Notes on Feature Based Group Forming

June 30th, 2011

If you’re thinking about launching an application that centers around group forming as a filtering mechanism, a couple of quick feature requests:

  • “Smart Sets” ala iTunes are unlikely to work because human features are less well understood then pop’s features, and will lead to frustration and abandonment, so don’t do that.

However …

  • most entities we refer to as human have a single geographic location at any given point in time. This is a useful, and well understood feature. Cityli.st was a project I start to exploit this fact that automatically maintained and updated Twitter lists based on a person’s location. (turns out Twitter lists don’t work very well, nor are they particularly useful.) Please dear god, build this into G+ already, especially if you’re going to have checkins.

  • update asymmetry is a fact of life. Some folks update multiple times a day, some folks update very few minutes, some folks drop 1000+ photos in a single upload session once every 6 months. Automatic grouping by rolling average of update frequency would be extremely useful.

  • shibboleths (aka something you have or something you know). A group of everyone who hit this onetime URL. A group of everyone who can take a photo the Empire State Building from where they’re currently standing.

  • time is another interesting human constant. People I’ve recently contacted, people I haven’t contacted in years, and people who I was at the same concert/bar/office with are all useful slices.

Thanks.

Photo by maryclaireroman

Tagged: Uncategorized

2 responses to “Notes on Feature Based Group Forming”

  1. Dan Brickley says:

    ““Smart Sets” ala iTunes are unlikely to work because human features are less well understood then pop’s features, and will lead to frustration and abandonment, so don’t do that.”

    …can you expand on that?

    In Google-speak, maybe I wouldn’t write a ‘circle definition’ as a query against people features (age, interest, whatever…). But still a smart set, er, circle, … could be composed from atomic pieces, where the atoms are other hand-curated or auto-generated sets?

    (move this thread back into G+ if you prefer btw…)

  2. Boris Mann says:

    But wait, it’s SO FUN manually creating geographic circles and organizing people into them, even though you KNOW that the data is there to do it automatically!