Spotify Data Visualization
Recently my teammates and I worked on a Spotify dataset for our Computation and Visualization class. We have tried to incorporate different types of visualizations ranging from time series analysis to word clouds. Hope you find it useful :)
The code to this project is also available on my github account — https://github.com/shrunalisalian/Spotify-Data-Visualization
This is my first post on Medium, so feel free to let me know if you have any suggestions. Happy studying!
You can download the dataset here https://www.kaggle.com/datasets/yamaerenay/spotify-dataset-19212020-600k-tracks This is a Data Visualization project in R.
We have divided the project into 3 domains.
- Trend Analysis
- Feature Analysis
- Time Series Analysis
There are a total of 12 visualizations, four in each category.
Before we begin, more about the dataset. There are 2 csv files giving information about — artist and tracks. The data ranges from 1920 to 2020. Since we have many artists who have not uploaded any tracks, the number of rows in the artists file is double that of the tracks file. We have combined the two files using an inner join.
Firstly we import all the required libraries for the visualization.
```{r loading libraries, include=TRUE,message=FALSE,warning=FALSE}
library(knitr)
library(dplyr)
library(tidyr)
library(stringr)
library(tidyverse)
library(reshape2)
library(ggplot2)
library(Hmisc)
library(magrittr)
library(lubridate)
```Reading the csv files into respective dataframes.
```{r reading data, include=TRUE,message=FALSE,warning=FALSE}
df1=read.csv("tracks.csv")
df=read.csv("artists.csv")
```Cleaning the data
```{r data cleaning begins, include=TRUE,message=FALSE,warning=FALSE}
#removing the brackets to unnest the the list of values from the artist_id
df1$id_artists = gsub("\\[", "", df1$id_artists)
df1$id_artists = gsub("\\]", "", df1$id_artists)
#splitting using the delimmeter ","
df1$id_artists = strsplit(df1$id_artists, ",")
df3 = tidyr::unnest(df1, id_artists)
df3$id_artists = gsub("'", "", df3$id_artists)
```Merging the two files using inner join
```{r using inner join to merge two csv files, include=TRUE,message=FALSE,warning=FALSE}
#using inner join joinning two table, artist and tracks
spotify_data <- inner_join(df3, df, by=c('id_artists'='id' ))
```Renaming all the parameters accordingly. I have shown two of the parameters here.
```{r renaming columns in new dataframe, include=TRUE,message=FALSE,warning=FALSE}
spotify_data <- spotify_data %>% rename("track_id" = "id")
spotify_data <- spotify_data %>% rename("track_name" = "name.x")
```Some more data cleaning
```{r data cleaning, include=TRUE,message=FALSE,warning=FALSE}
# adding a new attribute - track duration in minutes
spotify_data_cleaning <- spotify_data_cleaning %>% mutate(track_time = track_duration/60000)
spotify_data_cleaning <- spotify_data_cleaning %>% rename("track_time_mins" = "track_time")
# checking if any of the rows have all NA values
spotify_data_cleaning[rowSums(is.na(spotify_data_cleaning))!=ncol(spotify_data_cleaning), ]
# data column is not consist and has values like 1922 and 1922-01-01, in order to bring consistency we are adding "-01-01" to the dates given without month and day information
# date data ranges from 1921 - 2020
dates <- spotify_data_cleaning$track_release_date
dates <- as.POSIXct(dates, format = "%Y-%m-%d")
# first convert 1922 to 1922-01-01 and then convert it to date format and specify time zone
date_data <- spotify_data_cleaning$track_release_date
fixed_date <- gsub("1922","1922-01-01",date_data)
spotify_data_cleaning$track_release_date <- ifelse(nchar(spotify_data_cleaning$track_release_date) == 4, paste0(spotify_data_cleaning$track_release_date, "-01-01"), spotify_data_cleaning$track_release_date)
spotify_data_cleaning$track_release_date <- ymd(spotify_data_cleaning$track_release_date)
# release date done cleaning
# adding the changes to the original dataframe
spotify_data <- spotify_data_cleaning
```Since all the parameters are not equally scaled, we perform data scaling and bring the required features in the range of 0–1
```{r scaling track features, include=TRUE,message=FALSE,warning=FALSE}
# since the rest of the attributes have mean != 0, we think the best scaling approach should be min/max scaling
# min_max_scaled <- (data - min(data)) / (max(data) - min(data))
#track_key
