Spotify Wrapped Brainstorming
I have been a regular Spotify Premium user for about 5 years. First of all, I would like to say that I am very pleased with the experience and pleasure I have gained so far and I would like to express my gratitude. If I have to talk about myself briefly, I am a university student of Turkish origin who continues his education in computer engineering in 3rd year and economics in 2nd year. I work on artificial intelligence, financial technology, machine learning and data science. I have many research and studies in different fields.
·
Spotify
Wrapped brainstorm
I decided to write an article like this and share it with you, as I wish to do one of them on Spotify Wrapped just as an idea storm. Spotify Wrapped can be summarized as an innovation that provides its users with a report on their various listening histories at the end of each year. Wrapped can offer various results to the user thanks to data such as which type of music was listened to and how many minutes were listened to from the most listened artist during the year. As Spotify Wrapped was positively affected, I am aware that a few companies focused on data science were acquired by Spotify in order to use technological infrastructures such as AI or ML more efficiently. I am also aware that some new features are presented to the user and are being tried to be offered, with a very high probability of months of thought and billions of data used. However, as I continued to examine this Spotify Wrapped add-on, which I especially like a lot, I thought that it could be offered to users, and I came up with some extra ideas. Can the processed data be fully used? Is the data stored in databases fully used? Is it slow to develop the algorithms used? I followed inquiries like. At this stage, I will have a few suggestions that I am working on specifically for Spotify. I think the current artificial intelligence infrastructure is sufficient, but I have a few ideas that I think would be great if they were added to the Wrapped section.
·
part 1
As a feature that can be added to Spotify
Wrapped, it may be possible to develop an algorithm that includes predictions
about how the user spends his/her year. I think that data such as emotional
state or enjoying life can be obtained as output from data such as the type of
music that the user has listened to, the artist and which genre for how long. This
technology can be likened to fortune-telling applications that are tasked with
predicting the future when you upload the coffee left in the cup to the
application as a photo when you run out of coffee from your mobile phone. Of
course, the work to be done for Spotify Wrapped will contain a more complex
infrastructure that cannot be exactly similar, but it is not bad as an example.
In the final stage of the project, there will be many outputs prepared to be
presented to the audience. Let's list a few of them below as examples.
There is such a thing as the text that is
intended to be created, here is the weighted average definition of it. Let the
weight of this in the text be 0.1. The percentage weight of this information is
spread over each day. Since he goes to have fun on weekends, he should not be
here and participate in this formula. Different interpretations can also be
brought to reflect the various difficulties experienced during the working days
or during the day.
· Mike preferred to listen to very energetic music and artists for a month, especially on weekends à Mike is getting a more enjoyable result on weekends.
· Sarah mostly listened to chill music every evening for a month compared to her other listening habits à The output of this may be trying to get rid of the tiredness of the day.
· Tom listened to metal rock music on his way to work on weekday mornings à Let's say this sounds like his character is more emotional and uses music as energy fuel.
· Carmine listened to more music towards the end of the month à This can be interpreted as a constant decrease in his mood towards the end of the month in relation to spending his salary and financial adequacy. We can also summarize as using Spotify as a cheaper entertainment source.
·
Carla
listened to more enjoyable music during the summer months à If we interpret that he likes summer more, we can add
this to the data we will present in Wrapped at the end of the year. Outputs can
be interpreted as Hey Carla, you look like you loved this month!
Of course, these are all fairly simple
examples to form an opinion about the outputs.
From the beginning of the year until the end
of the year, Spotify will continue to benefit from different technologies while
improving its ability to receive and interpret user data day by day. Examples
of these technologies are innovations such as NLP. At the end of the process, a
personalized end-of-year message and different badges can be presented to
people who spent the year in different classes. Sharing it on social media can
increase the interaction between Spotify users. Music can be classified as
cheerful or sad, depending on how the year has passed for the users in
question.
If it is possible to integrate the existing
innovation ideas and bring them to a conclusion one by one, I think that this
can be realized after an estimated one year of work here. To put it
superficially for now, it can be thought that only a few new ones will be added
to the outputs obtained from the artificial intelligence fields used. The good thing
here is that not much needs to be added or removed from the existing database.
