But really, who is the G.O.A.T? (Part 1)

Fri, 02 Aug 2024 00:00:00 +0000

Look, I know some of you might not like hip hop and don’t get me wrong, I don’t think there’s anything inherently wrong with having different tastes and preferences in terms of music, but we are living in a time where we have multiple chart-topping hip hop artists of incredible talent who are redefining the genre while simultaneously bringing us back to the intricate flow and lyricism of old-school legends. The names of some of these artists should be familiar to you, even if you aren’t a hip hop head like me, but what makes their music good? Are there identifiable characteristics of their music that we can quantify to gain insights about what defines good hip hop artists?

What is it that defines good hip hop?

What better place to answer that question than with music from the big three: Drake, Kendrick Lamar, and J. Cole.

In Part 1 of this series, I’ll walk you through how to use the spotifyr package to extract audio features from Spotify for Developers. We will then create some tables and visualizations we can use to generate data-driven questions about music from the Big 3. To start, we will need to follow some simple steps outlined in this Spotify API Tutorial.

# lets load our packages
if (!require("pacman")) install.packages("pacman")
pacman::p_load(tidyverse,spotifyr,ggridges,ggimage,cowplot,
               gt,gtExtras,glue,ggfx,extrafont)

# specify client ID number from website
Sys.setenv(SPOTIFY_CLIENT_ID = "5de20cf621eb4e64b375af831a4f4ad4")

# specify client secret from website. I specified this above by writing this code and populating with my spotify client secret: spotify_client_secret = c(""xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"") 
Sys.setenv(SPOTIFY_CLIENT_SECRET = spotify_client_secret)

# get access token
access_token <- get_spotify_access_token()

Now that we have access to the Spotify API, let’s import audio features and metadata for each member of the Big 3 and then create a single object containing data from all three artists.

# let's get Kendrick (section.80 was fire) ...
kendrick_lamar <- get_artist_audio_features('Kendrick Lamar')

# ... then let's get the man who bodied Drake on 'First Person Shooter' ... 
j_cole <- get_artist_audio_features("J. Cole")

# ... and finally Drake (So Far Gone Mixtape was my anthem in 2009)....
drake <- get_artist_audio_features("Drake")

# ... and let's combine them ...
big_three <- bind_rows(kendrick_lamar,j_cole,drake)

Let’s inspect a small subset of the data, remove any unnecessary columns, and generate some datasets we can use to create tables and visualizations.

# randomly select three songs for each artist
big_three %>%
  group_by(artist_name) %>%
  slice_sample(n = 4)

## # A tibble: 12 × 39
## # Groups:   artist_name [3]
##    artist_name    artist_id  album_id album_type album_images album_release_date
##    <chr>          <chr>      <chr>    <chr>      <list>       <chr>             
##  1 Drake          3TVXtAsR1… 4czdORd… album      <df [3 × 3]> 2023-10-06        
##  2 Drake          3TVXtAsR1… 0ptlfJf… album      <df [3 × 3]> 2015-02-12        
##  3 Drake          3TVXtAsR1… 4Q7cRXi… album      <df [3 × 3]> 2023-11-17        
##  4 Drake          3TVXtAsR1… 7dqpveM… album      <df [3 × 3]> 2019-08-02        
##  5 J. Cole        6l3HvQ5sa… 2n3quCZ… album      <df [3 × 3]> 2019-07-05        
##  6 J. Cole        6l3HvQ5sa… 3tFjhFP… album      <df [3 × 3]> 2022-03-31        
##  7 J. Cole        6l3HvQ5sa… 3CCnGld… album      <df [3 × 3]> 2016-12-09        
##  8 J. Cole        6l3HvQ5sa… 4JAvwK4… album      <df [3 × 3]> 2021-05-14        
##  9 Kendrick Lamar 2YZyLoL8N… 4alcGHj… album      <df [3 × 3]> 2017-12-08        
## 10 Kendrick Lamar 2YZyLoL8N… 3DGQ1iZ… album      <df [3 × 3]> 2012              
## 11 Kendrick Lamar 2YZyLoL8N… 4alcGHj… album      <df [3 × 3]> 2017-12-08        
## 12 Kendrick Lamar 2YZyLoL8N… 1bkN9nI… album      <df [3 × 3]> 2011-07-02        
## # ℹ 33 more variables: album_release_year <dbl>,
## #   album_release_date_precision <chr>, danceability <dbl>, energy <dbl>,
## #   key <int>, loudness <dbl>, mode <int>, speechiness <dbl>,
## #   acousticness <dbl>, instrumentalness <dbl>, liveness <dbl>, valence <dbl>,
## #   tempo <dbl>, track_id <chr>, analysis_url <chr>, time_signature <int>,
## #   artists <list>, available_markets <list>, disc_number <int>,
## #   duration_ms <int>, explicit <lgl>, track_href <chr>, is_local <lgl>, …