key_min <- min(spotify_data_cleaning$track_key)
key_max <- max(spotify_data_cleaning$track_key)
key_range <- key_max - key_min
spotify_data_cleaning$track_key <- (spotify_data_cleaning$track_key - key_min)/key_range
#track_loudness
loudness_min <- min(spotify_data_cleaning$track_loudness)
loudness_max <- max(spotify_data_cleaning$track_loudness)
loudness_range <- loudness_max - loudness_min
spotify_data_cleaning$track_loudness <- (spotify_data_cleaning$track_loudness - loudness_min)/loudness_range
#track tempo
tempo_min <- min(spotify_data_cleaning$track_tempo)
tempo_max <- max(spotify_data_cleaning$track_tempo)
tempo_range <- tempo_max - tempo_min
spotify_data_cleaning$track_tempo <- (spotify_data_cleaning$track_tempo - tempo_min)/tempo_range
# track_time_signature
ts_min <- min(spotify_data_cleaning$track_time_signature)
ts_max <- max(spotify_data_cleaning$track_time_signature)
ts_range <- ts_max - ts_min
spotify_data_cleaning$track_time_signature <- (spotify_data_cleaning$track_time_signature - ts_min)/ts_range
spotify_data_cleaning <- spotify_data_cleaning[complete.cases(spotify_data_cleaning[,"track_release_date"]), ]
# summary(spotify_data_cleaning)
spotify_data <- spotify_data_cleaning
```Now that we have cleaned and scaled the data. Let’s dive into visualization.
Trend Analysis:
- Number of tracks on Spotify categorized year-wise
The aim of this visualization is to provide aquick and simple way to see how the number of tracks released by Spotifyhas changed over the years, broken down by decade. By seeing the size ofeach slice, it should be easy to quickly compare the number of tracksreleased in different decades and get an overall understanding of thedata. The donut chart depicts the evolution of the number of tracksreleased on Spotify, starting with 5304 tracks in 2006 and steadilyrising to 12331 tracks in 2020. However, there was a slight decrease inthe number of tracks released in 2017, which was almost a thousand lessthan the previous year.
```{r graph plots begin - donut chart, include=TRUE,message=FALSE,warning=FALSE}
spotify_performance <- spotify_data %>%
mutate(track_release_date = ymd(track_release_date))%>%
filter(track_release_date > ymd("2006-01-01"))
spotify_performance[c('Year','Month','Date')]<-str_split_fixed(spotify_performance$track_release_date,"-",3)
spotify_performance_plot <- spotify_performance[,c(2,3)]
spotify_performance <- spotify_performance %>%
group_by( Year) %>%
summarise(no_of_songs = n())
spotify_performance <- filter(spotify_performance, Year < "2021")
# Compute percentages
spotify_performance$fraction <- spotify_performance$no_of_songs/sum(spotify_performance$no_of_songs)
# Compute the cumulative percentages (top of each rectangle)
spotify_performance$ymax = cumsum(spotify_performance$fraction)
# Compute the bottom of each rectangle
spotify_performance$ymin = c(0, head(spotify_performance$ymax, n=-1) )
#compute label position
spotify_performance$labelPosition <- (spotify_performance$ymax+spotify_performance$ymin)/2
#compute label
spotify_performance$label <- paste(spotify_performance$Year, "\n value: ", spotify_performance$no_of_songs)
colors=rainbow(15)
ggplot(spotify_performance, aes(ymax = ymax, ymin=ymin, xmax=4, xmin=3,fill=Year))+
geom_rect()+
geom_label(x=3.5, aes(y=labelPosition, label=label), size=3.5) +
coord_polar(theta = "y")+
xlim(c(2,4))+
ggtitle("Trends of tracks released on Spotify")+
theme_void() +
theme(legend.position = "none") + scale_fill_manual(values = c("#9400D3","#FF69B4","#FFC0CB","#FFA07A","#E0FFFF","#ADD8E6","#87CEFA","#1E90FF","#20B2AA","#90EE90","#00FF7F","#7FFF90","#7FFF00","#DAA520","#FFD700"))
```The number of tracks uploaded on Spotify for each year has been consistently increasing. In 2013 we see a more than regular increase in the number of tracks being uploaded and the reason is, the prices of compact disc began to increase during the same time frame. This was combined with the growing popularity of streaming platforms. Next we saw a drop in the number of tracks uploaded for the first time in 2017. The reason for this is unclear. However, from what we know, Spotify went public in 2018 and recorded the largest ever operational costs in its history in 2017 due to worldwide expansion.
