Sentiments with Peach Pit

A class project trying sentiment dictionaries on lyrics of the album “Magpie” by Peach Pit.

Author

Jeneta Nwosu

Published

May 16, 2025

Modified

May 17, 2025

Question

“Magpie” by Peach Pit was my favorite album of the year in 2024. Peach Pit is known for having creative (maybe even zany?) lyricism, and I’m curious if the meaning behind what they write is parsable by a machine. Below, I predict what sentiments will be identified and then see if the results line up with my expectations.

As an example of the problem, the first track “Every Little Thing” sounds upbeat and tender. But when I looked up what it meant, I learned it’s written from the Devil’s perspective. Spooky! On a second read, the malevolence and sinister themes of temptation become clear.

Data

The lyrics were scraped from Genius using their API and the lyricsgenius python package. I replaced the “right single quote character” with a normal apostrophe before importing because it was breaking the cleaning process.

Prediction

# What sentiments are possible?
nrc_sentiments <- tidytext::get_sentiments("nrc")
nrc_sentiments %>%
  select(sentiment) %>%
  unique()

# A tibble: 10 × 1
   sentiment   
   <chr>       
 1 trust       
 2 fear        
 3 negative    
 4 sadness     
 5 anger       
 6 surprise    
 7 positive    
 8 disgust     
 9 joy         
10 anticipation

library(SentimentAnalysis)


Attaching package: 'SentimentAnalysis'

The following object is masked from 'package:base':

    write

harvard_iv <- SentimentAnalysis::DictionaryGI
names(harvard_iv)

[1] "negative" "positive"

While the Harvard IV dictionary assigns its words to either negative or positive sentiment, the NRC dictionary contains 10 possible emotions. The Harvard IV options will help simplify the meaning of the document, and the NRC will help capture nuances. My predictions below are mostly about the literal meaning of the lyrics, not the figurative.

Document	NRC Sentiment	Harvard Sentiment
Every Little Thing	anticipation, trust	positive
Yasmina	positive, joy	positive
Am I Your Girl	negative	negative
Little Dive	joy, positive*	positive
Outta Here	fear, anger, negative	negative
Did You Love Somebody	negative, anger, fear	negative
St. Mark’s Funny Feeling	negative, sadness	negative
Magpie	negative	negative
Nowhere Next To Me	disgust, sadness, negative	negative
Wax & Wane	negative, disgust	negative
Your Long Black Hair	negative, anticipation	negative

*Song is about doing drugs, but the words are positive

Descriptive

# Read in with track numbers to preserve the order for later
magpie_album_order <- read_csv('magpie.csv', col_select = "number":"song.lyrics")

New names:
Rows: 11 Columns: 3
── Column specification
──────────────────────────────────────────────────────── Delimiter: "," chr
(2): song.full_title, song.lyrics dbl (1): number
ℹ Use `spec()` to retrieve the full column specification for this data. ℹ
Specify the column types or set `show_col_types = FALSE` to quiet this message.
• `` -> `...1`

magpie <- magpie_album_order[1:nrow(magpie_album_order),2:3]
# Getting rid of punctuation that isn't within the text for the word cloud. Ex. "me," becomes "me", but "I'll" doesn't become "ill"
lyrics_no_punct <- tm::removePunctuation(magpie$song.lyrics,preserve_intra_word_contractions = T, preserve_intra_word_dashes = T)
docs <- tm::Corpus(tm::VectorSource(lyrics_no_punct))

dtm <- tm::DocumentTermMatrix(docs,
                              control = list(stopwords = TRUE,
                                             tolower = TRUE))


freq <- colSums(as.matrix(dtm))

library(wordcloud)

Loading required package: RColorBrewer

set.seed(169)
wordcloud::wordcloud(names(freq), freq, max.words = 100)

NRC

# Turning words into tokens but keeping as many words as possible, as there's only 10 songs
magpie <- magpie %>%
  tidytext::unnest_tokens(word, song.lyrics) 


# Joining NRC dictionary with Magpie words

nrc_words <- magpie %>%
  dplyr::inner_join(nrc_sentiments)

Joining with `by = join_by(word)`

Warning in dplyr::inner_join(., nrc_sentiments): Detected an unexpected many-to-many relationship between `x` and `y`.
ℹ Row 10 of `x` matches multiple rows in `y`.
ℹ Row 1573 of `y` matches multiple rows in `x`.
ℹ If a many-to-many relationship is expected, set `relationship =
  "many-to-many"` to silence this warning.

