I know next-to-nothing about IFDB beneath the hood, so I don’t want to second-guess anything here. But as promising as this new approach sounds, I have a feeling that getting it added to the service will mean some work to ensure it doesn’t take down the machine during peak loads (and that its recommendations are being acted upon by users, which is more difficult to measure).
Caching is the usual remedy, but it probably would have helped the old algorithm too. I assume there’s some constraint here making it non-trivial. (And, of course, any approach to caching must also make some decisions about cache invalidation.)
Oh yeah, that is a more interesting list! I’ve always wanted to try out Hound of Shadows, it’s a wonky Lovecraft game with character creation and skills and stuff.
(Thanks for posting my list too, but I’m in a similar boat to Victor inasmuch as it’s mostly a reminder of stuff I’ve played but never reviewed )
I looked up just over half of the alternate list for Victor. Some really interesting stuff! There were a couple of curiosities (one with no ratings at all), but there were some I hadn’t played yet that I will add to my list.
It might be counting the five-star external review, or it may possibly have reviews by people who violated the terms of service (like the Suzy game on the list by Endmaster).
I think it has its high rating because it’s a good game (I hope!) but also because I haven’t really been advertising it or pushing it hard. If it becomes more well known, people outside the target audience will start bumping into it and it’ll start getting a lot more negative reviews (I could see you personally giving it a 2 or 3!). I know there’s a couple of people who finished it who haven’t rated it who I suspect are trying to spare my feelings, already. So if you tweak the algorithm to avoid it showing up, I’d be fine!
Can you share the code of the algorithm? Is it a SQL query?
That seems to be what it’s doing. All of the games I see on these lists have very high ratings on IFDB’s “starsort” algorithm (which is what we call Evan Miller’s formula.)
What we currently do in the “IFDB recommends” list on the home page is just show a random assortment of games that have high starsort scores, excluding games that you’ve played, rated, or are “not interested” in.
Regarding the performance issue, we could simply make a “show me some IFs to play” button so that it only computes on demand and not every time the home page is loaded? It could be OK compromise, maybe? (Sorry if it has already been discussed.)
Yes, thank you! I was curious what it would say for me because I’ve played and rated a lot of the popular ones that showed up in the first few lists you shared. Looking at what it gave for me, looks like a lot of popular ones that I haven’t played yet but have been meaning to (or haven’t played yet and don’t intend to). I only see one I’ve actually played.
Edit: Just double checked and I have both rated Savoir-Faire and marked it as played; is it a glitch that it still shows up on the list?
I like this idea. Even if it were a completely separate thing from IFDB, it could still be interesting. If you wanted to get fancy, you could have options that would adjust the outcome, like “favor games that are popular” vs. “favor games that not many people have rated,” etc.
@kamineko, it looks like the ratings from deleted users are kept in the database, and those aren’t yet being screened out by what I’m doing.
@alyshkalia, I’m using a backup copy of the database from late July. Did you by any chance rate that rating/play after the backup date? (The last one on your list gets high marks from me.)
I’ve been looking at this, but I’m not sure it’s really viable. The tag data is pretty sparse overall, and the most prominent tags are relatively uninformative about the games themselves (as opposed to their development system, competition context, interface type, etc.).
Genre data is also quite poor – over a third of games in the database (including some surprisingly popular ones) do not have any genre data at all.
This thread reminds me of a phenom observed by my Goodreads friend.
He’d put in a rating for a Harry Potter book. So had everyone else in the world, so then he was bombarded with the friend requests, feeds and recommendations of prolific Harry Potter Readers. I forget how he fixed it - either by unrating the Harry Potter books, or throwing some lever(s) on those books pages.
I don’t know if Lost Pig is the Harry Potter book in this equation. Though the difference is, it’s being recommended because nobody claims to have played it
)