See Similar Recommendation Engines:

The new music sites say, “tell us what you like, we’ll tell you some music that sounds the same.” Why would I want that? Here’s a look at recommendation engines, and how to improve upon them.

Plenty of taste-based businesses want to sell you more or keep you happy with what you’re getting.  So far that’s mostly meant the media industry, but establishments like grocery stores and restaurants will probably join in before long.  They set up recommendation engines.

I am curious why, though, because every one I’ve tried worked thus:

  1. You either input something you like, or they extrapolate it based on the data they gather on you
  2. Their application looks for similar things
  3. It tells you those similar things, possibly with a sample

These bug me, because if I’m resorting to trying a recommendation engine, it almost always means I want something different.  I don’t want something similar, because then I would go to the one I already had or knew.

But it’s the easier problem to solve, to build a system that can determine similarities, than it is to actually elicit and provide the work or the taste to fit the person (and the person’s mood).

The latter would actually go something more like:

  1. Provide a random sample, preferably one that the user hadn’t encountered before
  2. The user provides feedback (too heavy, not enough color, overcooked)
  3. The system then provides new samples based on the feedback
  4. Repeat until satisfied

Note that step two needn’t be formal feedback, and could be accomplished through split sampling.  In that case, the user experiences two random offerings, and picks the one that’s a better match.

I sincerely hope that better recommendations come forward.  Every time I’ve used such a system I am utterly frustrated.

Examples are worth the time.  If you go on to Google Blogger and click the “Next blog” link at the top, you are typically forwarded to a similar blog.  That means (real example) if you were on a blog about digital scrapbooking written in Bulgarian, the next one will likely be another blog written in Bulgarian, and it will probably have a craft theme.  There’s no escape hatch that says, “oops, not Bulgarian, don’t speak it, sorry,” or “well digital scrapbooking isn’t my bag.”

The thing is, Bulgarian digital scrapbookers do really neat work.  But after I’ve looked at one, I didn’t click next because I got the wrong one.  I clicked next to see something completely different.  As there isn’t an alternative link for that, I feel stranded, so I return to the homepage, and go next from there.  That leads me down a similar path, but this time with middle class Christian blogs from Canada.  Again, once I’ve read one, I want to move on, and not to another one of the same type.

The other example comes from music.  The new music sites say, “tell us what you like, we’ll tell you some music that sounds the same.”  Why would I want that?  For gods sakes, I’m not trying to hear the same thing over and over.  Variety is the spice of life, after all.

There are other ways to implement my suggested alternative, though.  I might be equally happy to see, “last 10 shown to people that are not you,” or such.  It’s less overwhelming than the sites that let you browse others’ profiles and see what their entire taste is like, since you only get a single instance per each person (at least initially), and you do get more variety (provided they are chosen randomly enough to prevent spammers from trying to game them).

Do you feel differently?  Do you like the recommendation to give you things similar to what you already like?


Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.