Map-based music visualizations considered harmful

February 5th, 2011 § 0 comments § permalink

In order to make sense of huge collections of music the first idea that comes to people’s minds is often: Display all of these songs by using some minimalistic graphical abstraction (so they can fit on the screen). This approach is not only commonly taken in research but is by now also available in commercial products and apps where maps are commonly used as a way to one-up the boring old lists.

Last.fm Artist Map tuneglue sonarflow

Oooh, so pretty!

Just to be clear: I’m not against music maps per se – they’re great for discovery. The recent Aweditorium for iPad, for example, lets you find new indie music in a playful way and is really, really neat. The problems start, however, once you decide to use a map for anything beyond discovery: Organizing music collections, building playlists and so on.

My first problem with this approach is how music is represented in one’s mind: Maps are often used for visualizing personal music collections that, as the name implies, also contain all kinds of invisible personal stories stuck to the songs. While James Yuill and The Airborne Toxic Event might not be very similar music-wise, they have a certain connection in my mind as I kept thrashing their albums in the same summer. However, the layout of such a map is usually based on similarity, either content- or metadata-wise, and these algorithms can’t reflect that. Using a personal listening history as basis for the map could, for example, work much better (full disclosure: LastHistory was developed at our research group. Here’s the research paper (PDF)).

The second problem is that of visual representation of songs. Finding a specific artist or even title on a map is often close to impossible as there are mostly no clues for navigation (musical genre and sometimes even similarity are vague and often based on personal taste). So the only option for finding something is visually scanning the whole map. And depending on the representation of songs that can take a while. This scanning becomes more and more difficult the more abstract the representation and the more interaction is required: While it’s possible to find a song when all titles are shown with their album cover art or a photo of the artist, locating a song in a field of colored circles requires much more effort and finding it on a musical “terrain” is practically impossible. Aside from the visual representation, interactivity can also hinder this effort: requiring people to tap on circles to learn about their true nature (a.k.a. title and artist) will ensure that you only keep the most obsessive-compulsive part of your audience.

Mufin Player Treemap Personal Collection Treemap Personal Collection
Quick! Find Madonna!

Such a search is finally made even more frustrating when the app has no capabilities for searching or filtering. While text-based direct search might not be the perfect way to look for music (Query by humming comes to mind) it’s better than visually scanning the whole map. Having a way to filter this information by user-defined criteria (genre, lyrics, tags, release date and the rest of all that meta-data) is even better to let people find what they are looking for.

The most important rule (which is true for all map-based representations): If you absolutely have to use this abstraction MAKE SURE THAT THE MAP DOES NOT CHANGE! Humans are very good at remembering spatial relationships (you could probably tell me without looking how all these things in your apartment are arranged. See also Roman Room memorization technique) and user interfaces that tap into this ability should also support it. Once the map changes all of these intricately learned relationships are useless. Point in case: Aweditorium. I remembered hearing interesting Danish experimental pop there (it was Oh No Ono by the way) which was somewhere in the upper right corner of the map. But of course, once I restarted the app the whole map got reshuffled and I couldn’t find it. Why have a map if the spatial relationships are meaningless anyway? I have similar gripes with the actually pretty great Kindle app for the iPad, that re-calculates the layout of the text – no, not only every time the font size is changed – but every time the app is restarted! Again, remembering that some interesting tidbit of text is on the lower part of a page isn’t any use.

To sum up, here’s the list of map rules in short:

  • Use easily memorizable/recognizable representations for songs. Cover art or artist photos work well. Try to think of visual landmarks to support visual search.
  • Provide a text-based search for specific titles/artists and filters for genres.
  • Never ever change the layout of the map! If you have to change it to add or remove songs try to keep it as close to the original as possible and inform the user about it using animations.

Where Am I?

You are currently browsing entries tagged with maps at Augenmusik.

  • noise

    Error: Twitter did not respond. Please wait a few minutes and refresh this page.