# remove columns with titles that either 1) end with or 2) contain specific sequences
big_three_1 <- big_three %>%
  select(-ends_with("_id"),-contains("url"),-contains("href")) 

# let's also unnest the album_images column and filter out any remaining list columns
big_three_2 <- big_three_1 %>%
  unnest(cols = c("album_images")) %>%
  select(!where(is.list))
  
# let's take a look... 
head(big_three_2)

## # A tibble: 6 × 32
##   artist_name    album_type height url                  width album_release_date
##   <chr>          <chr>       <int> <chr>                <int> <chr>             
## 1 Kendrick Lamar album         640 https://i.scdn.co/i…   640 2022-05-13        
## 2 Kendrick Lamar album         300 https://i.scdn.co/i…   300 2022-05-13        
## 3 Kendrick Lamar album          64 https://i.scdn.co/i…    64 2022-05-13        
## 4 Kendrick Lamar album         640 https://i.scdn.co/i…   640 2022-05-13        
## 5 Kendrick Lamar album         300 https://i.scdn.co/i…   300 2022-05-13        
## 6 Kendrick Lamar album          64 https://i.scdn.co/i…    64 2022-05-13        
## # ℹ 26 more variables: album_release_year <dbl>,
## #   album_release_date_precision <chr>, danceability <dbl>, energy <dbl>,
## #   key <int>, loudness <dbl>, mode <int>, speechiness <dbl>,
## #   acousticness <dbl>, instrumentalness <dbl>, liveness <dbl>, valence <dbl>,
## #   tempo <dbl>, time_signature <int>, disc_number <int>, duration_ms <int>,
## #   explicit <lgl>, is_local <lgl>, track_name <chr>, track_number <int>,
## #   type <chr>, track_uri <chr>, album_name <chr>, key_name <chr>, …

# look's like we now have three rows per song: one row for each image, so let's create a new big three by removing the columns we just made and grabbing only distinct rows
big_three_3 <- big_three_2 %>%
  select(-height:-width) %>%
  distinct()

# .. let's also pull out the album data and images and store them in a separate data frame for later, but filter so we only keep the smallest images. 
albums <- big_three_2 %>%
  select(artist_name,album_release_year,album_name,height,width,url) %>%
  filter(height == 64) %>%
  distinct()

Now we have two datasets with different observational units: 1) an album-level dataset in which each row contains a single album with information about the album name, artist name, and release year and 2) a song-level dataset in which each row contains metadata for a single song. This raises an important question: what now? Let’s create a table with their discographies as we start thinking about next steps.

# let's make a table where each row is an album from a given artist and display the album cover
albums %>%
  gt(rowname_col = "artist_name") %>%
  cols_hide(c(height,width)) %>%
  tab_header(title = md("Artist Albums by Year"))%>%
  tab_stubhead(label = "Artist") %>%
  gt_img_rows(url,height = 50) %>%
  cols_label(
    artist_name = "Artist",
    album_name = "Album Name",
    album_release_year = "Release Year",
    url = "Album Cover") %>%
  cols_move_to_start(
    columns = c(url,album_name,album_release_year)) %>%
  cols_align(align = "center", columns = c("artist_name","url","album_release_year"))

