Game Recommendations from "The Bat" - Poll Data (from you!) Wanted

General
1

In my ongoing tinkering with the problem of automated game recommendations, I need some data from IFDB users to evaluate which approach is the most promising.

I have three different sets of 10 recommendations for games using 2024 IF Comp winner The Bat by Chandler Groover as the starting point. Each set uses different methodologies and different data sets, and there is a little overlap between them so that there are only 28 unique games across the three sets.

There’s a maximum of 20 choices for a poll, so the list has to be broken up into two. Please check all games that you think are good recommendations across both polls. Remember that the theoretical question being asked for recommendations is: “If I liked Game X, which other games might I like?” In this case, Game X is The Bat.

I think the following are good recommendations… (choose all that apply)
  • A Crimson Spring
  • A Matter of Heist Urgency
  • A Spliff in Time
  • An Account of Your Visit to the Enchanted House & What You Found There
  • Counterfeit Monkey
  • Curses
  • Dr Ludwig and the Devil
  • Eat Me
  • Heroes Rise: HeroFall
  • Invisible Parties
  • Max Blaster and Doris de Lightning Against the Parrot Creatures of Venus
  • Midnight. Swordfight.
  • Murder at the Manor
  • Pick Up the Phone Booth and Aisle
0 voters
(This is the previous question continued; choose all that apply)
  • Repeat the Ending
  • Sting of the Wasp
  • Superluminal Vagrant Twin
  • Taco Fiction
  • The Addicott Manor
  • The Frenetic Five vs. Mr. Redundancy Man
  • The Frenetic Five vs. Sturm und Drang
  • The Frenetic Five vs. the Seven Deadly Dwarves
  • The Origin of Madame Time
  • The Owl Consults
  • The Witch’s Apprentice
  • The Wizard Sniffer
  • Toby’s Nose
  • Zozzled
0 voters

Also, a follow-up question: If there are games that you would expect to see on this list but which aren’t listed, which are they?

1 Like
2

Following up on this: Given the limited data received, the best-performing model by far is the AM model, aka “Ask Mathbrush” – yes, I cheated and one of the “models” is a list that mathbrush gave me. (Thanks, @mathbrush!)

This is not particularly surprising, but since mathbrush is not scalable, I’m still looking for more data to compare-and-contrast the other two models (which are both actual algorithms). Also, if anyone wants to chime in about why they think particular games are good recommendations, that might be helpful in developing new ideas about how to make progress.

I will also note that some of the connections between games observed by mathbrush would not really be possible to derive from the data available in IFDB, since they depend on higher-level categorization based on more abstract features, e.g. tone and style. (There was some theoretical discussion last year about setting up tags for this type of thing, such as “mood tags,” but I would expect problems in practice due to their at least partially subjective nature.)

1 Like