2. Comparing sentiments of the tracks released on Spotify before and during COVID-19
Hypothesis: Covid-19, also known as the novel coronavirus, ahighly infectious respiratory illness swept across the world, affectingmillions of people and causing widespread panic and disruption. The Covid-19 pandemic has had a profound impact on every aspect of life,from the economy and public health to the music industry. The pandemicnot only affected physical health but also had a significant impact onmental health, particularly in terms of depression. Thus, we conducted an experiment to find the type of tracks released by Spotify in theyears 2019 and 2020.
Aim: Do a comparative study for the years 2019 and 2020 inorder to find a relation between the type of songs released on Spotifyand people’s mental health.
```{r covid graph boxplot with jitters, include=TRUE,message=FALSE,warning=FALSE}
spotify_pandemic <- spotify_data %>%
mutate(track_release_date = ymd(track_release_date))%>%
filter(track_release_date > ymd("2019-01-01") &
track_release_date < ymd("2020-12-31"))
spotify_pandemic <- spotify_pandemic %>% group_by(track_release_date,track_valence)
spotify_pandemic$mood <- ifelse(spotify_pandemic$track_valence > 0.5, "Happy","Sad")
spotify_pandemic$covid_year <- ifelse(spotify_pandemic$track_release_date < "2020-01-01", "Before Covid","Covid")
ggplot(spotify_pandemic, aes(x = covid_year, y = track_valence, colour = mood)) +
geom_boxplot(outlier.shape = NA) +
geom_jitter()+ggtitle("Comparing Emotional Tone in Music Before and During the COVID-19 Pandemic")
```The above graph is a boxplot with jitters plottedfor two time periods — pre-covid and covid. In order to analyze therelation between type of tracks and the pandemic we have made use of themetric — valence. Valence — A measure from 0.0 to 1.0 describing the musicalpositiveness conveyed by a track. Tracks with high valence sound morepositive (e.g. happy, cheerful, euphoric), while tracks with low valencesound more negative (e.g. sad, depressed, angry). During our analysis we found correlation between valence and other features. There is significant correlation between the following pairs
- valence — danceability
- 2) valence — loudness
- 3) valence — energy
Contrary to our belief we see no significant change in the type of tracks released on Spotify. Hence, our hypothesis was incorrect.
