Extending People Analytics to K-Pop
How data analytics and people analytics could potentially change the landscape of K-pop music industry.
What are the key determining factors that guarantee an artist’s success?
With the exploding popularity of BTS in the recent years, especially their growth in the Western music scene, the blog post (linked below) explored what sets BTS apart from the hundreds of K-pop groups, and from other Western artists.
In his blog post, the author concluded from his analysis on the music, rather than on the artist, that what makes BTS’s songs unique is in fact their ‘energy’, higher ratio of vocal parts vs lower ratio of instrumental parts, and the diverse tempo offerings from song to song, which resonate better with the mainstream audiences.
Though the author focuses on using data analysis to identify the ‘secret sauce’ that makes BTS’s songs unique, I believe that there are further applications of data and people analytics that would impact upstream portion of the K-pop value chain.
Mapping Backward to Music Production
In comparison to other traits and characteristics of the artists, their music and its success are the most quantifiable outputs. In the K-pop scene, it is less common for artists to self-produce/compose their releases, but rather their companies will source the music from different producers/studios.
Using the analysis in the referenced post, entertainment companies could not only identify the musical traits of the chart toppers, but also see the trend of listeners’ music consumption behavior over time. With this data, the companies could backtrack to identify the producers who have previously produced similar style of music, and source more songs from them, or even recruit them. The company could also guide their in-house producers to produce music that reflects the current trend.
The downfall of this application is that, if the new releases converge on the same traits that were once unique, will any of the songs stand out and make the same impact as BTS’s songs?
The referenced post ended by hinting further explorations on the lyrics and how they contribute to BTS’ uniqueness. Though the author didn’t mention his approach to this further exploration, I think he would use a form of NLP algorithm to identify the factors, such as: word counts, number of english words, overall sentiment of the song; that differentiate hit songs from other, less successful ones.
Beyond Music
Coming back to the main question of the article: what makes BTS successful in the Western market, in comparison to other K-pop groups? Undoubtably, artist’s success depends on more than just their music. Their live performance, physical appearance, music videos, social media presence, and many other components all contribute to their success in the music scene.
In the K-pop system, where artists started off as trainees and went through years of training before making their debut, entertainment companies incur huge upfront investments to groom the trainees. Only a handful however made their debut, and even less became profitable for the company. How could an entertainment company better identify which individuals to bring onboard to the training program to minimize costs and maximize future returns?
I believe this is where people analytics has a great potential. By assessing the incoming trainees’ tendency to persevere through the rigorous training program, their personality, as well as their talent, and mapping it to the ideal score based on successful artists in the industry, entertainment company could have a better screening method to recruit and invest in selected trainees that would potentially yield better returns for them down the road.
However, a problem for this approach is the fact that people are dynamic: perseverance could change, talent and skills could be improved with training. Thus, entertainment companies need to answer these preliminary questions before moving forward: Which independent variables to use? How to quantify or which proxy to use for each variable? Also, the company need to address the potential bias baked into the algorithm based on the algorithm makers’ idea of what an “ideal” artist is.
Referenced Post: https://towardsdatascience.com/the-data-science-of-k-pop-understanding-bts-through-data-and-a-i-part-1-50783b198ac2
I’m really intrigued by people analytics applied to music. Many professionals in the music industry would claim they have a special skill of spotting the “next big thing”. But to what extent is this uncovering hidden patterns in existing data (easy for algorithms to do), versus deciding that something completely new could be popular (difficult for algorithms to do). I hope we find out! My cousin is writing his high school capstone analyzing songs that go viral on TikTok, and I think there is much more research to come.
Great point! Also, given that a ‘hit’ or the ‘next big thing’ no longer limits to just the musical output, but also music videos, and tons of other factors, how much could the analysis be of help in identifying them vs does human gut instinct still have certain unexplainable advantage over this? Also, the behavior of the mainstream consumers changes at the rate that it wasn’t before, which makes predicting or even following the trend even harder.
This is very interesting – and a great example how the field of people analytics goes way beyond impacting the professional workplace setting!
