Kpop is gaining a huge amount of popularity internationally. Captained by BTS’ success in the West, Korean pop music has become one of the most recognized music genres/scenes in the world. But for all the renown Kpop has garnered, you might be surprised to learn that Kpop isn’t necessarily the most popular genre in Korea.
What do I mean? There are lots of famous Kpop groups, and even more unknown ones. But when you look at the domestic music charts, you’ll find that the groups most famous abroad rarely crack the top of the streaming charts. There are other artists that top these lists that many Kpop fans would struggle to recall. Sure, Kpop has become the face of Korean music. But the reality is that Korean pop music isn’t always the most popular genre amongst the general public. As large as fanbases like ARMY and BLINK are, they can’t always outcompete the general populace.
That isn’t to say it’s impossible for Kpop music to top these charts. Groups like BTS and Red Velvet have garnered enough success and produced enough quality songs that even the public love to listen to them. But when less popular groups are releasing new music, fans often emphasize the importance of having a public-friendly song. A song that the general public will find enjoyable. If you make a mark with the public and chart well, your song will inevitably begin showing up in another places: played in stores, background music for TV shows, etc. Charting performance is also a part of the scoring for music show wins. Having a hit that charts really well is a great indicator of a group’s popularity.
So what makes a popular, public-friendly song? Many people may point to the abundance of ballad songs that often top these domestic charts as an indication for what the public wants. To answer this question more analytically, we’ll turn to Spotify, one of the largest music platforms in the world. Spotify provides audio data on every song in their database: using this data, we will describe what makes a Korean song public-friendly. You can read more about these attributes here.
First, we compile a list of the top public friendly songs. To do this, I looked at Instiz chart, a music chart that tracks and ranks the most successful songs amongst various Korean music providers in realtime. They also give rankings for daily, weekly, and monthly popularity. There are other notable charts like the Gaon Digital Chart, but I used Instiz because of its connection to the achievement known as a Perfect All Kill, that rewards songs that perform exceptionally well on all notable music charts and platforms. I took all of the best songs (the number one ranked song) for each week of the Instiz chart and determined their audio features according to Spotify. The list of songs that topped the weekly charts is below:
From the list of audio features, we can try to remove some to simplify the analysis. Instrumentalness can be removed from this list since most of the values are all zero (most of the songs aren’t instrumentals). We also choose to ignore the mode, key, duration, and time signature. Though these are properties of the songs, I am focusing on the qualities of songs that an everyday listener could discern. I also omitted liveness since I found that it was more representative of which groups had more of their concert performances as tracks on Spotify (Blackpink had a huge share of their tracks on Spotify come from concert performances of their other tracks).
We will look at these features:
- Acousticness: how acoustic a song is (how little electrical amplification there is)
- Danceability: how suitable a track is for dancing
- Valence: measure of musical “positiveness” (e.g. sad songs are low valence and happy songs are high valence)
- Energy: measure of intensity (metal music is high intensity)
- Speechiness: measure of spoken word
- Tempo (scaled): beats per minute
Next, I made a program to scrape all of the albums for a few of the top Kpop artists (including boy and girl groups) and collect their audio features. Note that some singles/mini-albums may have been excluded. I chose these groups admittedly arbitrarily, but I believe I’ve taken a representative sample of the most popular groups domestically and internationally. I got the six audio features for each of the full album tracks for these artists, and averaged them. I represented the averages on a radar plot to compare them against one another. The result is below:
So the radar plot sums it up very nicely. To read and understand the plot: don’t worry too much about how each vertex relates to the other. Only compare the vertex locations of different colors. The pink shaded area corresponds to the audio features for the Instiz chart songs (highlighted so it sticks out). The Instiz chart songs have significantly higher average acousticness compared to the Kpop groups examined. To justify this result, consider groups like AKMU and BOL4, who often chart really well despite having a smaller fanbase (at least internationally). Their songs frequently feature relatively stripped down music and their live performances are often as simple as one member singing and another playing guitar (and also singing).
The other pop groups have higher averages in the expected areas: energy, danceability, and valence. What was surprising was how similar all of the radar plots for the pop groups were to one another. It does seem to support the common criticism that Kpop songs are very similar to one another. While there is a healthy variety of songs in my opinion, the results do seem to suggest that on average the songs of various Kpop artists are very similar to one another.
Some other interesting features: though the tempo vertex may seem larger than the other features, it is because I scaled the value. I was surprised to find that Instiz chart songs on average had a similar tempo as the other pop groups. Also, BTS has a higher average speechiness compared to the other groups, meaning they have a lot of spoken words in their songs. This can be hypothesized to be due to BTS’s strong rap-line and how involved the members are in composing their own songs.
