10_005_analyzing-my-streaming-history.Rmd

---
title: "Analyzing my streaming history"
description: |
  Identifying trends in my listening habits.
author:
  - first_name: "Joshua"
    last_name: "Cook"
    url: https://joshuacook.netlify.app
    orcid_id: 0000-0001-9815-6879
date: "`r Sys.Date()`"
output:
  distill::distill_article:
    toc: true
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = FALSE, comment = "#>", dpi = 500)

library(nakedpipe)
library(magrittr)
library(ggthemes)
library(ggrepel)
library(tidyverse)

theme_set(ggthemes::theme_fivethirtyeight())
pal_538 <- deframe(ggthemes_data[["fivethirtyeight"]])

walk(list.files("lib", pattern = "R$", full.names = TRUE), source)

streaming_history <- readRDS(here::here("data", "streaming_history.rds"))

min_song_length <- 60 * 1
```

The following is a dive into my streaming history from the past year or so.
For each track, I am given the time and date, artist name, track name, and for how long the track was played.
The data was processed in [Preparing the downloaded files](05_002_spotify-raw-data-preparation.html).

---

## Time spent listening to music

To begin, we can look at the distribution of the durations of the tracks.
For reference, I have annotated the plot with 3 of my favorite songs: [*Stacy*](https://open.spotify.com/track/07ZQLYn9x4x3L3vxStc1zr?si=PKUhdcYLQXmaiyU79C-HmQ) by [Quinn XCII](https://open.spotify.com/artist/3ApUX1o6oSz321MMECyIYd?si=aBfY9yDrSmmhxe1F8ThFww), [*Good Things Fall Apart*](https://open.spotify.com/track/3LxG9HkMMFP0MZuiw3O2rF?si=T9_Hg7_pRnWRGlctI9f5aQ) by [Illenium](https://open.spotify.com/artist/45eNHdiiabvmbp4erw26rg?si=2ZdCEqJFT2KU3PZ_iB71Vw) featuring [Jon Bellion](https://open.spotify.com/artist/50JJSqHUf2RQ9xsHs0KMHg?si=2SMZIa-MS3mHU0BLmsolkg), and [*Choices*](https://open.spotify.com/track/65z28RA7Nmzsog6Noln0sB?si=suGmKtjtREmIINnBU3HrJg) by [To Kill a King](https://open.spotify.com/artist/7yHBGaYHPXkJSVUtHiM816?si=iHFiAoUHRD2xBPQc1EgPUA) (you can listen to samples of these songs at the bottom of the page).

```{r}
notable_lengths <- tibble(
  song = c("Stacy", "Good Things Fall Apart", "Choices"),
  length = c("2:49", "3:37", "6:28"),
  x = c(169, 217, 388),
  y = c(0.010, 0.010, 0.0112)
) %.% {
  mutate(length = as.numeric(lubridate::ms(length)))
}

streaming_history %>%
  ggplot(aes(x = sec_played)) +
  geom_density(fill = pal_538["Medium Gray"], alpha = 0.7) +
  geom_rect(
    data = tibble(sec_played = 0),
    xmin = 0, xmax = min_song_length,
    ymin = 0, ymax = Inf,
    fill = "grey20",
    alpha = 0.2,
    color = NA
  ) +
  geom_vline(
    xintercept = min_song_length,
    color = "grey35",
    lty = 2,
    alpha = 0.8
  ) +
  geom_vline(
    aes(xintercept = length, color = song),
    data = notable_lengths,
    size = 1,
    alpha = 0.8
  ) +
  geom_label(
    aes(label = song, x = x, y = y, color = song),
    data = notable_lengths,
    hjust = c(1, 0, 0),
    nudge_x = c(-10, 10, 10),
    alpha = 0.7,
    label.size = 0,
    family = "sans",
    fontface = "bold"
  ) +
  scale_x_continuous(
    expand = expansion(mult = c(0, 0.02)),
    limits = c(0, 600),
    breaks = seq(0, 10, 2) * 60,
    labels = function(x) {
      round(x / 60, 1)
    }
  ) +
  scale_y_continuous(
    expand = expansion(mult = c(0, 0)),
    limits = c(0, 0.012)
  ) +
  scale_color_fivethirtyeight() +
  theme(
    axis.title = element_text(),
    legend.position = "none"
  ) +
  labs(
    x = "time spent listening to a song (min)",
    y = "density",
    title = "The distribution of play time",
    subtitle = paste(
      "Most songs were listened to for about 2-5 minutes. Some example song",
      "durations are labeled. The far-left peak (highlighted in grey) represents",
      "songs that were quickly skipped.",
      sep = "\n"
    )
  )
```

The durations range from 0 to around 10 minutes.
Presumably, the short durations correspond to skipping a track at the beginning.
For the purposes of further analysis, I will set the minimum length of a song as `r round(min_song_length, 1)` seconds, removing data points with shorter durations.
The longer songs were likely background noise tracks for various wilderness albums or coffee shop sounds.

We can also see how much time I spent listening to music by plotting the cumulative duration over time.
By the gradual reduction on the slop over time, we can see that I reduced the amount of time I listened to music over the course of the year.

