5  Lyric Sentiments

require(sentimentr)
require(taylor)
require(ggrepel)

## taylor_album_songs from taylor package
lyrics_df <- taylor_album_songs %>%
  unnest(lyrics) %>%
  select(album_name, track_name, track_number, line, lyric, element)

lyrics_df$row_id <- 1:nrow(lyrics_df)

sentences_with_id <- get_sentences(lyrics_df$lyric, lyrics_df$row_id) # this processes
## each line as its own sentence as it's using the row_id. 
sentiment_with_id <- sentiment(sentences_with_id)
sentiment_summary <- sentiment_with_id %>%
  group_by(element_id) %>%
  summarise(
    avg_sentiment = mean(sentiment, na.rm = TRUE),
    word_count = sum(word_count)
  )

## Join

lyrics_sentiment <- lyrics_df %>%
  left_join(sentiment_summary, by = c("row_id" = "element_id"))
## Aggregate by song to get net sentiment score
song_sentiment_scores_sentimentr <- lyrics_sentiment %>%
  group_by(track_number, track_name, album_name) %>%
  summarize(
    sum_sentiment = sum(avg_sentiment),
    total_sentiment_words = n(),
    avg_sentiment = sum(avg_sentiment) / n(), 
    .groups = "drop"
  )

albumOrder <- c("Taylor Swift", "Fearless (Taylor's Version)", 
                "Speak Now (Taylor's Version)", "Red (Taylor's Version)",
                "1989 (Taylor's Version)", "Reputation", "Lover",
                "folklore", "evermore", "Midnights",
                "THE TORTURED POETS DEPARTMENT")

lyrics_sentiment <- lyrics_sentiment %>%
  mutate(album_name = factor(album_name, levels = albumOrder))
lyrics_sentiment <- lyrics_sentiment %>%
  filter(!is.na(album_name))

## Plot the net sentiment scores
ggplot(lyrics_sentiment, aes(x = track_number, y = avg_sentiment, fill = album_name)) +
  geom_col(show.legend = FALSE) +
  facet_wrap(~album_name, ncol = 3, scales = "free") +
  scale_fill_manual(values = colorPaletteAlbums) +
  theme_minimal() +
  labs(
    title = "The emotional spectrum across Taylor Swift's discography",
    x = "Track Number",
    y = "Net Sentiment (Positive - Negative Words)"
  ) +
  theme(plot.title = element_text(hjust=0.5))

song_sentiment_scores_sentimentr <- lyrics_sentiment %>%
  group_by(track_number, track_name, album_name) %>%
  summarize(
    sum_sentiment = sum(avg_sentiment),
    total_sentiment_words = n(),
    avg_sentiment = sum(avg_sentiment) / n(), 
    .groups = "drop"
  )

surpriseSongsDressColours <- surpriseSongsDressColours %>%
  mutate(DressColourGroup = case_when(
    DressName %in% c("Pink", "Flamingo pink") ~ "Reds",
    DressName %in% c("Blue", "Ocean blue") ~ "Blues",
    DressName %in% c("Yellow", "Sunset orange") ~ "Yellows",
    DressName %in% c("Cotton candy", "Grapefruit", "Popsicle") ~ "Colourful",
    DressName == "Blurple" ~ "Purples",
    DressName == "Green" ~ "Greens"
  ))

dress_song_sentiment_scores_sentimentr <- surpriseSongsDressColours %>%
  left_join(song_sentiment_scores_sentimentr, by = c("Song title" = "track_name")) %>%
  filter(!is.na(avg_sentiment))  

dress_group_sentiments <- dress_song_sentiment_scores_sentimentr %>%
  group_by(DressColourGroup) %>%
  summarise(
    performance_weighted_sentiment = mean(avg_sentiment, na.rm = TRUE),
    n_performances = n(),
    .groups = 'drop'
  )

