From to vinyl to cassette, from CD to download, streaming technology changed the way we listen to music again. And we’re so happy that the wall blocking us from accessing a great piece of work produced in any corner of the earth eventually disappeared. As a result, we started to use it, and then probably switched from one streaming service to another one since the world of entrants is expanding – Spotify, Pandora, Apple Music, Tidal, GooglePlay, etc. Perhaps, one of the most important criteria for us to choose the frequently used application is whether it can get us exposed to the music in line with our tastes in the most effective manner, and this is exactly what they are fighting for – Music Discovery. This article will offer an overview of where these big streaming players have been with regard to the war and the foreseeable section of future.
The Pros for Human Curation
The issue about how to reach a make-sense discovery – Human Curation vs. Big Algorithm – had been taking up many headline columns in the whole last year, because each biggest player is making its effort respectively to figure out the best way of music discovery to differentiate itself from other competitors. And we did see that various viewpoints of professionals in the music industry directly impact the product development strategy in music recommendation.
According to the industry views I was able to access, there are three interesting ideas contended on the human beings side by now. First, music listeners like music curators to be their teachers. “I’ve met many special people who have a deep understanding of the music they love. Their empathy goes beyond the songs, into the history, the influences and connections between artists and genres, the visual components and the craft. Even though I myself know much about music, they provide illumination and enrichment.” (McDermott)
Secondly, they believe the real progress in music creation is always realized by the few people with the creative senses of hearing. “Mainstream society has shown throughout history that although they think they know what they will like, they really do not. They need people at the bleeding edges of creativity to show the way… In the music business you learn quickly that if you only give the people what they want when they want it, you will always behind and eventually become extinct. The most successful people in the music industry do not ask people what they want now, they give them what they will want next. A machine cannot stomp on concrete floors splashed with beer, sweat and blood to figure out what that is next.” (McDermott)
The third point is that humans are still deemed good at discovering brilliant new music and taking them to other humans as well. “I think the value of the DJ and what they can bring to the listening experience – above and beyond the blending tracks and making a seamless music experience – is to bringing to the forefront new artists and music makers that perhaps you have never heard of.” (Perez)
Apple Music, one of the latest entrants in the streaming-music war is proceeding with strengthening the Human Curation team in the music recommendation. In the meantime, the big algorithm was criticized that “a lot of those algorithms are just associating popular tracks with one another and it does take a long time for new music to enter the system and start being recommended to people.” This is a defensible perspective, but we shouldn’t fail to know that a decade of dedicated efforts have been put on the big algorithm behind “Pandora-style” recommendation, and it appears to be blurring the gap between human and machine. From the perspective of data science, many efforts have been made to improve the efficiency and precision of music delivery.
Music recommendation is aimed to improve the efficiency and precision of music delivery, which is a key step to help users discover music. To maintain the spirituality of sensing music in their special algorithm, Pandora “takes academically trained music experts and ask them to tag songs with dozens of pieces of metadata.” (Titlow) Since these tags are tied to musicological knowledge – “tonal qualities, instruments played, rhythmic nuances and hundreds of other details” – their system based on data science has blended many tastes of human beings.
Also, Pandora has been very careful to work on user and scenario segmentations. For example, Pandora found that not all the music listeners are proactive enough to discover new music – “People who put on a classical station or a station based on a Miles Davis song will hear more new music than people who listen to pop and rock stations.” Therefore Pandora got to know which subset of the users are worth further recommending or testing.
Pandora segments user behaviors down to one-by-one individuals through analyzing the different devices users would use in various scenarios. “The thumb-down you get on a Samsung player is very different than the thumbs-down you get at a desktop computer during working hours.” Apparently, with digging into the complexities in the dataset of users, Pandora team would open up more valuable possibilities in respect of customizing music discovery for users.
Predictive Modeling in the Big Algorithm
For many streaming services providing the functionality – music discovery, “predictive modeling” has been widely used among them. In the book Data Analysis for business: What You Need to Know About Data Mining and Data-Analytic Thinking, Provost and Fawcett explained how to use the theory of Euclidean distance to reason the most similar neighbor – the core of predictive modeling. And this experiment has been deepened down to using the relationship between similarity and distance to reasonably predict neighbors’ responses and behaviors.
Take this simple example. We could analyze the similarities amid the group of users, artists or music respectively and then produce recommendations. A user who loves artist A could be delivered the music of artist B resembling artist A algorithmically. Or, user A is invited to follow user B’s personal playlist because their tastes are very much alike. Or track B is recommended to a user who always loops track A because musicological algorithm already recognizes the similarities between them.
As a user listening to streaming music every day, I could not deny that I have become one of the beneficiaries in music discovery. Of course, in fact, sometimes I have to say that “why they guess that I should like this song,” which indicates an unchanged fact that music is truly a personal experience entertainment product – the uncertainties in listeners’ responses are tightly associated with the complexities of human beings. However, I’ve always been supporting the R&D people in the persistence of working on how to cater to the preferences of users more correctly, because this is why competitions exist – the team capable of tailoring their product best will attract more users and win this victory.
Concerning how to improve the accuracy of assessing these similarities, researchers in academia are also making positive progress on the algorithms used in music recommendation, other than crafting those musicological tags in a catalog. On the basis of the most typical method – Collaborative Filtering (CF), more and more new variables were selected to new algorithms during their new experiments. By far, users’ ratings (Mao) and emotion signal in microblogs (Deng) have proved that we’re producing better recommendation results with the help of academic research.
As I mentioned in the fourth paragraph, however, I still believe the human curation cannot be completely replaced with the big algorithm since the progress of art creation always derives from something people think they do not want at the very beginning. And which side do you support?