Steam Store Overhaul Highlights Still-Rubbish Recommendations

Valve have launched an overhaul of the Steam Store front page, highlighting handy features and slapping recommendations all over the place. In short, it would really like to recommend games to you, but it still not very good at them. I’ve seen the new store for a few days now but other folks in the RPS treehouse say this is new to them so hey, here the Steam Discovery Update 2.0 is, officially launched last night.

This handy page details the new store layout. You’ll find a handy ‘recently viewed’ list, big buttons for commonly-used features which used to have small text links, notices for when your games are patched, updates on what your chums are playing, and oh so many recommendations from all sorts of places.

That’s a big focus of this, getting more games – and more unfamiliar games – before players’ eyes, piecing together sources like your game history, your friends’ favourites, and games on Steam Curator lists. Keep scrolling down the Steam front page and it’ll generate recommendations forever. Do follow our Steam Curator to see our tops smashing wizard wizzo ace aces recommendations in there.

More lists and things on the Steam are now customisable and filterable too, but not quite enough for Discovery to be very useful.

Even as someone whose job includes watching a conveyor belt of new releases and remembering everything I see, like Brucie’s Generation Game without the cuddly toy (wahey!), I’d appreciate a good recommendation engine to highlight things I’ve overlooked or forgotten. That’s still not Steam.

“Due to your recent playtime in other Indie games”, it suggests, I may want retro action-RPG CrossCode or factory-building-factory-builder Factorio. I am grateful for the reminder that I’ve not got Factorio yet but sheesh, ‘Indie’ is a useless label in almost every situation and certainly here.

“Due to your recent playtime in other Singleplayer games”, I might like booby catgirl visual novel Nekopara Vol. 2. It is singleplayer, after all. The ‘multiplayer’ recommendations are just as useless. Not that genre tags are much better.

“Due to your recent playtime in other Strategy games”, by which I assume it means Civilization VI, it recommends Megadimension Neptunia VII, a JRPG.

Because Alien: Isolation is on my wishlist, it thinks I may like Aliens: Colonial Marines. Ah. No.

“Since you recently played Half-Life”, I may want Half-Life: Source.

I’m not scrolling down for yonks to get these, reaching for the bottom of an endless barrel – they’re some of the top suggestions. It’d take a lot of training to improve the recommendations and they’ll still be built on faulty ground. Steam’s user-applied tags are not good.

Recommendation engines are difficult, to be sure. It’s easy to spot broad trends in someone’s activity but picking out the specifics of why they do or don’t like particular things is mighty tricky. My decade-old Amazon account thinks I may want bourbon (correct), a spy camera (what?), and the 54-disc JAG box set (oh). knows almost every song I’ve listened to in five years but fails to recommend tunes I really dig. Google Now tells me daily about rugby, a sport I doubt I’ve ever even Googled. Spotify is… actually getting quite good with its ‘Discover Weekly’ and other generated playlists, suggesting some real bangers often from artists I’d never heard. Nice, Spotify.

I think Steam, bless its heart, still has a lot of work to do. Given how much Steam knows about what I buy, what I play, and even how I play (as a point of comparison, at least, through achievements), and the same for people I play with, and for everyone else on Steam, lots of its recommendations seem really clumsy.

Still, it is better than a lot of other virtuastores. I will say that Itch‘s recommendations are solid but maybe my tastes align well with The Average Itch User.


  1. Darth Gangrel says:

    Doesn’t matter that Steam has rubbish recommendations, I still have too many unplayed games and a wishlist with games I’d otherwise forget (because they’re not e.g. Deus Ex-quality). Reading RPS and Swedish PC Gamer also makes me aware of any game I’d like to try in the postbacklog future.

  2. Jambe says:

    Hm. I find Steam’s recommendations about on par with Pandora/Spotify, which is to say maybe one in five recs is something I’m interested in. I’ve may’ve looked through a few hundred things in the queue since they introduced that feature, so perhaps I’ve trained it a bit?

    I like that the redesign’s supposed to surface less-prominent titles. I wonder if they’ll release any stats as to whether sales have been affected (lol, ofc they won’t).

