Graduate work in the sphere of game recommendations. (GA inside)

Bump!

Well, I rated the "content-based" algorithm one and a half stars, because half the games it recommended were already in my library. That's a bit of a problem.

Friendly bump for now :)

Tried content based. From 9 suggestions, I already owned 4, 1 would be interesting and 4 are pretty much the last kind of games I'd play (CoD).
User based was much better. But mostly games from series I already own or games from my wishlist.
Item based seems to be all over the place. Just random suggestions. SVD was decent, in comparison.

error 500, might try later when it's back up and I don't forget
edit: actually some really interesting recommendations!

error 500 :/

Well, content based worked out best. The other threw things a my, I had a look at and decided it was nothing for me. So it probably makes sense to show them to me, because they seem to be to my liking... but aren't :]

No idea what the algorithm is but at least in my case I'd say it doesn't work particularly well. Usually only 1 to 5(very rarely 5) of the games recommended would be ones I'd even consider and 90% of the time they're already part of my wishlist or installements from franchises I partially own which kinda beats the point. I do see a correlation between games I own and have hours logged in and the types being recommended so it's not THAT far off either.
Have you considered along with the pool rating to give the option to rate individual recommendations with the algorithm, saving that data for that session and adjusting to that? Also if this is based solely on hours played there's a big pitfall there to avoid, idling. That generates incorrect data so achievements earned might be a better metric used in conjunction with the rest of the data.

I tried all algorithms multiple times and the one that came the closest were mostly Content and Item Based. User and SVD rarely got more than 2 right, in fact they even brought up games I have marked as "Not Interested" quite often(I know you can't pull that data, I'm just emphasizing how mismatched it was :3 )

Anyway best of luck with your project!

Did it with every combination possible, both popular and without, including auto.

Im not sure this method (without any questionaries) is the best.
Surely it works for what it focuses most - an overall formula- but it avoids gathering data that could help you change the algorithms; All it does is telling wich of the versions is 'best' right now - and only by aproximation, i mean, with a bigger data set(more testers) and perhaps even in a different day the result could be different (depending on how the algorithms work). For example as of late i have been playing a lot of retroish games and rogue-likes. Couple weeks back i was into fps. A month ago rpgs. Considering how many weren't really played before, how much of what i won i actually like or not...

Theres also a fuzzy definition of recommendation. We can interpret it as 'kind of games you like/might like' or more specific/better 'games you might like that you're unaware of'.

No matter what I do, I get Server Error (500)

Edited;

I was getting "Server Error (500)" over and over but I just tried it again and it worked. I got 12 recommendations on auto settings. One was an RPG from my wishlist. There was Timberman and Five Nights at Freddy's and a cheap multi-player FPS - none of which look like games I would play. There was also a bunch of RPGmaker games and visual novels that I wouldn't claim a key for if they were free.

Content-based got me a 50/50 mix of stuff that was on my wishlist and stuff that looked like complete poop. User-based was almost complete poop. Item-based was actually complete poop. Whatever SVD is got me a few games that were already on my wishlist, a bunch of HOGs, a couple more RPGmaker games and some games based on strange regional sports from the USA that I'm sure I would find repugnant if I actually understood them.

I'll admit I'm probably not the best test subject for this kind of thing though - given that I'm forced to do a fair amount of gaming in offline mode stats from my profile might be a bit off...

"Server error (500) here"

Bump!

It should really filter out owned games from the recommended results. Without this it is useless.

Also, it should filter out "hidden" games before running algorithm.

error 500 right now but it sounds interesting so I will try again later

Bump

ERR_CONNECTION_REFUSED
localhost rejects)

it says error 500:( Maybe I will try some time after thanks:D!

yeah, thanks for checking and testing. I should fixed something now.

I just tried it and it worked really fast, here is my feedback. However, I chose not to "rate" any algorithm because I didn't want to spoil it with my (perhaps) unexpected requirements & higher standards:

• Auto - 1/12
• Content-based - 1/12
• User-based - 2/12
• Item-based - 0/12
• SVD - 0/12

Why so low? I use Linux and if the game isn't supported on my OS I have no interest in it (in fact, I don't even know most of the games that were recommended). Perhaps the algorithm could be tweaked ... in my case, 98% of my playtime is from games available on Linux - this should be a pretty obvious indicator of a preference (likely, anything 90% and above would be).

The not well known game recommendation I didn't know a lot of it but I knew a few of the games that I had or would be interested. The popular game option for all the choices had a lot of games I knew and was or still interested in. So I thought it worked fine.

Bump

All filters work almost the same for me:

Most games i am interested in, majority on my wishlist
Few games i played similar, so i could find them interesting and i understand why they are there
And the rest i have no interest in and have no idea what are they doing there

Just a suggestion, maybe add filter for wishlisted games(to include them in recommendations or not) :)

beump

500 error
/shrug