Our goal at Localify.org is to help people discover local artists within their communities. However, it is difficult to design an effective “Local Music Recommendation” user experience (UX) for a number of reasons:
- most local artists are likely obscure and likely unfamiliar to most users. As a result, showing the artist names will not provide much information.
- genre, emotion, instrument, and usage tags can be be vague and non-specific, but they can be consumed quickly.
- images, such as promotional photos and album covers, can tell us something about an artists but also can be vague, non-specific, and sometimes misleading.
- listening to music may be the best way for a user to form an opinion about an artist but listening music is relatively slow compared to scanning text or looking at images.
For the past, few year we have been designing and redesigning Localify with these limitations in mind. We put many of these thoughts into a Late-Breaking Demo paper for the 2024 International Conference on Music Information Retrieval (ISMIR ’24) in San Francisco this November.
Here are a couple of the highlight from our recently relaunched Localify.org web application.
Artist Card
In the image below, we see an “artist card”. It was designed to contain fast-to-consume contexual information without overwhelming the user. The card includes
- artist name: needed for identification and future recall
- genre information: a mixture of coarse (e.g., “blues”) and more fine-grained (e.g., “americana”) labels for quick scanning
- “similar-familiar” artists: these artists are known to the user and are included as part of the of the user’s profile. These artists are known landmarks and are intended to explain the recommenation.
- “%-match” score: gives the user a measure of how confident we are at recommending this artist.
- artist image: the images encode information about genres, style, intended audience, etc. It is also useful for future recall.
- audio sample: a quick 30-second sample so that the user can get a listen if interest.
As you can see, the card is packed with useful contextual information. A user can also “bookmark” an artists to save for future exploration. If the users clicks on the card, they are shown a “detailed” artist page which contains more information about upcoming events, (non-personalized) similar artists, and links out to music services and social media sites.
Personalized Playlists
While the artist cards are designed for rapid (but perhaps surface-level) contextualization, we also provide personalized playlists that attempt to contextualizate the local artists using their music.
Our playlist algorithm introduces the user to a local artist by first playing a song from a “similar-familar” artist, and then playing an acoustically-similar song from the local artist. Acoustic similarity is measured as distance between the familiar artist’s song and the top song by the local artist in terms of energy and valence acoustic features.
The purpose of alternating between familiar songs and songs by local artists is to balance the amount of familiarity and novelty while creating a coherent playlist. Over time, a user might start to appreciate the music by the local artists as their songs become more familiar (i.e., Mere Exposure Effect.)
Late-Breaking Demo
If you would like to learn more, check out our 2-page demo paper:
Localify.org: Contextualizing Long-Tail Music for Local Artist Discovery
Paul Gagliano, Griffin Homan, Cassandra Raineault, Ruth Ayambem, Bridget Burns, Douglas Turnbull
International Symposium on Music Information Retrieval (ISMIR ’24)
Late Breaking Demo
San Francisco, November 2024