Only the received data has to be inserted into a different algorithm, resulting
in a deeper stack of results.
This add-on, which is deemed appropriate to
be added to Spotify Wrapped, can be considered to have a very high chance of
catching the truth with its feeding feature from many different places. For
example, we can also consider it to be used on Spotify or a user's organic
searches. A few helpful sentences can be shared here for the path to be
created. Weekly discovery can be given as a great example of this work,
artificial intelligence, which can present a different list at the beginning of
each week, may also give a motto or a text in the future. This can also make
the user consider posting on social media.
The data such as song preferences, playlists, geographical location of listeners used to feed the existing technology infrastructure can be used in this topic as well as in all suggestions made at the moment. Users can be classified over time, just as on the other side. Since the algorithms that will provide the texts prepared for Spotify Wrapped will work better in the future, a study that can be done with a simple and short text with a trial version towards the end of 2022 may arouse quite curiosity on the part of the user. As in any sector, it is necessary to be open to such gradual innovations in order to improve the user experience and get ahead of the competitors with such innovative ideas.
·
part
2
In the next stage, a text guide and pool is
created by combining the algorithm that gives feedback by seeing the behavioral
trends as a whole, just like the total number of plays for the songs and
recording to the list, with NLP. Data such as acoustic different noise levels,
accents, age groups can be used for this text.
Phrases used to describe artists classify
them by song genre, while an output can also be obtained in mood estimation.
Examples of this are data such as which songs are discussed or described on
social media and various platforms, and descriptive terms about artists. For
example, in which language does the user listen to music more? With these,
ready-made text infrastructures can be created.
It can give meaningful results by assigning the weights of the obtained vectors and terms. Considering that the system does not have a specific dictionary and is constantly renewing itself, it can be understood that the process will not interfere at all even if the artist changes the genre and shares a song. RestNet, which can keep working up to 150 layers without damaging the algorithm, can help here.
Here, sound models try to evaluate raw sound
notes directly. It will analyze the song and allow users who listen to similar
songs to be included in the recommendation group. Of course, we will try to
search for the truth with bidirectional control. These sound models will also
help in identifying and recognizing new generation artists by the model.
In order to avoid spam here, it may be
necessary to use a very reliable source or a lot of sources to pull the data.
For this, common filtering, natural language processing and sound models can be
studied in detail. Azure infrastructure can be used for these.
Acquisition of blockchain company Mediachain
Labs. It also ensures that the process can be used more accurately and
transparently for royalty payments. Artificial intelligence can now distinguish
who is singing the song from within the music. Data centers have the power to
process data. It was Niland, a music technology company that provides music
search and discovery engines based on deep learning and machine listening
algorithms to fuel its machine learning model, and its acquisition by Spotify
could make them work more efficiently.
Especially as of 2020, machine learning, powered by both user data and external data, has become the core of Spotify's offerings, helping artists better understand and reach their audiences and be discovered, while helping Spotify stay on top of the music streaming space. The customer base and predictive recommendations that keep users coming back are now very effective and will continue to be such platforms.
·
part
3
However, I think that Spotify should make it
possible for those who have been users for a while to download the songs as an
excel list and keep it as a souvenir, maybe leave it to their children in the
future. After all, the only thing that separates humans from other living
things is the instinct of thinking, which pushes the person to want to be
permanent. It is not difficult to guess that this is not something technically
difficult.
It shows that developments in such deep learning will continue in our age when there are developments such as understanding the spoken, understanding the text and so on. As can be expected in the following process, there is a lot of learning to be done.
·
references
https://outsideinsight.com/insights/how-ai-helps-spotify-win-in-the-music-streaming-world/
https://www.datasciencecentral.com/profiles/blogs/6448529:BlogPost:1041799
https://www.analyticssteps.com/blogs/how-spotify-uses-machine-learning-models
https://ichi.pro/tr/cnn-mimarisi-resnet-nasil-calisir-ve-neden-30895772711476