Artist	Album Name	Release Year
Artist Albums by Year
Kendrick Lamar	Mr. Morale & The Big Steppers	2022
Kendrick Lamar	Black Panther The Album Music From And Inspired By	2018
Kendrick Lamar	DAMN. COLLECTORS EDITION.	2017
Kendrick Lamar	DAMN.	2017
Kendrick Lamar	untitled unmastered.	2016
Kendrick Lamar	To Pimp A Butterfly	2015
Kendrick Lamar	good kid, m.A.A.d city (Deluxe)	2012
Kendrick Lamar	good kid, m.A.A.d city	2012
Kendrick Lamar	Section.80	2011
Kendrick Lamar	Overly Dedicated	2010
J. Cole	Might Delete Later	2024
J. Cole	D-Day: A Gangsta Grillz Mixtape	2022
J. Cole	The Off-Season	2021
J. Cole	Revenge Of The Dreamers III: Director's Cut	2020
J. Cole	Revenge Of The Dreamers III	2019
J. Cole	KOD	2018
J. Cole	4 Your Eyez Only	2016
J. Cole	Forest Hills Drive: Live from Fayetteville, NC	2016
J. Cole	2014 Forest Hills Drive	2014
J. Cole	Born Sinner (Deluxe Version)	2013
J. Cole	Cole World: The Sideline Story	2011
Drake	For All The Dogs Scary Hours Edition	2023
Drake	For All The Dogs	2023
Drake	Her Loss	2022
Drake	Honestly, Nevermind	2022
Drake	Certified Lover Boy	2021
Drake	Dark Lane Demo Tapes	2020
Drake	Care Package	2019
Drake	Scorpion	2018
Drake	More Life	2017
Drake	Views	2016
Drake	What A Time To Be Alive	2015
Drake	If You're Reading This It's Too Late	2015
Drake	Nothing Was The Same (Deluxe)	2013
Drake	Nothing Was The Same	2013
Drake	Take Care (Deluxe)	2011
Drake	Thank Me Later	2010
Drake	So Far Gone	2009

Pretty cool, right? We were able to render album cover images inside of our table because the Spotify data contains hyperlinks to these images. Let’s see if we can also include an image of the artists in our table. We don’t have any links to head shots in our dataset, but maybe we can do a simple Google search to find an image we can import, render, and integrate into a new table.

# lets use the magick package to grab a single picture containing equally sized head shots for each member of the Big 3
big_three_pic <- image_read("https://hiphop-n-more.com/wp-content/uploads/2024/03/drake-cole-kendrick.jpg")

# let's grab drake (note that there are ways to do this using image analysis, but we're doing it manually here for simplicity)
drake_pic <- image_crop(big_three_pic, "485")       

# let's grab J. Cole as well... 
j_cole_pic <- image_crop(big_three_pic,"485x0+986")

# and finally... Kendrick...
kendrick_pic <- image_crop(big_three_pic,"485x0+490")

# let's write these images locally so we can pull them into gt
image_write(kendrick_pic, path = "kendrick_pic.png", format = "png")
image_write(drake_pic, path = "drake_pic.png",format = "png")
image_write(j_cole_pic, path = "j_cole.pic",format = "png")

# because we are using images that are stored locally, we will need tospecify image paths and reference these paths in our dataset
image_paths <- list(
  value1 = "kendrick_pic.png",
  value2 = "drake_pic.png",
  value3 = "j_cole.pic"
)

# let's conditionally add the image path for each artist to rows containing songs from that artist
albums_with_headshots <- albums %>%
  mutate(image_path = case_when(
    artist_name == "Kendrick Lamar" ~ image_paths$value1,
    artist_name == "J. Cole" ~ image_paths$value3,
    artist_name == "Drake" ~ image_paths$value2))

# we will need to include a specialized function to make this work for our images, since they are stored locally 
albums_with_headshots %>%
  gt(rowname_col = "artist_name") %>%
    text_transform(
    locations = cells_body(columns = "image_path"),
    fn = function(x) {
      # loop over the elements of the column
      map_chr(x, ~ local_image(
        filename = .,
        height = 50
      ))
    }) %>%
  cols_hide(c(height,width)) %>%
  tab_header(title = md("Artist Albums by Year"))%>%
  tab_stubhead(label = "Artist") %>%
  gt_img_rows(url,height = 50) %>%
  cols_label(
    image_path = "Artist Photo",
    album_release_year = "Release Year",
    artist_name = "Artist",
    album_name = "Album Name",
    url = "Album Cover") %>%
  cols_move_to_start(
    columns = c(image_path,url,album_name)) %>%
  cols_align(align = "center", columns = c("url","image_path","album_release_year"))

