

ibo <- afrisenti %>%
  filter(language_iso_code == "ibo")

Unnesting words and whatnot

ibo_top <- ibo %>%
  unnest_tokens(word, tweet, token = "ngrams", n = 2) %>%
  group_by(word) %>%
  filter(!(word %in% c("https t.co"))) %>%
  filter(!str_detect(word, "user")) %>%
  summarize(freq = n(), pos = mean(label == "positive"), 
            neg = mean(label == "negative"),
            neut = mean(label == "neutral")) %>%
  arrange(desc(freq)) %>%
  head(10) %>%
  pivot_longer(pos:neut, names_to = "sentiment", values_to = "proportion") %>%
  mutate(english = case_when(
    word == "nke a" ~ "this one",
    word == "ndi igbo" ~ "Igbo people",
    word == "na eme" ~ "and do",
    word == "ka m" ~ "like me",
    word == "ike gwuru" ~ "tired",
    word == "happy birthday" ~ "happy birthday",
    word == "gozie gi" ~ "bless you",
    word == "gi na" ~ "you and",
    word == "chukwu gozie" ~ "God bless",
    word == "a na" ~ "a"
  ))

ibo_top %>%
  ggplot(aes(x  = reorder(word, desc(freq)), y = proportion, width = freq, fill = sentiment)) + 
  geom_bar(stat = "identity", position = "fill", color = "grey") + 
  geom_text(y = 0.12, aes(x = reorder(word, desc(freq)), label = english)) + #,
  #           hjust = 0, vjust = .025, angle = 90, stat = "unique") + 
  coord_flip() + 
  facet_grid(reorder(word, desc(freq)) ~ 1, scales = "free_x", space = "free_x") + 
  theme_void() + 
  xlab("")  +
  scale_fill_brewer(palette = 4) + 
  labs(title = "Most common 2-grams in Ibo tweets",
       subtitle = "width proportional to # of tweets\ncolored by proportion of tweets with each sentiment",
       caption = "TidyTuesday AfriSenti\n GitHub: @hardin47\nData from @shmuhammad2004")

Breakdown of Ibo tweets in terms of the sentiment of the most common 2-grams.

praise()

[1] "You are world-class!"