3. What drives popularity of a track? Can we predict the trend of a track released on Spotify?
```{r HEXBIN PLOT, include=TRUE,message=FALSE,warning=FALSE}
spotify_track_artist <- spotify_data[,c(3,21,24)]
# colnames(spotify_track_artist)
# Load the hexbin library
library(hexbin)
# Create the hexbin object
hexbin_obj <- hexbin(y = spotify_track_artist$track_popularity, x = spotify_track_artist$artist_popularity, xbins = 15)
# Create the hexbin plot
ggplot(data = spotify_track_artist, aes(x=spotify_track_artist$artist_popularity,y=spotify_track_artist$track_popularity))+geom_hex(bins=18)+
scale_fill_gradient(low = "white", high = "red")+
labs(title = " Hexbin Plot for track popularity vs artist popularity", x="artist popularity",y = "track popularity")
```This Hex Binplot helps us in understanding the relation between the popularity of a track and the popularity of an artist. Hexbin plots take in lists of X and Y values and returns what lookssomewhat similar to a scatter plot, the entire graphing space has beendivided into hexagons (like a honeycomb) and all points have beengrouped into their respective hexagonal regions with a color gradientindicating the density of each hexagonal area. From this graph we undestand that highest majority of the artists have very low or negligible popularity and hence popularity of theirtracks is low. At the same time, artists with high popularity have hittracks. Thus if a new song is released on Spotify, given that the track artist is popular the track has a good chance of becoming a hit.
Feature Analysis
- Relationship between the features of a track
```{r heatmap - graph-1 relation between song characteristics, include=TRUE,message=FALSE,warning=FALSE,echo=FALSE}
df_sample=spotify_data_cleaning%>%select(track_danceability:track_time_signature)
library(ggplot2)
library(reshape2)
library(plotly)
correlation_matrix <- cor(df_sample)
# Melt the correlation matrix into a dataframe
melted_correlation_matrix <- melt(correlation_matrix)
melted_correlation_matrix<-melted_correlation_matrix%>%rename("Track_characteristics"="Var1")
melted_correlation_matrix<-melted_correlation_matrix%>%rename("Track_characteristics_1"="Var2")
# Plot the heatmap
p<-ggplot(data=melted_correlation_matrix, aes(x=Track_characteristics, y=Track_characteristics_1, fill=value)) +
geom_tile(color="black") + geom_text(aes(label = value), color = "black", size = 1)+
scale_fill_gradientn(colors = hcl.colors(3, "RdYlGn")) +
theme_minimal() +
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, size = 12, hjust=1))+ggtitle("Feature Analysis - Correlation Matrix")
coord_fixed()#for square tiles i have use coord_fixes() and heatmap can be formed with the geom_title itself
ggplotly(p)
```We see a significant correlation between
a) i) valence — energy ii) valence — danceability iii) valence — loudness
b) Energy — Loudness The track loudness and energy are directly proportional.
c) What factors contribute to the tracks popularity for a popular artist? Track characteristics do not play a significant role in determining the popularity of an artist’s tracks.
d) Popular tracks of a leading artist Word cloud visualization indicates that the track titles “hold_on” and “anyone” appear frequently for the artist Justin Bieber.
e) What variation in track characteristics can be observed across the top 10 music genres? The “kleine hoerspie” genre possesses the highest values for several track features, including “track_danceability,” “track_energy,” “track_key” , “track_loudness,” “track_mode,” “track_speechiness,” and “track_acousticness.” This allows us to observe the variations in each track characteristic among the top 10 music genres.
Analyzing the correlation between energy and loudness further.
```{r scatterplot to show correlation between track loudness and energy,include=TRUE,message=FALSE,warning=FALSE}
library(ggplot2)
library(RColorBrewer)
ggplot(data=spotify_data, aes(x=track_loudness, y=track_energy, color=track_mode)) +
geom_point(size=1) +
# scale_color_gradientn(colors=magma(5)) +
theme(legend.position="none")+ggtitle("Correlation between track energy and track loudness")
```The plot showcases the correlation between different characteristicsof the song tracks, indicating how they are related to one another. Ifthe values are close to 1, then the features are positively correlatedand are depicted in green, such as between “track_loudness” and”track_energy.” On the other hand, if the values are close to -1, then the features are negatively correlated and are shown in red, forinstance, between “track_acousticness” and both “track_energy” and”track_loudness”.