# Summarizing sentiment
magpie_counts <- nrc_words %>%
  dplyr::group_by(song.full_title, sentiment) %>%
  dplyr::count()

mag_sent_nrc <- magpie_counts %>%
  ungroup() %>%
  arrange(song.full_title, -n) %>%
  mutate(sentiment_nrc = paste(sentiment,paste0("(", n , ")"))) %>%
  group_by(song.full_title) %>%
  summarize(sentiment_nrc = toString(sentiment_nrc)) %>%
  ungroup()

mag_sent_nrc

# A tibble: 11 × 2
   song.full_title          sentiment_nrc                                       
   <chr>                    <chr>                                               
 1 Am I Your Girl           negative (2), sadness (2), trust (2), disgust (1)   
 2 Did You Love Somebody    negative (9), anger (6), fear (6), joy (6), positiv…
 3 Every Little Thing       positive (8), trust (7), joy (6), negative (5), sad…
 4 Little Dive              negative (15), positive (14), anticipation (8), fea…
 5 Magpie                   negative (9), positive (7), joy (5), sadness (3), s…
 6 Nowhere Next To Me       negative (12), sadness (9), anger (8), disgust (8),…
 7 Outta Here               negative (14), anticipation (11), positive (10), di…
 8 St. Mark's Funny Feeling negative (14), anticipation (9), disgust (9), sadne…
 9 Wax & Wane               negative (3), positive (3), sadness (3), surprise (…
10 Yasmina                  negative (6), anger (4), joy (4), positive (4), tru…
11 Your Long Black Hair     anticipation (11), negative (9), sadness (8), fear …

Harvard IV

# Turning dictionary into dataframe
harvard_iv <- sapply(harvard_iv, '[', 1:2005)
harvard_iv <-data.frame(harvard_iv)
harvard_sentiments <- harvard_iv %>%
  pivot_longer(colnames(harvard_iv), values_to = 'word',names_to = 'sentiment',)

# Joining Harvard dictionary with Magpie words
harvard_words <- magpie %>%
  dplyr::inner_join(harvard_sentiments)

Joining with `by = join_by(word)`

Warning in dplyr::inner_join(., harvard_sentiments): Detected an unexpected many-to-many relationship between `x` and `y`.
ℹ Row 30 of `x` matches multiple rows in `y`.
ℹ Row 1806 of `y` matches multiple rows in `x`.
ℹ If a many-to-many relationship is expected, set `relationship =
  "many-to-many"` to silence this warning.

# Summarizing sentiment
magpie_counts_2 <- harvard_words %>%
  dplyr::group_by(song.full_title, sentiment) %>%
  dplyr::count()

mag_sent_harv <- magpie_counts_2 %>%
  ungroup() %>%
  mutate(sentiment_harv = sentiment) %>%
  select(-sentiment) %>%
  group_by(song.full_title) %>%
  filter(n == max(n)) %>%
  ungroup() %>%
  mutate(sentiment_harv = ifelse(song.full_title == "Magpie", "negative and positive", sentiment_harv)) %>%
  unique()

mag_sent_harv

# A tibble: 11 × 3
   song.full_title              n sentiment_harv       
   <chr>                    <int> <chr>                
 1 Am I Your Girl               3 negative             
 2 Did You Love Somebody       25 negative             
 3 Every Little Thing          13 positive             
 4 Little Dive                 20 positive             
 5 Magpie                      16 negative and positive
 6 Nowhere Next To Me          15 positive             
 7 Outta Here                  16 positive             
 8 St. Mark's Funny Feeling     9 positive             
 9 Wax & Wane                  12 positive             
10 Yasmina                     18 negative             
11 Your Long Black Hair        15 positive