```{r}
streaming_history %.%
  {
    mutate(
      cum_sec_played = cumsum(sec_played),
      cum_hour_played = cum_sec_played / 60 / 60
    )
  } %>%
  ggplot(aes(x = end_time, y = cum_hour_played)) +
  geom_ribbon(
    aes(ymin = 0, ymax = cum_hour_played),
    alpha = 0.2,
    fill = spotify_dark
  ) +
  geom_line(size = 1, color = spotify_green) +
  scale_x_datetime(
    date_breaks = "2 months",
    date_labels = "%b %Y",
    expand = expansion(add = c(0, 0))
  ) +
  scale_y_continuous(expand = expansion(mult = c(0, 0.02))) +
  theme(
    axis.title.y = element_text()
  ) +
  labs(
    y = "time (hours)",
    title = "Cumulative time spent listening to Spotify"
  )
```

```{r}
streaming_history %<>% filter(sec_played >= min_song_length)
```

This trend is also supported by the number of tracks I listened to per day.
The smoothened trend line indicates a lull in the fall of 2020.

```{r}
streaming_history %.%
  {
    mutate(date = lubridate::floor_date(end_time, "day"))
    count(date)
  } %>%
  ggplot(aes(x = date, y = n)) +
  geom_point(alpha = 0.8) +
  geom_smooth(
    method = "loess",
    formula = "y~x",
    color = spotify_green
  ) +
  scale_x_datetime(
    date_breaks = "2 months",
    date_labels = "%b %Y",
    expand = expansion(add = c(0, 0))
  ) +
  theme(
    axis.title.y = element_text()
  ) +
  labs(
    y = "number of songs",
    title = "Daily number of songs listened to"
  )
```

The variability of my listening patterns also followed the same general trend.
As the plot of the number of different artists and tracks per month below indicates, I had a peak of diversity in April and May of 2020 followed by a general decline.

```{r}
sec_axis_coef <- 2
artist_color <- pal_538[["Red"]]
track_color <- pal_538[["Blue"]]

streaming_history %.%
  {
    mutate(mo_year = lubridate::floor_date(end_time, "month"))
    group_by(mo_year)
    summarise(
      num_artists = n_distinct(artist_name),
      num_tracks = n_distinct(track_name)
    )
    ungroup()
    mutate(num_tracks = num_tracks / sec_axis_coef)
  } %>%
  ggplot(aes(x = mo_year)) +
  geom_line(aes(y = num_artists), color = artist_color, alpha = 0.7) +
  geom_line(aes(y = num_tracks), color = track_color, alpha = 0.7) +
  geom_point(aes(y = num_artists), color = artist_color) +
  geom_point(aes(y = num_tracks), color = track_color) +
  scale_y_continuous(
    name = "num. artists",
    sec.axis = sec_axis(~ . * sec_axis_coef, name = "num. tracks")
  ) +
  scale_x_datetime(
    date_breaks = "2 months",
    date_labels = "%b %Y"
  ) +
  theme(
    axis.title.y = element_text(color = colorspace::darken(artist_color)),
    axis.title.y.right = element_text(color = colorspace::darken(track_color))
  ) +
  labs(
    title = "Variability in artists and songs",
    subtitle = "The number of different artists (red) and songs (blue) listened to each month."
  )
```

### Favorites

Next, I want to look into who my most played artists and songs have been.
The next two plots show the top artists and individual songs from my streaming history, and the third plot shows the top songs from each of my favorite artists.