2. What factors contribute to the popularity of the track of a popular artist?
```{r , include=TRUE,echo=FALSE,message=FALSE,warning=FALSE}
df_s=spotify_data_cleaning%>%filter(artist_popularity==max(artist_popularity))
df_s[c('Year','Month','Date')]<-str_split_fixed(df_s$track_release_date,"-",3)
df_s
df_s$track_popularity <- df_s$track_popularity /100
p1<-ggplot(df_s, aes(x =track_popularity , y = track_loudness)) +
geom_line() +
scale_color_discrete(name = "track_popularity") +
xlab("track_popularity") +
ylab("Variable 2") +
ggtitle("Scatter Plot of track_popularity and track_loudness")+ geom_smooth(formula = y ~ x,method = "lm")
p2<-ggplot(df_s, aes(x =track_popularity , y = track_danceability)) +
geom_line() +
scale_color_discrete(name = "track_popularity") +
xlab("track_popularity") +
ylab("track_danceability") +
ggtitle("Scatter Plot of track_popularity and track_danceability")+ geom_smooth(formula = y ~ x,method = "lm")
p3<-ggplot(df_s, aes(x =track_popularity , y = track_energy)) +
geom_line() +
scale_color_discrete(name = "track_popularity") +
xlab("track_popularity") +
ylab("track_energy") +
ggtitle("Scatter Plot of track_popularity and track_energy")+ geom_smooth(formula = y ~ x,method = "lm")
p4<-ggplot(df_s, aes(x =track_popularity , y = track_valence)) +
geom_line() +
scale_color_discrete(name = "track_popularity") +
xlab("track_popularity") +
ylab("track_valence") +
ggtitle("Scatter Plot of track_popularity and track_valence")+ geom_smooth(formula = y ~ x,method = "lm")
p5<-ggplot(df_s, aes(x =track_popularity , y = track_tempo)) +
geom_line() +
scale_color_discrete(name = "track_popularity") +
xlab("track_popularity") +
ylab("track_tempo") +
ggtitle("Scatter Plot of track_popularity and track_tempo")+ geom_smooth(formula = y ~ x,method = "lm")
p6<-ggplot(df_s, aes(x =track_popularity , y = track_instrumentalness)) +
geom_line() +
scale_color_discrete(name = "track_popularity") +
xlab("track_popularity") +
ylab("track_instrumentalness") +
ggtitle("Scatter Plot of track_popularity and track_instrumentalness")+ geom_smooth(formula = y ~ x,method = "lm")
p7<-ggplot(df_s, aes(x =track_popularity , y = track_acousticness)) +
geom_line() +
scale_color_discrete(name = "track_popularity") +
xlab("track_popularity") +
ylab("track_acousticness") +
ggtitle("Scatter Plot oftrack_popularity and track_acousticness")+ geom_smooth(formula = y ~ x,method = "lm")
p8<-ggplot(df_s, aes(x =track_popularity , y = track_speechiness)) +
geom_line() +
scale_color_discrete(name = "track_popularity") +
xlab("track_popularity") +
ylab("track_speechiness") +
ggtitle("Scatter Plot of Two Continuous Variables and track_speechiness")+ geom_smooth(formula = y ~ x,method = "lm")
p9<-ggplot(df_s, aes(x =track_popularity , y = track_mode)) +
geom_line() +
scale_color_discrete(name = "track_popularity") +
xlab("track_popularity") +
ylab("track_mode") +
ggtitle("Scatter Plot of track_popularity and track_mode")+ geom_smooth(formula = y ~ x,method = "lm")
p10<-ggplot(df_s, aes(x =track_popularity , y = track_key)) +
geom_line() +
scale_color_discrete(name = "track_popularity") +
xlab("track_popularity") +
ylab("track_key") +
ggtitle("Scatter Plot of track_popularity and track_key")+ geom_smooth(formula = y ~ x,method = "lm")
p11<-ggplot(df_s, aes(x =track_popularity , y = track_liveness)) +
geom_line() +
scale_color_discrete(name = "track_popularity") +
xlab("track_popularity") +
ylab("track_liveness") +
ggtitle("Scatter Plot of track_popularity and track_liveness")+ geom_smooth(formula = y ~ x,method = "lm")
library(gridExtra)
library(grid)
grid.arrange(p1,p2, p3, p4 ,p5,p6,p7,p8,p9,p10,p11,nrow=4, ncol=3)
```The line graph indicates that track characteristics do not play asignificant role in determining the popularity of an artist’s tracks.