So this plot is fairly definitive in discerning the factor a Kpop song needs to chart well. Another way to further study this hypothesis is to look a the specific song that has charted for well for some of these pop groups. Amongst the groups I examined, Red Velvet and BTS have both topped the Instiz chart for multiple weeks in 2019.
Unfortunately, looking at these comparisons did not emphatically support my hypothesis. In fact, they seemed to have opposite trends. While Psycho had less energy and valence compared to the average Red Velvet song, Boy with Luv had the opposite change in audio features. Neither song had a significant increase in acousticness compared to the rest of their artists’ discographies. But at least with BTS, it was almost guaranteed they would chart well given their incredible success recently, so Boy with Luv could’ve been any kind of song and potentially done as well. And the decrease in valence for Psycho compared to the rest of Red Velvet’s songs could be a big factor in its general popularity, though I suspect that Wendy’s injury may have played a role in helping the song trend.
So let’s move on from looking at BTS, and instead look at Red Velvet more closely. Red Velvet’s Psycho distinguishes itself further when we compare it to the other releases they had this year (this is where the BTS dataset falters since they only had one major release this year).
Here the data is more representative of the conclusion we had presented previously: Psycho has more acousticness and less energy than the other 2019 Red Velvet releases, reinforcing the idea that more acousticness and less energy are important for a song to top the Instiz chart. Umpah Umpah has a much higher valence peak compared to the other songs and Zimzalabim has more speechiness.
Next, I tried looking at the Spotify features for some of the repeat artists at the top of the weekly Instiz charts.
AKMU and BOL4 have significantly higher levels of acousticness in their discographies and a distinct decrease in energy relative to the other Kpop groups. While their presence on the Instiz chart does influence how similar their radar plots will be to the Instiz chart plot, their entire discography seems to more closely mimic the shape of the Instiz radar plot. In particular, BOL4, despite topping the chart for only four of the 52 weeks (with two songs!), has a radar plot that is basically identical to the Instiz chart result. Regardless of how serious the relationship between acousticness and public friendliness is, it’s clear that BOL4 matches the general public’s taste perfectly.
Next I look at the relative histograms for a few select groups (Red Velvet, Twice, and BOL4), to see if there is a trend in distribution of audio features that points towards success on Instiz charts. For example, if acousticness is an important quality for Instiz chart, does BOL4 have a greater distribution of their tracks on the higher end of acousticness than Twice or Red Velvet? The average already has some similar level of analysis, but a histogram could provide some new information.
First, let’s disregard some of the less consequential plots: tempo isn’t too insightful since Red Velvet and Twice have similar tempo to the Instiz songs (though BOL4 notably has more slower songs). Valence also features heavy overlap between all of the examined groups. Speechiness has a slightly interesting feature where both Instiz songs and BOL4 songs have a heavier density of songs lacking speechiness compared to Twice and Red Velvet. But all of the groups have a general trend towards low speechiness songs.
The other three traits do verify the aforementioned conclusions. In the acousitcness histogram, there is a clear divide between pop artists (Red Velvet/Twice) and public friendly songs (BOL4/Instiz). The ordering of how acoustic the songs are follows the trend I was hoping to examine when choosing these groups: Twice, who had no entries that placed first on the Instiz weekly charts, had the highest density of low acoustic songs. Red Velvet, who had one entry, had a similar distribution to Twice, but a relatively lower density of low acoustic songs and slightly higher density of high acoustic songs. BOL4, with two entries, has a much larger shift in their discography towards high acoustic songs. And the Instiz chart dataset had the distribution with the highest density of acoustic songs. So one can hypothesize that a discography with a higher density of acoustic songs will be more likely to top the weekly Instiz chart.
Similarly, the trend is reversed for danceability and energy. Especially with the energy histogram, you can see that the order we established in the previous paragraph holds: Twice has a large density of songs that are high in danceability and energy. Red Velvet has the next largest, then BOL4, and lastly the Instiz chart songs. The radar plots made it appear that acousticness and energy were the main differences between Kpop and public friendly music, but the histograms make it clear that danceability is also very different between these groups of songs. Still, they pull from the same datasets so the conclusions are expected to be similar. It’s just another interesting way to examine the data.
So at the end, this analysis probably didn’t surprise too many people. Acousticness and less intense (lower danceability and energy) songs are the most suited for the Korean public’s tastes. Of course, there is some ambiguity from how much one trusts Spotify’s algorithms in determining these qualities. But it was still an interesting study to get some form of data to backup some of the more “common sense” notions that fans may have. There are many great sites that provide codes/tutorials to help you scrape your own data if you’re interested. Also, check out this BTS focused study for an even deeper analysis of Spotify audio features and how one can use machine learning to identify BTS songs.
And a note to any companies looking to make a public friendly track: just copy BOL4.
Simplified infographic here.
Originally published at http://paukshop.wordpress.com on February 7, 2020.