```{r}
plot_barcount <- function(df, x, y,
                          x_lbl, title,
                          nudge_text = 11,
                          expand_x = 0.08,
                          wrap_y = Inf) {
  df %>%
    mutate(label = scales::comma({{ x }}, accuracy = 1)) %>%
    ggplot(aes(x = {{ x }}, y = {{ y }})) +
    geom_text(
      aes(label = label),
      hjust = 0,
      family = "Helvetica",
      size = 3.5,
      nudge_x = nudge_text
    ) +
    scale_x_continuous(expand = expansion(c(0, expand_x))) +
    scale_y_discrete(labels = scales::label_wrap(wrap_y)) +
    theme_fivethirtyeight() +
    theme(
      panel.grid.major.y = element_blank(),
      axis.title.x = element_text()
    ) +
    labs(
      x = x_lbl,
      y = NULL,
      title = title
    )
}

streaming_history %.%
  {
    count(artist_name, sort = TRUE, name = "n")
    head(10)
    mutate(artist_name = fct_rev(fct_inorder(artist_name)))
  } %>%
  plot_barcount(
    x = n,
    y = artist_name,
    x_lbl = "number of times listened",
    title = "Top-10 most listened to artists",
    expand_x = 0.11,
    wrap_y = 30
  ) +
  geom_col(fill = spotify_dark2, color = NA)
```

```{r}
streaming_history %.%
  {
    count(track_name, artist_name, sort = TRUE, name = "n")
    head(10)
    mutate(track_name = fct_rev(fct_inorder(track_name)))
  } %>%
  plot_barcount(
    x = n,
    y = track_name,
    x_lbl = "number of times listened",
    title = "Top-10 most listened to songs",
    nudge_text = 2,
    expand_x = 0.06,
    wrap_y = 30
  ) +
  geom_col(aes(fill = artist_name), color = NA) +
  scale_fill_excel_new(
    "Atlas",
    labels = function(x) {
      str_wrap(x, width = 25)
    },
    guide = guide_legend(nrow = 1, title = "Artist")
  ) +
  theme(
    legend.margin = margin(),
    legend.title = element_text(size = 11),
    legend.key.size = unit(4, "mm")
  )
```

Interestingly, my top songs are dominated by [Jon Bellion](https://open.spotify.com/artist/50JJSqHUf2RQ9xsHs0KMHg?si=2SMZIa-MS3mHU0BLmsolkg) though my most played artist was [Imagine Dragons](https://open.spotify.com/artist/53XhwfbYqKCa1cC15pYq2q?si=GcRvAHkmTCuCChoe1nJULA).
This likely indicates that I listened to many different songs from [Imagine Dragons](https://open.spotify.com/artist/53XhwfbYqKCa1cC15pYq2q?si=GcRvAHkmTCuCChoe1nJULA) while I tended to listen to fewer [Jon Bellion](https://open.spotify.com/artist/50JJSqHUf2RQ9xsHs0KMHg?si=2SMZIa-MS3mHU0BLmsolkg) songs very frequently.

```{r, fig.height=6}
ignore_top_artists <- c(
  "Background Noise From TraxLab", "Ambient Sounds from I'm In Records"
)

top_artists <- streaming_history %.% {
  filter(!artist_name %in% !!ignore_top_artists)
  count(artist_name, sort = TRUE)
  head(6)
  pull(artist_name)
  unlist()
  unique()
}

streaming_history %.%
  {
    filter(artist_name %in% !!top_artists)
    group_by(artist_name)
    count(track_name, sort = TRUE)
    slice(1:5)
    ungroup()
    mutate(track_name = fct_rev(fct_inorder(track_name)))
  } %>%
  plot_barcount(
    x = n,
    y = track_name,
    x_lbl = "number of times listened",
    title = "Top songs of my favorite artists",
    nudge_text = 3,
    expand_x = 0.20,
    wrap_y = 30
  ) +
  geom_col(fill = spotify_dark2, color = NA) +
  facet_wrap(~artist_name, ncol = 2, scales = "free") +
  theme(strip.text = element_text(face = "bold"))
```

Plotting the number of times I listened to my top songs in each month, [*Good Things Fall Apart*](https://open.spotify.com/track/3LxG9HkMMFP0MZuiw3O2rF?si=uY4xje_uSi24UUAB21wcbg) (orange) peaked in February, probably after I first discovered it in late January.
Then, In April, I got hooked on [*Pre-Occupied*](https://open.spotify.com/track/0i4Zkb6Ts5JqLJIAMNzSSl?si=Tz9423Y9ThSfHQK5UBcJkA) by [Jon Bellion](https://open.spotify.com/artist/50JJSqHUf2RQ9xsHs0KMHg?si=iURyF8AzTCOL86CmmGnJ3w) (purple).