## recall the CA from before
colorSentimentScores$colourGroup <- colorGroups[colorSentimentScores$colour]
cols.df <- as.data.frame.matrix(table(colorSentimentScores$colourGroup, colorSentimentScores$meaning))
coa <- FactoMineR::CA(cols.df)

## here we flip so that dim 1 is "-ve --> +ve"
colour_coa_table <- coa$row$coord %*% diag(c(-1,1)) |>
  as.data.frame() |>
  rownames_to_column("colour_category") |>
  select(colour_category, colour_sentiment = V1) |>
  as_tibble()


colour_sentiments <- colour_coa_table %>%
  mutate(
    DressColourGroup = case_when(
      colour_category == "blues" ~ "Blues",
      colour_category == "colorful" ~ "Colourful", 
      colour_category == "greens" ~ "Greens",
      colour_category == "purples" ~ "Purples",
      colour_category == "reds" ~ "Reds",
      colour_category == "yellows" ~ "Yellows",
      TRUE ~ NA_character_
    )
  ) %>%
  filter(!is.na(DressColourGroup))

dress_group_sentiments <- dress_song_sentiment_scores_sentimentr %>%
  group_by(DressColourGroup) %>%
  summarise(
    performance_weighted_sentiment = mean(avg_sentiment, na.rm = TRUE),
    n_performances = n(),
    .groups = 'drop'
  )

combined_analysis <- dress_group_sentiments %>%
  left_join(colour_sentiments, by = "DressColourGroup") %>%
  select(DressColourGroup, performance_weighted_sentiment, colour_sentiment, n_performances)

Let’s plot.

song_performance_counts <- dress_song_sentiment_scores_sentimentr %>%
  group_by(`Song title`, DressColourGroup) %>%
  summarise(
    avg_sentiment = first(avg_sentiment),  
    performance_count = n(),              
    .groups = 'drop'
  )

plot_data <- song_performance_counts %>%
  left_join(colour_sentiments, by = "DressColourGroup") %>%
  filter(!is.na(colour_sentiment))  


## Top songs (highest sentiment) for each color group
top_songs <- plot_data %>%
  group_by(DressColourGroup) %>%
  slice_max(avg_sentiment, n = 1, with_ties = FALSE) %>%
  ungroup()

## Bottom songs (lowest sentiment) for each color group  
bottom_songs <- plot_data %>%
  group_by(DressColourGroup) %>%
  slice_min(avg_sentiment, n = 1, with_ties = FALSE) %>%
  ungroup()

## Medium songs (closest to median for each color group)
medium_songs <- plot_data %>%
  group_by(DressColourGroup) %>%
  mutate(
    median_sentiment = median(avg_sentiment, na.rm = TRUE),
    distance_from_median = abs(avg_sentiment - median_sentiment)
  ) %>%
  slice_min(distance_from_median, n = 1, with_ties = FALSE) %>%  # Closest to median per color (medium)
  ungroup()

highlight_songs <- bind_rows(
  top_songs,
  bottom_songs,
  medium_songs) %>%
  distinct(`Song title`, DressColourGroup, .keep_all = TRUE)

colourGroupPalette <- c(
  "blues" = "#5A8BC8",      
  "reds" = "#D17A8A",       
  "purples" = "#A675C8",  
  "greens" = "#7AB87A",     
  "yellows" = "#E6C470",    
  "colourful" = "#D19A70"
)

