Since there was some recent discussion about how the old one was discontinued because it didn’t work, I was tinkering with a new way to generate a recommended list using IFDB data. The basic idea is that you might like the same things that are liked by people who like the same things you like.
It’s giving pretty different results for different people, and a test query for me was chock full of games that are already on my wishlist, so it might be working OK. The tendency is to suggest high-ranking games that you have not rated, so if you haven’t rated them, the classic hits tend to dominate the top of the list. I think that’s desirable – for someone theoretically just starting out, these would be among the most likely to please them.
The #1 suggestion for @mathbrush was The Elysium Enigma. Anyone else want some custom recommendations?
I forgot that I never reviewed that (or deleted my review). I don’t know why, I wrote a whole article on it. But that looks like the algorithm is working based on the data it has!
I adjusted it to exclude games that are marked as played even if there is no rating, as well as games on the “not interested” list of and games known to be authored by the person for whom recommendations are being determined.
More than half of those are somewhere in my “Very Urgent Play-Soon” list !, which has also grown much too large to be considered as urgent when it comes to practicality.
Some (Slouching Towards Bedlam, To Hell in a Hamper,…) I have played some time ago but are still waiting to be replayed and reviewed.
Your list brought to my attention that I had apparently forgotten to rate Fail-Safe, although I liked it very much.
And a whole lot has now blipped onto my radar as maybe being worth a closer look than I thought.
Hm… I’ve played almost half of those to completion, including the entire top 6, so this mostly tells me that I should be better at posting ratings and reviews. Which is actually useful – if this were a public feature, I could post those ratings and reviews, and then check out the new list.
(I’ve actually got an external piece about Photopia listed on IFDB, but of course your algorithm can’t know that.)
I don’t think all of it is particularly geared towards me as a player; e.g., Lost Pig is good, of course, but not really my type of game. It would be interesting to know why it came up second in my list.
The short version is: “Because it was really liked by people who really liked the same games that you really liked.”
I figured as much about it serving as a ratings reminder function. From what I’ve seen about other top reviewers’ lists, it seems like it would be a widespread phenomenon. (But maybe that’s not a bad thing?)
If the priority for recommendation is given to highest ratings instead of most rated, then the lists become quite different. If anybody wants to see these chancier recommendations, let me know – there are certain newer and/or more obscure games that tend to end up at the top of the list. (FYI, @mathbrush: The game that I’ve seen most frequently appearing at the top with this methodology is Never Gives Up Her Dead.)
Oh yeah, that would also be interesting. A Classics list vs a Gems list, something like that. It could certainly lead to some more surprising discoveries. (So give it, please! )
I personally would be interested in something that recommended NGUHD first. There are just so many ways to find out about some of these more-frequently mentioned works
Because I’m too lazy to dig out the old thread, I might be misremembering, but:
(<nerd stuff>)
I thought the other problem with the old algorithm was also that it took excessive system resources to compute the list, and was bogging down the server. Do you think your algorithm is more performant?
)