  3. Premium User Badge

    Aerothorn says:

    As someone who wrote a paper on Steam in library school: the problem is that their metadata is incredibly limited (which Steam is obviously aware of) but they don’t know how to solve it. For instance, let’s look at the list of all game genres, according to Steam:

    Massively Multiplayer

    Anyone can tell you that there is an enormous amount of diversity within these genres (and, as Alice notes, Indie isn’t even a genre – it’s just a bucket Steam throws the presumably-hard-to-categorize indie games in). What Steam needs is a deeper controlled vocabulary for video games that represents them at a much more granular level. This doesn’t solve the problems of representation (nothing does!) but it gets us a lot closer to a smart engine.

    The problem is that, with that level of granularity, many developers can’t do it themselves; it’s not reasonable to ask them to be cataloging librarians and read a controlled vocabulary list for all of the fields Steam wants them to fill (let’s assume we were doing this for things other than just genre). At which point Valve has to do it in-house.

    Which…honestly, isn’t that hard, they have the money. But you know Valve, they want to crowd-source everything, and if that means their recommendations are crappy so be it. That said, I do understand the hesitancy; having to catalog every game that goes through greenlight, one that’s likely to sell no more than 1000 copies, probably isn’t a great use of their time (from a business perspective – speaking as a librarian, representing information accurately is always worthwhile :) )

    • Kefren says:

      Our librariany ways are often ignored. I remember when many search engines dropped things like include/exclude, phrase searching and so on.
      “How are people meant to find things?” I asked.
      “By letting us track all your behaviour and try and second-guess you,” was the answer.
      Good luck with that when many of the things I search for have no connection to previous searches.

    • bglamb says:

      You say that, but Spotify does really well and I’m pretty sure it isn’t catalogued like you suggest Steam should be.

      • Premium User Badge

        Aerothorn says:

        bglamb, it’s an apples and oranges comparison. Spotify is an all-you-can-eat model for songs, which take 2-5 minutes to consume. The result is a typical spotify user has listened to thousands of songs, and it’s easy to then cross-reference that with other users listening to thousands of songs and build a taste profile (and even this isn’t perfect – it tends to break down when your preferences are weird, like preferring certain types of lyrics – but as you note it works pretty good).

        Steam, on the other hand, is a single-sale platform for video games, which are generally longform. The average Steam user doesn’t have that many games (the powerusers that make up RPS commenters aren’t really representative) and thus it’s much harder to draw recommendations from that. This also leads to a bit of a vicious cycle, in which the lack of discoverability means people don’t play the vast majority of games on Steam, which means those don’t get included in ‘taste profiles’ *even if* the user would like them (because they don’t know they exist). And the system can only work if users trust it, say “this is outside my comfort zone but I’ll give it a shot.” Again, with a free song, this is easy to do, with a paid video game not so much. So a good video game recommendation engine would have to be a lot more reliant on metadata.

      • Merus says:

        It actually is: Discover Weekly is primarily built on playlists – Spotify do pay people to make playlists, but they also draw from other playlists made by their users.

        The advantage of playlists is that people try to curate them into songs that are similar, which ironically makes them better for recommendation engines than actual recommendations.

        They do some other things – they try and build a profile on what kind of things you like, and have an algorithm to identify outliers – but playlists are the core data set.

    • Nucleus says:

      It’s should be quite easy to do tbh, the whole cataloging process – just ask the community some questions about games – what kind of genre would you put it in? what’s the viewpoint perspective? real-time or turn based? what’s the setting or the time period? is it like these three other games?

      And they have the mechanics in place for that too, just look at steam controller preset sharing, and can be made quite granular and effective, if given some thought on what games are and how they differ from each other.

    • Hedgeclipper says:

      Steam has a ton of meta data though – they’re just going about it the wrong way. What they should do is ignore the genre tags completely (for the reasons you covered) and look at what people play – they’ve got a huge set of data on play times and also wishlist/follow/not interested information from millions of people – they should be looking at what people with similar profiles of game play times and wishlists to you and toss up recommendations based on things people with similar tastes liked.

      • Merus says:

        Did I play a lot of Civ VI because they’re similar, or because they’re for completely different audiences but I happen to be in both? If someone else played a lot of Civ VI, does that mean they should be recommended Dark Souls?

  4. NetharSpinos says:

    I’ve never really given much thought to Steam’s recommendations, in much the same way I would brush off a salesperson trying to recommend me…well, anything really. (Incidentally, Youtube’s recommendations seem solely engineered to inspire bewilderment and wrath.) I’m not too keen on the new layout either, especially the “trending amongst friends” bit. I couldn’t give a flying fuck about what my friends are playing as our collective tastes rarely seem to coincide.