3. What is the occurrence rate of track names for a leading artist?
```{r graph-4 analysing frequency of tracks names of an popular artist over the years using wordcloud,include=TRUE,message=FALSE,warning=FALSE}
df_s_1=spotify_data_cleaning
df_s_1[c('Year','Month','Date')]<-str_split_fixed(df_s_1$track_release_date,"-",3)
df_s_1<-df_s_1%>%filter(artist_popularity==max(artist_popularity))
df_s_1_grp<-df_s_1%>%filter(artist_popularity==max(artist_popularity))%>%group_by(track_name)%>%summarise(total_popularity=sum(track_popularity))
c<-df_s_1%>%group_by(track_name)%>%summarise(num=n())
df_s_1 <- df_s_1 %>%
mutate(track_name = gsub(" ", "_", track_name))
#install.packages("tm")
library(tm)
#Create a vector containing only the text
text <- df_s_1$track_name
# Create a corpus
docs <- Corpus(VectorSource(text))
dtm <- TermDocumentMatrix(docs)
matrix <- as.matrix(dtm)
words <- sort(rowSums(matrix),decreasing=TRUE)
df <- data.frame(word = names(words),freq=words)
set.seed(1234) # for reproducibility
#install.packages("wordcloud")
library(wordcloud)
#install.packages("RColorBrewer")
library(RColorBrewer)
suppressWarnings(wordcloud(words = df$word, freq = df$freq, min.freq = 1,
max.words=500, random.order=FALSE, rot.per=0.35,
colors=brewer.pal(8, "Dark2")))
```The wordcloud visualization indicates that the track titles “hold_on”and “anyone” appear frequently for the artist Justin Bieber. The fontsize in the wordcloud allows us to determine which track titles occurmost frequently and which do not. Hence, the wordcloud visual providesinformation about the occurrence rate of track names for a leadingartist.
4. What variation in track characteristics can be observed across the top 10 music genres?
```{r graph-2 stacked bar graph how the track characteristics changes based on genres(top),include=TRUE,message=FALSE,warning=FALSE}
library(dplyr)
library(tidyr)
library(stringr)
library(tidyverse)
library(reshape2)
genre_data <- spotify_data_cleaning%>%filter(artist_genres!="[]")%>%group_by(artist_genres)%>%summarise(sum_popularity = sum(track_popularity))%>%top_n(10, sum_popularity)%>%arrange(desc(sum_popularity))%>%rename(genres=artist_genres)
df_sample=spotify_data_cleaning%>%select('track_danceability','track_energy','track_key','track_loudness','track_mode','track_speechiness','track_acousticness','track_instrumentalness','track_liveness','track_valence','track_tempo','track_time_signature','artist_genres')
long_df <- melt(df_sample, id.vars = "artist_genres")
grouped_df <- long_df %>%group_by(artist_genres,variable) %>%summarise(average = mean(value))
df_sample_1<-inner_join( genre_data,grouped_df, by=c('genres'='artist_genres' ))
colors <- c("#0429b3", "#2853b2", "#3c76b1", "#5199b0", "#65bcaf", "#79dfae", "#8df2ad", "#a1f5ac", "#b5f8ab", "#c9fbaa", "#deffa9", "#f2ffa8")
#colors <- c("#0a3d91", "#186faf", "#2da4d4", "#48c7e0", "#63e1eb", "#7ff5f5", "#9bf9ff", "#b7feff", "#d2ffff", "#edffff", "#f7f7f7", "#ffffff")
p<-ggplot(df_sample_1, aes(x =genres, y = average, fill = variable)) +
geom_bar(stat = "identity") +
ggtitle("Stacked Bar Plot of Two Categorical Variables and One Continuous Variable") +
xlab("genres") +
ylab("average")+
theme(axis.text.x = element_text(angle = 90, vjust = 1, hjust=1))+scale_fill_manual(values = colors)
ggsave("plot.png", p, width = 10, height = 10, units = "in")
p
```According to the Stacked bar chart, it can be noted that the “kleinehoerspie” genre possesses the highest values for several track features,including “track_danceability,” “track_energy,” “track_key,””track_loudness,” “track_mode,” “track_speechiness,” and”track_acousticness.” This allows us to observe the variations in eachtrack characteristic among the top 10 music genres.