ggplot(plot_data, aes(x = colour_sentiment, y = avg_sentiment)) +
   geom_point(aes(size = performance_count, colour = tolower(DressColourGroup)), 
             alpha = 0.4, stroke = 0.8) +
    geom_point(data = highlight_songs,
             aes(size = performance_count, colour = tolower(DressColourGroup)), 
             alpha = 0.9, stroke = 1.5) +
    geom_text_repel(data = highlight_songs,
                  aes(label = `Song title`, colour = tolower(DressColourGroup)), 
                  size = 6, 
                  max.overlaps = Inf,
                  box.padding = 0.5,
                  point.padding = 0.3,
                  min.segment.length = 0.1,
                  show.legend = FALSE) +
  scale_size_continuous(name = "Performance\nCount", 
                        range = c(2, 12),
                        breaks = c(1, 2, 3, 4, 5),
                        guide = guide_legend(override.aes = list(alpha = 1))) +
  scale_colour_manual(values = colourGroupPalette,
                      name = "Dress colour\nGroup") +
    labs(title = "",
       subtitle = "",
       x = "Dress Colour Sentiment",
       y = "Lyric Sentiment",
       caption = "") +
  theme_minimal() +
  theme(
    plot.title = element_blank(),
    axis.title.x = element_text(size = 16, 
                                margin = margin(t = 10)),
    axis.title.y = element_text(size = 16),
    axis.text = element_text(size = 14),
    legend.title = element_text(size = 12),
    legend.text = element_text(size = 11),
    legend.position = "bottom",
    legend.box = "horizontal"
  )

What about group-wise correlation?

## We only have 6 observations, not great, but let's go ahead anyway!
## Pearson's correlation
cor.test(combined_analysis$performance_weighted_sentiment, 
                             combined_analysis$colour_sentiment)

    Pearson's product-moment correlation

data:  combined_analysis$performance_weighted_sentiment and combined_analysis$colour_sentiment
t = -5.4996, df = 4, p-value = 0.005329
alternative hypothesis: true correlation is not equal to 0
95 percent confidence interval:
 -0.9935629 -0.5403340
sample estimates:
       cor 
-0.9397859 

## Spearman's correlation
cor.test(combined_analysis$performance_weighted_sentiment, 
                             combined_analysis$colour_sentiment,
                             method = "spearman")

    Spearman's rank correlation rho

data:  combined_analysis$performance_weighted_sentiment and combined_analysis$colour_sentiment
S = 54, p-value = 0.2972
alternative hypothesis: true rho is not equal to 0
sample estimates:
       rho 
-0.5428571 

## Spearman's correlation with 95% CIs
correlation::cor_test(data = combined_analysis,
                      x = "performance_weighted_sentiment", 
                      y = "colour_sentiment",
                      method = "spearman")

Parameter1                     |       Parameter2 |   rho |        95% CI
-------------------------------------------------------------------------
performance_weighted_sentiment | colour_sentiment | -0.54 | [-0.94, 0.51]

Parameter1                     |     S |     p
----------------------------------------------
performance_weighted_sentiment | 54.00 | 0.266

Observations: 6

“Billboard Hot 100™.” n.d. https://www.billboard.com/charts/hot-100/.

Bortoli, Mario De, and Jesús Maroto. 2001. “Colours Across Cultures: Translating Colours in Interactive Marketing Communications.” In European Languages and the Implementation of Communication and Information Technologies (Elicit) Conference.

Kaufman, Gil. 2025. “Orange You Glad Taylor Swift Is Back?” Billboard, August. https://www.billboard.com/music/pop/taylor-orange-life-showgirl-album-memes-elmo-target-brands-1236042681/.

Olson, Samantha. 2022. “All the Easter Eggs in Taylor Swift’s Bejeweled Music Video.” Seventeen, October. https://www.seventeen.com/celebrity/music/a41766334/taylor-swift-bejeweled-music-video-easter-eggs/.

Rinker, Tyler W. 2021. “Sentimentr: Calculate Text Polarity.”

Rowley, Glenn. 2023. “Taylor Swift’s the Eras Tour Easter Eggs: Here’s What Fans Have Spotted.” Billboard, May. https://www.billboard.com/lists/taylor-swift-eras-tour-easter-eggs/the-red-taylors-version-t-shirt/.

Savage, Mark. 2024. “Taylor Swift: As the Eras Tour Bows Out, What Will She Do Next?” BBC, December. https://www.bbc.com/news/articles/ckg1r0v838vo.

Thompson, W Jake. 2025. “Taylor: Lyrics and Song Data for Taylor Swift’s Discography.”