The problem is choice (and therefore the opportunity)

… or more specifically, too much choice. It use to be that choice was scare, news was expensive to produce and distribute. Similarly (largely due to communication costs) was movies, music, books, and everything else you can imagine.

For years, we’ve been designing digital products that serve us selection, the more the better but recently we have finally caught up with the realisation that scarcity is not our problem anymore, it’s filtering and discovery.

We’re seeing this (obvious) trend with YouTubes decision to limit it’s options, search (Google) results being delivered on Cards (that limit the number of results that can be delivered), curated news, music, movie, … lists, and even hard wiring a single option to remove the effort required to make a decision (Amazon’s Dash).

And here lies the opportunity, Recommendation Engines that can balance between exploiting what they know about the user and exploring the unknown (i.e. enabling the discovery of new items and better learn the user).

One aspect of recommendation engines I’m interested in experimenting with is model based recommendations i.e. those that use some domain knowledge increase performance. One specific area is around social situations (part of the broader theme of contextual awareness) i.e. ‘group recommendations’ (cannot use ‘social’ as it’s already used as a way describe recommendation engines that incorporate some element of the users social network); group recommendations would use the collective preference of the inferred group to recommend music, food, movies, …

E.g. you’re going on a date, you open up your favourite restaurant locator app – it takes your history and recommends something near-by you might like … somethings amiss – what if your date is a vegan. Theoretically (ignoring scaling for millions) it shouldn’t be too hard to extend our current recommendation engines to handle this.