Using Genius to get lyrics

We can use the genius package to query music lyrics in R. We use the command genius_lyrics() and to anlayze the lyrics from a song and artist. For this example we can get the lyrics for “Mr. Brightside” by The Killers. Next we can use Julia Silge and David Robinson’s tidytext package to unnest tokens and obtain 2-word “ngrams”. We will use this later on to make our Markov chain generator.

genius_lyrics("the killers", "mr. brightside") %>% 
    unnest_tokens(bigram, lyric, token = "ngrams", n = 2) %>%
    count(bigram) %>% 
    arrange(desc(n)) %>% 
    head(10) %>% 
    kable() 

Analyzing sentiment

Next we can use the sentiment lexicon from Bing Liu et al which is included in the tidytext package. We will use this to ‘inner_join’ with our lyric collection.

bing <- get_sentiments("bing")

bing %>% 
  head(10) %>% 
  kable()

Getting Cuz I Love You lyrics

Now that we learned how more about the genius package we can use the ’genius_album()` command to get the lyrics from Cuz I Love You.

Lizzo <- genius_album(artist = "Lizzo", album = "Cuz I Love You")

Lizzo %>% 
  head(10) %>% 
  kable()

Now we can get the 2-word “ngrams”.

Lizzo %>% 
    unnest_tokens(bigram, lyric, token = "ngrams", n = 2) %>%
    count(bigram) %>% 
    arrange(desc(n)) %>% 
   head(10) %>% 
  kable()

Lizzo Cuz I Love You sentiment analysis

Next we can take the Lizzo lyrics and get rid of stop words by using an anti_join() and inner_join() with the bing sentiment dataframe and add up the top 10 ngrams and visualize it.

Lizzo %>% 
  unnest_tokens(word, lyric) %>%
  anti_join(stop_words) %>%
  inner_join(bing) %>%
  count(word, sentiment, sort = TRUE) %>%
  group_by(sentiment) %>%
  top_n(10) %>%
  ungroup() %>%
  mutate(word = str_replace_all(word, "fuck", "f*ck"), 
         word = str_replace_all(word, "bitch", "b*tch"),
         word = reorder(word, n),
         word = fct_lump(word, n = 10)) %>%
  ggplot(aes(word, n, fill = sentiment)) +
  geom_col(show.legend = FALSE) +
  facet_wrap(~sentiment, scales = "free_y") +
  labs(y = "Cuz I Love You: Words that contribute the most to positive and negative sentiment",
       x = NULL) +
  coord_flip() +
  theme_minimal()

It seems like there’s plenty of “love” and positive word sentiment in this album!

Markovify

We can use the python markovify package and use the reticulate to use python in our R session.

# library(reticulate)
# use_python("/usr/local/bin/python")
# py_install("markovify")

Alex Bresler has ported markovify to an R packaged named Markovifyr.

# devtools::install_github("abresler/markovifyR")
library(markovifyR)

First get the a corpus of lyrics only.

Lizzo_corpus <- Lizzo %>% 
  pull(lyric)

Lizzo_corpus %>% 
  head(10) 

##  [1] "I'm cryin' 'cause I love you, oh"       
##  [2] "(Ya ya ya, ya ya, ya ya)"               
##  [3] "Never been in love before"              
##  [4] "What the fuck are fucking feelings, yo?"
##  [5] "Once upon a time, I was a ho"           
##  [6] "I don't even wanna ho no mo'"           
##  [7] "Got you something from the liquor store"
##  [8] "Little bit of Lizzo and some Mo"        
##  [9] "Tryna open up a little mo'"             
## [10] "Sorry if my heart a little slow"

Now we can create a Markov model with our corpus and specify the overlap.

Lizzo_markov_model <-
  generate_markovify_model(
    input_text = Lizzo_corpus,
    markov_state_size = 2L,
    max_overlap_total = 25,
    max_overlap_ratio = .85
  )

markovify_text(
  markov_model = Lizzo_markov_model,
  maximum_sentence_length = NULL,
  output_column_name = 'textLizzoBot',
  count = 25,
  tries = 100,
  only_distinct = TRUE,
  return_message = TRUE) %>% 
  kable()

## textLizzoBot: Mirror, mirror on the Goose
## textLizzoBot: Fuck it up to the dance floor
## textLizzoBot: Will you be my lover
## textLizzoBot: Oh, if I try to tell you, then you real like a girl, then you real like a girl
## textLizzoBot: Oh, I get an amen?
## textLizzoBot: And I don't need to cry
## textLizzoBot: Trust, baby, you can't make me crescendo, I'm going up
## textLizzoBot: I'm cryin', I'm cryin', hey
## textLizzoBot: I don't need to cry
## textLizzoBot: If I try to hurt me
## textLizzoBot: Fly you out of my life
## textLizzoBot: I'm about to add a little lost sometimes
## textLizzoBot: Lit up like a girl, then you real like a girl, cry like a girl, cry like a girl
## textLizzoBot: Wanna put you on a mink fur?
## textLizzoBot: You and you and you and you and you and you and you and
## textLizzoBot: The fact is I'm leaving, so just let me feel you close
## textLizzoBot: Smile if I try to change, hey
## textLizzoBot: Louis down to my juicy parts
## textLizzoBot: Shit, fuck, I didn't care
## textLizzoBot: So if you fight like a crystal ball
## textLizzoBot: And she never tell me to say?
## textLizzoBot: That's exactly how I roll
## textLizzoBot: Oh no, don't you get a little slow

library(magick)
Lizzo_meme <- image_read('https://upload.wikimedia.org/wikipedia/commons/0/0f/Lizzo_-_Palace_Theatre_-_St._Paul_%2827266085317%29.jpg') %>% 
  image_scale("800") %>% 
  image_annotate("FLY YOU OUT OF MY LIFE", font = "Helvetica",
                 size = 65, location = "+5+350", color = "white" )
print(Lizzo_meme)