# ^^The song "Magpie" is equally positive and negative

Final Results

# Sentiments from both dictionaries in original order
mag_sent <- mag_sent_nrc %>%
  left_join(mag_sent_harv[1:nrow(magpie_counts_2),c(1,3)], by = 'song.full_title') %>%
  right_join(magpie_album_order[1:nrow(magpie_album_order), 1:2]) %>%
  arrange(number)

Joining with `by = join_by(song.full_title)`

mag_sent

# A tibble: 11 × 4
   song.full_title          sentiment_nrc                  sentiment_harv number
   <chr>                    <chr>                          <chr>           <dbl>
 1 Every Little Thing       positive (8), trust (7), joy … positive            1
 2 Yasmina                  negative (6), anger (4), joy … negative            2
 3 Am I Your Girl           negative (2), sadness (2), tr… negative            3
 4 Little Dive              negative (15), positive (14),… positive            4
 5 Outta Here               negative (14), anticipation (… positive            5
 6 Did You Love Somebody    negative (9), anger (6), fear… negative            6
 7 St. Mark's Funny Feeling negative (14), anticipation (… positive            7
 8 Magpie                   negative (9), positive (7), j… negative and …      8
 9 Nowhere Next To Me       negative (12), sadness (9), a… positive            9
10 Wax & Wane               negative (3), positive (3), s… positive           10
11 Your Long Black Hair     anticipation (11), negative (… positive           11

Discussion

NRC picked up on a broader range of sentiments per song than I expected. I only made a few predictions per song, and many ended up being correct, but not complete. I also failed to predict the number of songs that ended up with positive themes; the same was true for the Harvard IV results. I think on that front, the NRC assignments are better than the Harvard ones. “Wax and Wane” is in no way a positive song.

However, I think the results of both dictionaries benefit from lucky/surface level guesses. For example, in “Little Dive” the NRC dictionary identifies “dandy” with “disgust” and “negative” presumably referring to the pejorative term for a fashionable man, instead of the sense it is actually used in the song, which is a positive feeling. For some reason, “train” gets a positive identification by the Harvard IV dictionary in Your Long Black Hair when I would argue it’s by default neutral word (and in context, negative — “I been thinking up a one-night train from this world”). While there’s a valid argument for considering the song as positive, that’s not it.

I conclude that text analysis using dictionaries like these is a fun tool for a cursory investigation of themes in song lyrics, but especially for complex works like “Magpie” it’s probably best to use a more advanced tool like the human brain. And if you must use a dictionary, NRC performs better!

See the full final table below.

Document	NRC Sentiment	Harvard Sentiment
Every Little Thing	positive (8), trust (7), joy (6), negative (5), sadness (5), anticipation (4), fear (3), disgust (1)	positive
Yasmina	negative (6), anger (4), joy (4), positive (4), trust (4), anticipation (1), disgust (1), fear (1), sadness (1)	negative
Am I Your Girl	negative (2), sadness (2), trust (2), disgust (1)	negative
Little Dive	negative (15), positive (14), anticipation (8), fear (7), sadness (7), surprise (7), trust (7), disgust (5), anger (4)	positive
Outta Here	negative (14), anticipation (11), positive (10), disgust (9), anger (8), fear (6), sadness (6), surprise (5), joy (4), trust (3)	positive
Did You Love Somebody	negative (9), anger (6), fear (6), joy (6), positive (6), sadness (4), disgust (2), anticipation (1), trust (1)	negative
St. Mark’s Funny Feeling	negative (14), anticipation (9), disgust (9), sadness (9), fear (8), anger (7), positive (7), trust (7), joy (5), surprise (5)	positive
Magpie	negative (9), positive (7), joy (5), sadness (3), surprise (3), anger (2), anticipation (2), disgust (2), trust (2)	negative and positive
Nowhere Next To Me	negative (12), sadness (9), anger (8), disgust (8), fear (8), joy (7), positive (7), trust (4), anticipation (3), surprise (3)	positive
Wax & Wane	negative (3), positive (3), sadness (3), surprise (3), trust (3), anticipation (2), joy (2), fear (1)	positive
Your Long Black Hair	anticipation (11), negative (9), sadness (8), fear (5), disgust (4), anger (3), joy (2), positive (2), surprise (2), trust (2)	positive