Time Series Analysis
- What is the inclination of Spotify users towards particular artists, and does the year of release play a role in shaping these preferences?
```{r,include=TRUE,message=FALSE,warning=FALSE}
spotify_data$track_release_year = substr(spotify_data$track_release_date, 1, 4)
spotify_data$track_release_year<- as.numeric(spotify_data$track_release_year)
library(ggplot2)
library(dplyr)
# Group the data by decades
spotify_data_grouped <- spotify_data %>%
mutate(track_release_decade = as.integer(floor(track_release_year / 10) * 10)) %>%
group_by(track_release_decade)
spotify_data_grouped$track_release_decade<- as.numeric(spotify_data_grouped$track_release_decade)
spotify_data_grouped<- spotify_data_grouped[complete.cases(spotify_data_grouped), ]
spotify_data_grouped <- subset(spotify_data_grouped,track_release_decade!=1900 )
library(ggplot2)
df_filtered_1 <- spotify_data_grouped %>%
distinct(artist_name, .keep_all = TRUE)
# Get the top 5 artists by popularity for each decade
df_top5_artist <- df_filtered_1 %>%
group_by(track_release_decade) %>%
top_n(5, artist_popularity) %>%
ungroup()
# Plot the dot plot
ggplot(df_top5_artist, aes(x = reorder(artist_name, track_release_decade), y = artist_popularity)) +
geom_point(aes(color = as.factor(track_release_decade))) +
scale_color_discrete(name = "Release Decade") +
ggtitle("Top 5 Unique Artists by Popularity for Each Decade (1920-2020)") +
xlab("Artist Name") +
ylab("Popularity") +
coord_flip()
```We plot the top 5 artists in each decade between 1920–2020. This plot also shows an increase in popularity for artists in more recent yearscompared to those in the early ones.This could be due to a few differentfactors such as the evolution of the music industry in terms of genrepreferences, better technology used to create music, globalization ofpopular genres, etc.
2. What is the inclination of Spotify users towards different genres of music? Does the decade that this genre originatedhave an impact on whether people still listen to it?
```{r,include=TRUE,message=FALSE,warning=FALSE}
genre_data <- spotify_data_grouped %>%
mutate(artist_genres = str_remove_all(artist_genres, "[\\[\\]']")) %>%
separate_rows(artist_genres, sep = ",")
genre_data$artist_genres <- ifelse(genre_data$artist_genres == "", NA, genre_data$artist_genres)
genre_data<- genre_data[complete.cases(genre_data), ]
genre_data <- group_by(genre_data, track_release_decade, artist_genres)
genre_data <- summarise(genre_data, popularity = mean(artist_popularity))
genre_data <- genre_data %>%
group_by(track_release_decade) %>%
top_n(5, popularity)
```
```{r,include=TRUE,message=FALSE,warning=FALSE}
ggplot(genre_data, aes(x = track_release_decade, y = reorder(artist_genres, track_release_decade), fill = popularity)) +
geom_tile() +
xlab(label ="Decade")+
ylab(label ="Genre")+
scale_x_continuous(breaks=seq(1920, 2020, by=10)) +
scale_fill_gradient(name = "Popularity",low = "green", high = "black") +
theme_minimal()
```We plot the top 5 genres by popularity that originated in each decadebetween 1920–2020. What is interesting to note is that the most populargenres of all time on Spotify are those that originated in the 2000s. From the previous plot(PLOT1), we saw that artists of older decades didnot fare well in terms of popularity, but the same trend does not applyto genres. This shows that artists of recent years have adapted older genres into their style of music to make it more appealing to theaudience.
