The Problem with Algorithm-Based Recommendations

Twenty or thirty years ago, most people watched movies and television selected from a finite basket of options.  The biggest shows were on the major cable networks, and movie theaters or Blockbuster were your only options for finding movies.  If you wanted to travel off the beaten path, you could watch some of the smaller channels if you had a good TV package, try to find a low budget theater, or scour the dark corners of Blockbuster. These options were limited. With the advent of the internet, this has all changed.

Now, Netflix, Hulu, and other streaming services give you access to hundreds of movies that may not have been in theaters or available OnDemand.  You can watch shows that previously would have either never been aired or would have been shown on a smaller channel at 1am. As in the music industry, you can now even access amateur-made shorts and series from creators that got their start publishing content for free on Youtube or other similar sites. Now, whenever you sit down to watch something, you’re selecting from an almost unlimited database of content.

This same technology has also led to the creation of numerous resources available to make sense of all this consumable content. You can find unlimited articles from supposed experts with tastes inevitably different from your own. You can browse Netflix’s suggestions for you. You can look movies up on IMDb or Rotten Tomatoes to see how their user base feels. All of these resources, plus all of the available content, make “what do I want to watch?” harder to answer than ever.

There is obviously an enormous profit incentive for figuring out how to direct consumers to content that they like in this new age of choice, and many different companies are hard at work. Netflix actually employs people to watch their content and rigorously “tag” with descriptive indicators to help them understand what you might be responding to. Their suggestion algorithms then combine this with your viewing data to sort you into one of over a thousand different categories of viewer, so they can try to pick what you like most.

Though I am in general a fan of big data solutions, the main reason these types of solutions don’t work well in art is that your taste is too subjective and variable to predict. We can like a movie or show for so many different reasons that drawing correlations from the available data is impossible at this point. Because Netflix (and experts, and raters at IMDb, etc.) don’t know YOU, they have to start from the content of the movie and work backwards to try to predict which selections you’ll like best. For this reason, a recommendation from a friend or family member, for you specifically, is still the most reliable source of information for finding something that you will like.

That’s our philosophy at Reklist. You get these valuable recommendations all the time. Sometimes you put them in a note on your phone, sometimes it’s in a text thread, sometimes you’re sure you won’t forget. Reklist stores these recommendations for you so that they’re all in one place for when you’re actually sitting down to watch. You can label your Reklist because some of your recommendations are a little more tantalizing than others. You can mark them watched with a rating so you’ll remember better in a year from now when someone else brings it up again. Reklist allows you to cut out all the noise and start with the highest value data: recommendations from those who know you best.