How to design an experience for the global community by leveraging machine learning?
Smule connects people all over the world to sing and create music together through our Sing! app.
Our users feel that Sing! is like a home — inclusive, honest and welcome. Because of this cozy feeling of home, our community has grown tremendously as an attractive home for people all over the world.
Here’s an example of how Sing! users share and enjoy each other’s recordings:
Recently, we are experiencing some growing pains. New users say that Sing! doesn’t have that homey feeling anymore. To understand what the new users were seeing and experiencing, I stepped back and revisited Sing! as a first time user.
Putting my feet into a new user’s shoes allowed me to understand their experience and what content he/she sees in Sing! By viewing their journey through this lens, I learned a few key lessons on how to design for a global community.
As a new user from the U.S., I found a lot of content that was not in English. Some content was in Indonesia, Spanish and other languages I didn’t know. Sing! enables users to create recordings with singers from all over the world. With our fast-growing user base, there is a huge diversity of music, social and languages coming together all at once. Although the community is generally friendly and supportive of each other, it’s hard to understand and collaborate when there is a language barrier.
While music has no boundaries, language and communication do present barriers, especially when expressed through song lyrics. When new users see a recording in another language from someone they do not know, they are confused. This feeling of confused dissipates the collaborative feeling that we aim to foster.
When a first-time user sees people singing in a language they do not recognize, it can be an alienating and unwelcoming experience. An obvious solution might be to show content to users only in the language based on their location. However, as our mission is to connect people all over the world, we think that solution would be too limiting. For example, we found that users from the U.S. who can sing and speak in both Indonesian and English would like to sing with users from both countries as well. If we only show content in one language to these users, they will miss out on a whole region of the world, and we would not be accomplishing our mission.
So we took a step back and asked the question from another angle. Instead of asking, “How can we show people content in their language?” we asked, “How can we help people find songs they like to sing?” We then tweaked our song recommender based on a combination of user locale and preference.
When the user first starts the app, we use the user’s “locale,” eg. their location + language (for example, US-Spanish locale vs. US-English locale) to recommend songs based on popular genres and topics in his/her locale. Once users begin to interact with a song, we are able to learn about their language preferences and show them content in languages they can understand. Indonesian users now see content in both English and Indonesian, and U.S. users see content in English.
With this change, we saw global new user engagement improve. And more importantly, our users are able to meet and collaborate across the globe, not limited to their region.
Beauty is in the eye of the beholder. On Sing!, users create their own cover versions of a song. For example, users can create an acoustic version of “Let It Go” or a piano version “Let It Go.”
There are probably over thousand versions of “Let It Go” on Sing! and each of them is unique as they come from singers all over the world. That’s cool, and for those singers to be able to share their version with family and friends, that’s especially cool. However, you probably won’t be interested in hearing different versions of “Let It Go” from people you don’t know.
When we first designed the recommender, we thought that it would be great for a new user to hear a few different versions of the same song so they can see the diverse community, and even be inspired to sing their own version of the song. For each song, we recommended five different recordings for new users to explore.
However, since we have little information about new users and their network, we risk recommending songs users don’t like. For example, a new user may see five or more versions for a song they don’t like. To introduce a new user to Sing! by showing multiple recordings of that song can backfire because the user can lose interest, get frustrated, and drop out before they can explore more Sing! content.
The solution for this is rather simple. We changed the recommender to show new users only one version of a song, and shows more diverse, varied content. This way new users are more motivated to explore, listen to and discover many different songs and content in Sing!’s community.
Sing! has a rich variety of content. Users can create solo, duet, or group recordings in audio or video format. Among these, video or partner artist recordings provide the most engaging experiences.
However, new users are not creating video or partner artist content in their first creation experience. Based on our user data, we learned that new users tend to create their first songs in solo audio format. “Join” is a unique concept in Sing! where users collaborate with each other by creating a duet where they sing part one or part two of a song. Many users said they started participating in the Sing! community after their first collaboration with another Sing! user.
Through the audit, I learned that having new users explore more video duet content will help them learn the concept of “Join” more quickly. I also learned audio recordings are less self-explanatory than video, and that partner-artist recordings encourage new users to create duet recordings because they are more motivated when they can sing the song with an artist.
Because of these findings, we set a product goal in 2017 that showing more video and partner artist recordings in Sing! to engage new users.
Working with data team to design an experience for the global community is a unique design challenge. It is not a typical design problem such as delivering a new UX flow or UI design. It is an “experience design” that integrates product features and algorithms to best deliver content for different types of users. What is the right content, language, format? How do we present these in the way that users can understand and engage with? That becomes as the key to this success.
Our Sing! community is fun, creative, welcoming and encouraging. With its growth, we want our new users to experience the same emotions Sing! users had when we began four years ago, no matter where they are and what languages they speak.