The discovery and recommendation problem for Indie Games: Recommending games that you’re likely to finish

I have a confession: I trust Netflix more than my friends. When someone recommends a movie or a show, I check Netflix first to see my personal rating on Netflix. Basically, Netflix has solved the recommendation problem for television.

The Discovery and Recommendation Problem for Indie Games

The discovery problem for indie games boils down to this simple statement: Find games that a player is likely to finish.

(Note: I’m talking about single-player games that have a definite end. The discovery problem for multi-player games is VERY different from single-player games.)

Games have a bigger time commitment than television. Games can also have bugs, difficulty spikes or implement bad design decisions that force players to give up without seeing a game to its end. At the same time, players are more likely to recommend a game that they’ve finished.

A discovery system for indie games has to take all this into account and show recommendations that a player is likely to finish.

Obstacles to creating a Discovery and Recommendation system for Indie Games

  1. Classification is hard: Classifying indie games is hard. Indie games are still constantly innovating making it hard to classify them in any coherent fashion. Even if you can perfectly classify games, something like the graphical style might discourage you from playing a game all the way through to the end.
  2. Measuring time commitment: Unlike television, it’s impossible to determine the time commitment needed to finish a game. This depends on many factors including your skill level, whether you are willing to use walkthroughs, if you are a completionist etc.
  3. No database of indie games: The information on indie games databases like indiedb.com is not publicly accessible. On the other hand, in order for an indie game to get on Steam (which does have a public API), it needs to be greenlit. The end result is that there is no centralized database with information on all released indie games.
  4. Psychology might be more important: When Netflix sponsored a contest to improve their recommendation engine, there were two schools of thought: statistics and psychology. In the end statistics won over the psychology in improving the recommendations that Netflix provider. However we play and recommend games based on feeling a lot more than logic.
  5. Measuring engagement: Measuring engagement for Netflix is relatively simple. If you watch half of all episodes of a TV show, you’ve shown an engagement of 50%. Netflix can determine down to the second when you stop watching a show. How do you get this information for games? Do you have players rate their experience? Do you have developers implement a tracking API and if so, what’s their payoff? 
  6. Non-standard experience: Bugs and glitches, difficulty spikes and other factors all make the game experience different for different people.

None of these obstacles are insurmountable, but they seem to be a lot harder than even when Netflix accomplished.

Related Articles

  1. How Netflix Reverse Engineered Hollywood

Advertisements