  5. The Sombrero Kid says:

    It’s still filled with shovelware no one in their right mind would buy but you can now add the “Free to play” tag to a blocklist.

    The steam front page used to be 1 click to buy the game you already knew you wanted to buy. Now they seem to think they are press AND a shop, inexplicably unable to notice the obvious conflict of interests. So everything is geared towards showing you games you’ve never heard of.

    Clue: If I’ve never heard of it, I am not going to buy it & I don’t and never will trust steam as an independent source of criticism.

    • Someoldguy says:

      I don’t know if I’m in any way typical, but them shoving a load of titles in my face does get me looking up a few of them to see what reputable reviewers thought. Much in the same way that users here mentioning in a review comment that game X would be so much better if it was like game Y gets me looking up that game. Only if that then looks good do I check the Steam reviews to see if more recent player opinions sync with those of reviews that may be months or years old.

  6. Kefren says:

    I’d like an option to tick a box and not get tracked by software, and get no recommendations, and just leave me to it.

    Even better if that was the default everywhere, and only if you actually wanted Google and Amazon and Steam and 101 other sites to track all your behaviour and preferences did you tick a box. (Goes back to dreaming).

    • Someoldguy says:

      There are plugins and apps that curtail trackers significantly, but you’ll never get them all. It is nice though to visit websites and not be presented for ads about some item similar to something you looked at on Amazon a couple of hours earlier.

  7. DaceX says:

    To be fair, JAG was great. Loved it when I was younger,back in the late 90s, probably too young to understand it all. But that intro alone was gold. A 54-disk-collection might be a bit excessive, though.

    • DEspresso says:

      Only one Episode stayed with me, one which was set in the Yugoslavian Wars where a plane for unknown reasons first decided to use the exact same flight path to and from the target and with a burning engine decided No I will not land in any of the NATO Countries surrounding me, I will instead try crashlanding on a Carrier parked in the southern Adria.

      I guess that happens when you use parttime Lawyers instead of fulltime pilots.

  8. SuperJonty says:

    What I want is the option of saying that I’m not interested in something because I own it elsewhere (GOG, uPlay etc) and not because I don’t like it, so that it would hopefully recommend stuff I like but just not stuff that I have already…

  9. Creeping Death says:

    “Due to your recent playtime in other Singleplayer games”, I might like booby catgirl visual novel Nekopara Vol. 2. ”

    This is a silly recommendation. Why wouldn’t you recommend Vol 1? Who starts in the middle of a story?

    … Unless you already have vol 1 I guess.

    • kud13 says:

      Goodreads does this ALL THE TIME.

      “Because you’ve read volume 3 of a 6 book series, we think you’d like volume 2 in this trilogy”.

      That’s why I mostly get my book recommendations elsewhere.

    • charlesg says:

      Counterpoint: Saint’s Row 3

      The first two parts were pretty rubbish. Story-wise, it’s not really a big deal to miss out on them either.

      With games that simply aren’t story-driven (e.g. the Civilization series), the order in which they were released may not represent which one you should start with.

    • moridin84 says:

      They must have picked up for somewhere else that he’s into “booby catgirls”. The first volume has the flat kind I think.

  10. Sic says:

    I don’t understand why they base anything at all on genre. It makes no sense.

    They have the data for how much time players spend on any game, so why not base recommendations on those? Your time in-game is compared with every other player on Steam, and when the system finds someone very much like you (that enjoyed the same games), what differed between you could easily be communicated (Joe and Jill have played ~100 hours CIV6, ~50 hours Stardew Valley and ~20 hours Wolfenstein: The Old Blood—however, Joe never played To The Moon (Jill did, ~10 hours), and Jill never played The Room (Joe did, ~10 hours), thus, those games are recommended.)

    It would, obviously, also serve as a replacement for Tinder.

    • moridin84 says:

      I hate puzzle and adventure games so I’m happy to exclude them entirely.

  11. Sargonite says:

    I actually quite like the way you can exclude certain tags from the front page. I banned tags for three of the most annoyingly common kinds of games I encountered, and my page became considerably more interesting.