Artist	Album Name	Release Year
Artist Albums by Year
Kendrick Lamar	Mr. Morale & The Big Steppers	2022
Kendrick Lamar	Black Panther The Album Music From And Inspired By	2018
Kendrick Lamar	DAMN. COLLECTORS EDITION.	2017
Kendrick Lamar	DAMN.	2017
Kendrick Lamar	untitled unmastered.	2016
Kendrick Lamar	To Pimp A Butterfly	2015
Kendrick Lamar	good kid, m.A.A.d city (Deluxe)	2012
Kendrick Lamar	good kid, m.A.A.d city	2012
Kendrick Lamar	Section.80	2011
Kendrick Lamar	Overly Dedicated	2010
J. Cole	Might Delete Later	2024
J. Cole	D-Day: A Gangsta Grillz Mixtape	2022
J. Cole	The Off-Season	2021
J. Cole	Revenge Of The Dreamers III: Director's Cut	2020
J. Cole	Revenge Of The Dreamers III	2019
J. Cole	KOD	2018
J. Cole	4 Your Eyez Only	2016
J. Cole	Forest Hills Drive: Live from Fayetteville, NC	2016
J. Cole	2014 Forest Hills Drive	2014
J. Cole	Born Sinner (Deluxe Version)	2013
J. Cole	Cole World: The Sideline Story	2011
Drake	For All The Dogs Scary Hours Edition	2023
Drake	For All The Dogs	2023
Drake	Her Loss	2022
Drake	Honestly, Nevermind	2022
Drake	Certified Lover Boy	2021
Drake	Dark Lane Demo Tapes	2020
Drake	Care Package	2019
Drake	Scorpion	2018
Drake	More Life	2017
Drake	Views	2016
Drake	What A Time To Be Alive	2015
Drake	If You're Reading This It's Too Late	2015
Drake	Nothing Was The Same (Deluxe)	2013
Drake	Nothing Was The Same	2013
Drake	Take Care (Deluxe)	2011
Drake	Thank Me Later	2010
Drake	So Far Gone	2009

This looks better, but let me take you on a little side trip before we continue exploring the data. To give you a sense of how old I am, I started listening to Drake when he put out the So Far Gone Mixtape (not shown here). This came out prior to the release of the studio version of the album (his first album shown here). I also have fond memories of blasting songs from Section.80 (Kendrick Lamar’s first album) through the two ten-inch subwoofers I had in my trunk during my early twenties. Unfortunately, I didn’t stumble across J. Cole’s music until he released Born Sinner, but you get the idea. That’s enough nostalgia and “back in my day” talk for now. Let’s move on and make our audio features figure.

# let's start by creating a list of deluxe albums so we can remove them, since they are probably redundant with the non-deluxe versions 
deluxe_albums_to_remove <- c("Nothing Was The Same (Deluxe)","DAMN. COLLECTORS EDITION.","good kid, m.A.A.d city (Deluxe)")

# remove any albums in our list from the data
big_three_4 <- big_three_3 %>%
  filter(!album_name %in% deluxe_albums_to_remove)

# let's pivot the data from wide format to long format so we can make a pretty visualization
metrics <- big_three_4 %>%
  select(artist_name,album_name,album_release_year,track_name,danceability:tempo) %>%
  pivot_longer(cols = danceability:tempo,names_to = "metric",values_to = "score")

# lets make a list of the features we want
metrics_we_want <- c("tempo","loudness","acousticness","danceability","energy","liveness","speechiness","valence")
# visualize 

metrics %>%
  filter(metric %in% metrics_we_want) %>%
  ggplot(aes(x = score,y = artist_name,color = artist_name,fill = artist_name))+
  geom_density_ridges()+
  facet_wrap(~metric,ncol = 4,scales = "free_x")+
  ylab("Artist Name")+
  xlab("Score")+
  theme_classic()+
  theme(legend.position = "none",
        axis.title = element_text(size = 16),
        axis.text = element_text(size = 16),
        strip.text = element_text(size = 16),
        axis.text.x = element_text(size = 12,angle = 45, vjust = 1, hjust=1))

Well, it looks like there’s a wide range of scores across each audio feature, but the Big 3 are remarkably similar. I was surprised to see how similar the distribution of Kendrick Lamar’s scores were to Drake and J. Cole. If you’ve heard To Pimp a Butterfly, untitled unmastered., and DAMN, you know that Kendrick Lamar isn’t afraid to push the boundaries and experiment in his music, so I thought that might manifest in the distribution of scores across these audio features. I would argue, though, that a defining feature of the Big 3’s success is their willingness to try new things, push the boundaries, and invariably set new trends in hip hop, so I might not be giving J. Cole and Drake enough credit. Interesting, but the point is that the common scores across audio features as well as the amount of variation across scores are quite similar across all three artists.