```{r}
top_songs <- streaming_history %.% {
  count(artist_name, track_name, sort = TRUE)
  head(6)
  select(track_name, artist_name)
}

streaming_history %.%
  {
    mutate(date = lubridate::floor_date(end_time, unit = "month"))
    count(date, artist_name, track_name)
    right_join(top_songs, by = c("track_name", "artist_name"))
  } %>%
  ggplot(aes(x = date, y = n, color = track_name)) +
  geom_line() +
  geom_point() +
  scale_color_excel_new(
    "Atlas",
    labels = function(x) {
      str_wrap(x, width = 25)
    },
    guide = guide_legend(nrow = 2, title = NULL)
  ) +
  scale_x_datetime(date_breaks = "2 months", date_labels = "%b %Y") +
  labs(
    title = "The frequency of my top songs over time",
    subtitle = "The number of times each song was played per month."
  )
```

Below is a similar plot but showing how often I listened to my favorite artists each month.

```{r, fig.height=10}
pal <- c(excel_new_pal("Atlas")(6), "other" = pal_538[["Dark Gray"]])

streaming_history %.%
  {
    mutate(
      artist_name = fct_other(artist_name, keep = top_artists, other_level = "other"),
      date = lubridate::floor_date(end_time, "week")
    )
    count(artist_name, date, name = "n")
  } %>%
  ggplot(aes(x = date, y = n, color = artist_name)) +
  facet_wrap(artist_name ~ ., scales = "free_y", ncol = 1) +
  geom_col(
    aes(color = artist_name, fill = artist_name),
    position = "dodge",
    alpha = 0.5
  ) +
  geom_smooth(
    method = "loess",
    formula = "y~x",
    alpha = 0.7,
    se = FALSE,
    span = 0.5
  ) +
  scale_color_manual(values = pal) +
  scale_fill_manual(values = pal) +
  theme(
    strip.text = element_text(face = "bold"),
    legend.position = "none"
  ) +
  labs(
    title = "Trends in listening to my favorite artists"
  )
```
Most artists followed the general trend of more listening early in the year, trailing off in the Fall.
This was no true, however, for [Quinn XCII](https://open.spotify.com/artist/3ApUX1o6oSz321MMECyIYd?si=xNmBqLMsQ0KtNDonz6Jy3w) ([pronounced "Quinn 92"](https://en.wikipedia.org/wiki/Quinn_XCII#2020-present:_A_Letter_to_My_Younger_Self)) because he dropped a new album, [*A Letter To My Younger Self*](https://open.spotify.com/album/1dVw3jSdgZp7PfGhCEo32t?si=TOsNYpa2Txm4BwO6pZD7OQ), on July 10, 2020.

You can see this switch in the following plot of which Quinn XCII songs I listened to most.
In July, there was a spike in the total number of Quinn XCII songs played, and the top individual song switched to [*Notice Me*](https://open.spotify.com/track/6mVzv4E4oc7Fcw6yUTfzub?si=FVP-1x8QQSCqrc1A3UYqQw) (yellow).

```{r}
pal <- c(excel_new_pal("Office Theme")(6), "other" = pal_538[["Medium Gray"]])

streaming_history %.%
  {
    filter(artist_name == "Quinn XCII")
    mutate(
      track_name = fct_lump_n(
        track_name,
        n = length(pal) - 1,
        other_level = "other"
      ),
      date = lubridate::floor_date(end_time, "month"),
    )
    count(date, track_name)
  } %>%
  ggplot(
    aes(x = date, y = n, color = track_name, fill = track_name)
  ) +
  geom_col(position = "stack") +
  scale_color_manual(
    values = pal
  ) +
  scale_fill_manual(
    values = pal
  ) +
  scale_x_datetime(
    date_breaks = "2 months",
    date_labels = "%b %Y",
    expand = expansion(add = c(0, 0))
  ) +
  theme(
    legend.position = "top",
    legend.key.size = unit(0.5, "cm"),
  ) +
  labs(
    y = "times listened",
    color = NULL,
    fill = NULL,
    title = "Trend of favorite songs by Quinn XCII"
  )
```

---

<iframe src="https://open.spotify.com/embed/track/1pyGo7knT3LE2I1FJIrwdX" width="100" height="100" frameborder="0" allowtransparency="true" allow="encrypted-media"></iframe>

<iframe src="https://open.spotify.com/embed/track/3LxG9HkMMFP0MZuiw3O2rF" width="100" height="100" frameborder="0" allowtransparency="true" allow="encrypted-media"></iframe>

<iframe src="https://open.spotify.com/embed/track/65z28RA7Nmzsog6Noln0sB" width="100" height="100" frameborder="0" allowtransparency="true" allow="encrypted-media"></iframe>