    • foszae says:

      I can’t say i’m satisfied with the filtering yet. The first thing i asked to block was ‘VR’ because i just have no interest in wearing clunky headgear that cuts me off from the rest of things i would normally pay attention to while i’m gaming. And yet, there’s hardly a day goes by that Steam doesn’t suggest that i’d be interested in seeing some Virtual Reality title or another. And that’s hardly the only tag that gets ignored by the way.

      Granted, i haven’t had anything pop up in the last couple days on my new front page, but the discovery queue still seems to completely disregard the filters i’ve set.

      • malkav11 says:

        If you go to your account preferences you can turn off display of VR, software, videos, and a couple of other categories across the store in general. Which is basically my favorite thing about the update because I don’t own a VR headset and so could not possibly care less about VR content. (I also have no interest in buying movies digitally or paying for utility software, but they’re a lot less prevalent anyway.)

  12. kud13 says:

    Please get a tag for MOBAs. And stop recommending them “because you own RTS X, Y, Z”

  13. Don Reba says:

    Amazon still mostly recommends replacements for the things I just bought. It’s as ridiculous as ever:

    • I see you like buying CPUs. Here’s a good one!
    • I noticed you just bought a white set of RAM sticks. Here’s a matching blue set!
    • That’s a nice unisex training shirt you just bought. Want a pearl necklace to go with it?

  14. thenevernow says:

    “‘Indie’ is a useless label in almost every situation and certainly here.”

    Half-agreed. But indie is an approach to game making which one can appreciate in general.

    ““Due to your recent playtime in other Singleplayer games””

    Ok, this is idiotic.

    ““Due to your recent playtime in other Strategy games”, by which I assume it means Civilization VI, it recommends Megadimension Neptunia VII, a JRPG.”

    This is because people abuse or misuse tags, but Steam trusts them anyway.

    “Because Alien: Isolation is on my wishlist, it thinks I may like Aliens: Colonial Marines. Ah. No.”

    This is a fair recommendation, at least from Steam’s point of view.

    ““Since you recently played Half-Life”, I may want Half-Life: Source.”

    This is also kinda fair, if a bit obvious.

  15. sneetch says:

    Apparently you can finally filter out early access games, that’s worth it for me.

    So tired of all the half-baked shite that pops into my queue, “we need your help us make Zombie Survival FPS no. 58624” no thanks, I already have a job.

  16. IshtarGate says:

    Blimey is that Alice’s playlist? Considering what little I know of Alice’s taste in music, that’s a playlist I shall be perusing with extreme interest.

    • Alice O'Connor says:

      Not quite. That’s part of the latest Discover Weekly playlist that Spotify automatically generated for me. Some good songs there, mind.

  17. melnificent says:

    The problem with how steam recommends is that it doesn’t account for playtime. Both over the short, medium and long term. Combining the 3 lists could give better results, even with the limited tags available.

    If they kept a list of the most played games per month, per user too that would increase the accuracy of prediction. Per week won’t make much difference as games are a consumed over a longer period than a 5 minute track or even 90 minute film.

    I noticed that with an increased library (nearly 3k) the recommendations break down completely. But the above would be the valve alternative to Netflix (paid staff to tag everything properly and consistently).

    • melnificent says:

      I forgot to add, achievements should also be a part of this, after all grinding trading cards will add a few hours of playtime to a game.

    • malkav11 says:

      According to Steam’s page about the new update, it does account for playtime.

  18. SuicideKing says:

    “Top Sellers” may also not be a global list, by default, for some countries. Don’t know how to tell what I’m looking at.

    • sunburned says:

      Yes, they have included a language filter – if you don’t enable English as a secondary language then your recommendations and lists will be very limited if you are let’s say based in Denmark or Poland. You will not see any titles which only have an English version – at least in most cases, if you don’t select the tiny ‘show me all XYZ games’ on the bottom right of each list.

  19. Mandrake42 says:

    Store aside, I wish they would add a way to organise your library by genre when you are looking for something to play. The current library feature is really cluttered when you have a lot of games. GoG have a simple genre drop down box that will filter your library to only games that match that criteria. If they can’t fix the shambles that is the store they could at least fix this ;)

  20. charlesg says:

    I’ve developed software that suggests recommendations.

    Steam appears to be employing a very basic algorithm, which just compares the game tags. As pointed out, the tags are limited and often wrong and inaccurate. MOBA’s are grouped with “Strategy”, and “Free to Play” includes both Free games and Freemium games. And some tags are blocked, like “Pay2Win”. Some are overly broad, like “Singleplayer” and “Indie”.