3. What is the popular attitude towards explicit contentin music?
```{r,include=TRUE,message=FALSE,warning=FALSE}
grouped_data <- spotify_data_grouped %>%
group_by(track_release_decade) %>%
summarise(mean_popularity = mean(track_popularity),
mean_explicit = mean(track_explicit))
ggplot(grouped_data, aes(x = track_release_decade, y = mean_explicit, fill = mean_popularity)) +
geom_col(show.legend = FALSE) +
scale_fill_gradient(low = "green", high = "black") +
scale_x_continuous(breaks=seq(1920, 2020, by=10)) +
labs(x = "Decade", y = "Mean Explicitness",
title = "Explicitness of Music Over the Decades",
subtitle = "1920-2020") +
theme_minimal()
```To understand this better, we plot the mean for explicitness over thedecades from 1920–2020. The mean in this case would represent theproportion of 1s (explicit) in the data set. So, for a given decade, ifthe mean of the explicitness variable is 0.6, this would mean that 60%of the tracks in the data set for that decade are explicit. Byaggregating the mean explicitness per decade, we can see how theproportion of explicit tracks has changed over time and whether thereare any trends or shifts in attitudes towards explicit content.Thisshows us that the cultural attitude towards explicit music and anartist’s freedom of expression has increased largely, with most popularsongs in the recent decades of the 2000s having popular tracks withexplicit lyrics.
To further understand the portion of explicit lyrics in top songs ofthe artists that have dominated each decade of release, we plot theexplicitness for the top song of each artist by decade.
```{r,include=TRUE,message=FALSE,warning=FALSE}
top5_artist_song <- df_top5_artist %>%
group_by(track_release_decade, artist_name) %>%
top_n(1,track_popularity) %>%
ungroup()
top5_artist_song$track_explicit <- ifelse(top5_artist_song$track_explicit == 0, "no", "yes")
ggplot(top5_artist_song, aes(x = track_release_decade, y = reorder(artist_name, track_release_decade), color = track_explicit)) +
geom_point(size = 2) +
scale_color_manual(values = c("green", "black"), guide = FALSE) +
labs(x = "Decade", y = "Artist Name", color = "Explicitness") +
ggtitle("Explicitness of the Top Songs of Top 5 Artists based on the Decade the song was released") +
theme_minimal()
```We notice that most popular artists of recent decades do not in facthave explicit lyrics in their top tracks. This shows us that thoughexplicit music is popular, it is not a necessity for the success of thesong. This could be due to various factors such as the demographic ofthe audience, Spotify’s recommendation algorithms, etc.
In conclusion, our data visualization project based on Spotify’s track data has revealed several interesting insights into the music industry. Through the use of various graphical representations, we were able tounderstand the distribution of different audio features of songs, thepopularity of different genres, and the trend of listening behavior overtime. Our findings showed that danceability, energy, and tempo areimportant factors that contribute to the popularity of a song, whilehip-hop and pop music are the most streamed genres on Spotify. Moreover, we observed a steady growth in the number of streams over the years,indicating an increasing trend in the usage of Spotify and digital musicstreaming services.
In summary, this project has provided a comprehensive view of themusic industry and the trends that are shaping it. By utilizing datavisualization techniques, we have been able to effectively communicatecomplex information and derive meaningful insights. We hope that thesefindings will be useful for artists, music labels, and otherstakeholders in the industry to make informed decisions and shape thefuture of music.