These similarities might reflect aspects of their music that are attributable to being a hip hop artist. After all, these audio features are used to score music across the entire range of genres on Spotify, from Swedish Yodeling to Egyptian Black Metal (yes, both of those are real things). Assuming this is the case, we might expect to see a similar pattern for other hip hop artists as well, even ones who differ from the Big 3 in terms of their style and audience. Lets import and visualize audio features from another rapper’s music so we can compare their music with the Big 3, and let’s choose a rapper who appeals to a different listener base: 6ix9ine. 6ix9ine is an artist that many consider to be less talented than the members of the Big 3 (I know, I’m a hip hop elitist, I can’t help it), so it will be interesting to compare the distribution of scores when we include him in our visualizations. Let’s start by exploring 6ix9ine’s discography.

# import his data and then follow the same data cleaning steps above
six_nine <- get_artist_audio_features("6ix9ine") %>%
  select(-ends_with("_id"),-contains("url"),-contains("href")) %>%
  unnest(cols = c("album_images")) %>%
  select(!where(is.list))

# 
six_nine_2 <- six_nine %>%
  select(-height:-width) %>%
  distinct()

# let's see how many albums he has and how many tracks he has on them
six_nine_albums <- six_nine %>%
  select(artist_name,album_release_year,album_name,height,width,url) %>%
  filter(height == 64) %>%
  distinct()

# print albums
six_nine_albums %>%
  gt(rowname_col = "artist_name") %>%
  cols_hide(c(height,width,album_release_year)) %>%
  tab_header(title = md("Artist Albums by Year"))%>%
  tab_stubhead(label = "Artist") %>%
  gt_img_rows(url,height = 50) %>%
  cols_label(
    artist_name = "Artist",
    album_name = "Album Name",
    url = "Album Cover") %>%
  cols_move_to_start(
    columns = url) %>%
  cols_align(align = "center", columns = c("artist_name","url"))

Artist	Album Cover	Album Name
Artist Albums by Year
6ix9ine		BLACKBALLED
6ix9ine		Leyenda Viva
6ix9ine		TattleTales
6ix9ine		Day69: Graduation Day
6ix9ine		DUMMY BOY (Revised)

I mean, I don’t like his music, but I gotta give him credit for these album covers. These images are like the trap versions of the Candyland board game.

It’s also important to note that he has far fewer albums than members of the big three, so to make things fair, let’s randomly sample five albums from each of them and then compare those with albums from 6ix9ine.

# let's create an object with our randomly sampled albums
random_albums <- metrics %>%
  select(artist_name,album_name) %>%
  group_by(artist_name) %>%
  slice_sample(n = 5) %>%
  select(album_name)

# filter out albums that we didn't randomly sample
big_three_random_albums <- big_three_3 %>%
  filter(album_name %in% random_albums$album_name) 
  
all_data <- bind_rows(six_nine_2,big_three_random_albums) %>%
  select(artist_name,album_name,album_release_year,track_name,danceability:tempo) %>%
  pivot_longer(cols = danceability:tempo,names_to = "metric",values_to = "score") %>%
    mutate(artist_name = as.factor(artist_name),
           artist_name = fct_relevel(artist_name,c("Drake","J. Cole","Kendrick Lamar","6ix9ine")))
  
all_data %>%
  filter(metric %in% metrics_we_want) %>%
  ggplot(aes(x = score,y = artist_name,color = artist_name,fill = artist_name))+
  geom_density_ridges()+
  facet_wrap(~metric,ncol = 4,scales = "free_x")+
  ylab("Artist Name")+
  xlab("Score")+
  theme_classic()+
  theme(legend.position = "none",
        axis.title = element_text(size = 16),
        axis.text = element_text(size = 16),
        strip.text = element_text(size = 16),
        axis.text.x = element_text(size = 12,angle = 45, vjust = 1, hjust=1))+
  scale_y_discrete(breaks = c("6ix9ine","Kendrick Lamar","J. Cole","Drake"))

There do seem to be some differences. 6ix9ine’s music seems to be slightly lower in speechiness and slightly higher in danceability. That’s not entirely surprising. We also see that the distribution of responses are still pretty similar for the Big 3, even after randomly selecting five albums for each of them from their entire discographies.

This is great, but at this point, we’re really just reading tea leaves, so in Part 2 we will really dig into these and other metrics to quantify and statistically compare these artists’ music and maybe, just maybe, this will help us resolve the G.O.A.T. debate in hip hop. Stay tuned.

Posts | Academic

But really, who is the G.O.A.T? (Part 1)

What is it that defines good hip hop?