    Merely comparing game tags doesn’t require a lot of computing power. The total amount of data you need to look at is pretty small, it’s just the list of all games on Steam and all the tags for each game. With that data, you can generate a list of recommended games to go with each game, sortable with a “similarity score”, which is probably based on the amount of similar tags. All that’s left is to pick a threshold beyond which a game is insufficiently similar, and you remove these from the list.

    You can see very quickly how this leads to rubbish recommendations, especially with the tags being so inaccurate. But it’s easy to program and doesn’t take a lot of computing power.

    A much more effective and interesting way of recommending games, is to take a single game, e.g. Civ VI, and then look at other games owned by people who own Civ VI. You’re now actually looking at data defined by people’s purchase decisions (and hence, their personal tastes) so you get much more relevant and interesting recommendations.

    It should be obvious that this system means there’s a shitload of data to look at, and it will take a lot of CPU power to calculate the recommendations. And you’ll also need to re-run this program periodically, because people buy new games. And not to mention it’s harder to program and optimize.

    Maybe you want to include the amount of hours played. And exclude games that haven’t been played in a while. It’s not like games are removed from your library if you don’t play them.

    And maybe you want to consider that some games are often sold bundled together. Ultimate General: Gettysburg has been in a Chivalry: Medieval Warfare and Brothers – A Tale of Two Sons. These games are very dissimilar, but often seen together.

    But again, more data, more computing power, more complex to code. But absolutely possible. It may be worth it for Valve, as it can increase game sales.

    • amaranthe says:

      Agreed, it’s no surprise it’s rubbish when it’s not really doing any kind of true calculation.

      It’d be great if they’d implement a real system, aiming for the gold standard that is Netflix (not saying Netflix is perfect, but as far as estimated ratings and recommendations go it is pretty darn spot on).

      To that end, they’d need a better rating system. A simple Yes/No would recommend is not nearly good enough, especially when only a small minority actually bother to rate their games. Given a scale system and the knowledge that it would affect recommendations, I think people would participate.

      And I don’t think hours played is a great measure — it’s alright, but I think it’d complicate things by then having to compare player X’s hours played versus player Y’s hours played versus a player average, and that still really isn’t indicative of how much they liked the game (maybe they are low on hours because they haven’t finished, or play faster than the average person. Conversely maybe they’re high on hours because they leave their game running on pause all the time). A genuine rating system would remove most of that problem.

      And then the genre breakdowns and inaccuracies become less important, because while you can still get ratings based on games in that genre, it’s much more of a relative problem now — how much did you like games lumped in this category versus the other ones, compared to other players with similar interests?

      It definitely is complex computationally, but doesn’t seem all that complex heuristically.

      • charlesg says:

        For Netflix it’s easier to determine if you watched a thing or not.

        The rating/review system on Steam is pretty useless. It’s mostly used for sharing jokes and catchphrases.

        I find the negative reviews tend to be a bit more descriptive and helpful in deciding if the game’s for you. I’ve been spoiled by the Wot I Think articles on RPS. They’re on the long side, but really helpful as they’re very nuanced. “I really liked these parts of the game because … , but these other parts I didn’t like because …”

        I do like how you can see how many hours someone’s played a game, next to the review, so you know there’s more going on than they’re saying:

        “This game is bad!” 1265 hours played.

        “This game is great and has good replay value” 2.6 hours played (this one is very common with early access games)

        I’d like to see how much money someone spent on the game. Some games are pretty good value for money if you get them in a sale, but not for the full price. I’m looking at you No Man’s Sky.

        We can now see if someone got the game for free, it’s a start.

  21. sunburned says:

    As an indie dev I would say the update is half-bad. Valve managed to de-facto hide all English language titles on all the other language store pages. Which is an amazing achievement considering that most smaller titles start of with English only. Not an issue if you peruse Steam in English, but if you dare to use it for example in German “Achtung!” then you might miss out on a lot of games. We did quite well in Germany before the update … “Hände Hoch!”

    • charlesg says:

      That’s odd considering they also have added an option to see reviews in more than one language. I speak more than one language so I thought this was a welcome addition.

      Especially for extremely niche appeal products which have only a tiny amount of reviews. E.g. German-oriented Railworks DLC.

      This feature was added some time ago though, before the store overhaul.