A related concept that I find exciting is how the music industry is leveraging data to not only identify potential hits but to actively shape audience preferences. Starting from the early days of the radio, record labels have experimented with many factors beyond with song/artist characteristics. They would test the timing, order, and frequency of new songs to understand when and how often new songs needed to be heard in order to be recognizable and eventually turn into hits. There is also a network element to it, as the most cost-effective way to spread new music was through word of mouth.
In today’s era of subscription services, labels have even more tools at their disposal. They can access extremely precise metrics around listening habits and track in real-time how songs become viral. Furthermore, as subscription services themselves are concerned with growing and retaining their audiences, they employ many algorithms to ensure that the next song in our playlist is one that would keep us listening. As these tools are only going to improve, it is scary to think about whether we are being served songs that we actually like; or whether we actually are “tricked” to like songs that are served to us when we’re most likely to enjoy them. And as long as we enjoy them, does that even matter?
That’s super fascinating! Isn’t it bizarre how many sources of data the music industry could utilize to ‘shape’ the scene and our likings? Also, the point regarding the network effect is super valid. These days, music distribution depends on more than just a good song, but so many other components, which their impacts are definitely amplified by the network effect.
Since I enjoy opera singing very much, I was very interested to learn that people analytics can be applied to k pop as well! The method seems to be NLP, in different lyrics. I wonder whether it could work for songs in different languages, e.g. Italian, French, German, the ones predominantly used in opera singing as well? Indeed it is very interesting that the tone, and “energy” of the words is captured as well, and I wonder whether in the future people analytics can be used to capture not only the words, but the tone, and mood of the song, as sometimes the words in the song signify one emotion, while the use of major and minor keys, rhythm and body language of the performer suggests otherwise.
It never crossed my mind that people analytics could be used in k-pop or the music industry in general. I’m so glad that you shared this article!
The output you mentioned (success of music released) suggests that the purpose of creating music is to sell and win awards. That probably is the true goal of record labels like Big Hit Music, but I wonder if that’s the goal of the people who create art. As a non-musician, my best guess is that singers would want their songs to provoke thought and emotion among their audience. If they wanted to test that, they could test a songs on a group of participants and use emotion recognition technology to assess how much emotion the song inspired among the participants.
Really valid point on the purpose of music creation. I think for this particular article, and thus my extrapolation of it, we’re looking at the commercialization and business success of music, rather than the aesthetics of the song. However, I totally agree with you that for the artists/composers/musicians, their goals are definitely different from the companies’. I wonder how NPL or other tools could analyze emotions of the singers, and what proxies could we use to represents those aspects of what constitute as a ‘happy’ sound vs ‘sad’ sounds. (ratio of majors to minors chords?)
Carla, I agree that the output emphasizes revenue generation over art form. You know it’s significant when Forbes writes about you in those revenue and global growth terms: https://www.forbes.com/sites/tamarherman/2018/12/26/k-pop-claims-space-in-global-markets-while-looking-at-future-in-2018/?sh=499e18bc3998.
While most popular music is a revenue making business from the get-go, K-pop specifically may owe this label to its origin story. From what I know, the government of South Korea heavily invested in the promotion of Korean entertainment industry after the Asian financial crisis of 1997. Heavily in debt to IMF and other donors, South Korea looked for ways to generate more national income, and the then president saw great potential in exporting Korean culture. This article I found explains it in more detail: https://www.thedailyvox.co.za/how-a-financial-crisis-created-k-pop/. I had only vaguely heard of this before but the article filled in the gaps in my knowledge a bit.
Amazing commentary Korn, thanks for sharing! I saw a similar analysis at the ABA’s project to predict the “next big thing” but this article is very specific and interesting to see all these visualizations.
Totally agreed on your points mentioned about the article. Especially, I cannot agree more for the points about i) why don’t we apply the popularity of the songs as an outcome variable rather than the uniqueness of BTS (at the end of the day, other K-pop artists should be equally “unique” in some other variables), and ii) NLP analysis for the lyrics could be very insightful.
Also it is interesting that there are many available Korean-specific lexicon for sentiment / NLP analysis! Data science in Korea seems very advanced:) (for example: Korean-specific BERT model (KR-BERT) http://ling.snu.ac.kr/)