Live data from Hacker News

Evolving beyond 100B recommendations a day at Reddit

reddit.com

71–80 of 114 posts

Re: Evolving beyond 100B recommendations a day at Reddit

#71

I didn't realize that Reddit even recommends you content. And I can't tell the difference between before and after these recommendations are coming in. The blog post feels like one that Youtube made about their recommender system and how it improved but, their recommender system doesn't work. Reddit has been sort of this weird place where they face the issues that Digg and Twitter have and want to slowly transition i…

Reddit's recommendations are total garbage. But YouTube is actually even worse.

For instance, Eddie Van Halen recently died. So I clicked on a few links from Twitter to check out some concert highlights from his life and career.

Now my entire YouTube recommendations feed is nothing but Van Halen and 1980s rock music.

The Go-Go's recently got inducted into the Rock and Roll Hall of Fame. I clicked on a few links from Twitter to see some highlights.

Now my entire YouTube recommendations feed is nothing but the Go-Go's and 1980s pop music.

And not only that, but because I don't use Google search anymore, the only signal that Google can get from me is my YouTube watching, so my Google Now (swipe-left sidebar on Android phones) is nothing but Van Halen and the Go-Go's.

It's the dumbest freaking recommendation engine on planet Earth. It's ridiculous how bad it is.

Re: Evolving beyond 100B recommendations a day at Reddit

#72
post #5

Reddit is literally the last organisation on Earth I would take engineering advice from. I appreciate they are operating at a very large scale and that has its own challenges, but it is by far and away the worst engineered product I use. I really do mean from soup to nuts, Reddit is just garbage software. The modern front end is obviously and plainly terrible. The backend is slow and suffers very frequent downtime, o…

The front-end is a disaster, especially on mobile. It provides a paradigmatic example for people who believe SPAs are an invention from Satan himself. I actually love React and could have an SPA built that is a 100x better - but that is neither here nor there. The problem is that some features are half-assed or simply not implemented. The comment tree is a disaster, and pretty much in every interaction I discover new…

I’ve noticed sometimes it’s dismissals and sometimes not.

Re: Evolving beyond 100B recommendations a day at Reddit

#73
post #64

Earlier quoted context omitted.

I used to be an active mod on a somewhat active subreddit, not going to say which one. I can say that it was rather suspicious to get pinged to an AHS thread made about a post that was less than 7 minutes old. There's a report button under every post, so the idea that this person couldn't have just used that instead of posting a big dramatic post makes me highly suspicious of AHS.

There's no feedback loop on reports. They disappear into the ether, possibly never to be acted upon. It shouldn't be surprising users have zero faith they'll accomplish anything on their own.

Sure, but I think it's fair to wait longer than 10 minutes before doing a big post maligning a subreddit for "letting hate slide". The post being 7 minutes old before it was posted on AHS is not an exaggeration.

Re: Evolving beyond 100B recommendations a day at Reddit

#74
post #69
post #64

Earlier quoted context omitted.

I used to be an active mod on a somewhat active subreddit, not going to say which one. I can say that it was rather suspicious to get pinged to an AHS thread made about a post that was less than 7 minutes old. There's a report button under every post, so the idea that this person couldn't have just used that instead of posting a big dramatic post makes me highly suspicious of AHS.

AHS could have the best intentions but karma addicted folks jumping on such a post to get a blitz of upvotes is always going to happen. I'm amazed that the reddit mobile app pushes notifications about upvote thresholds by default - it is incredibly trashy.

I was pretty surprised because I just looked back at the AHS thread and there are hundreds of upvotes in there. It must've been euphoric

Re: Evolving beyond 100B recommendations a day at Reddit

#75
I once looked at the little notification tray on the new Reddit UI and saw a recommendation for a post. It piqued my interest, so I clicked it. It took me to the subreddit, not the specific post. I tried clicking different parts of the notification, same result every time. I tried scrolling down on the subreddit (in the new UI, ugh) and did not the find the post after thirty posts or so. I tried to copy the title from the notification, so I could paste it into Google, but user text selection was disabled by the CSS. I used “Inspect Element” to get a copy of the text, and put it into Google, which finally took me to the post.

All that engineering work to produce 100B recommendations, and it’s not even been put to good use.

Re: Evolving beyond 100B recommendations a day at Reddit

#76
post #5

Reddit is literally the last organisation on Earth I would take engineering advice from. I appreciate they are operating at a very large scale and that has its own challenges, but it is by far and away the worst engineered product I use. I really do mean from soup to nuts, Reddit is just garbage software. The modern front end is obviously and plainly terrible. The backend is slow and suffers very frequent downtime, o…

The other huge problem with reddit is that they aren't profitable, so far as I know. It would be one thing if the problem was just that their site was ugly, slow, frequently down, stuffed full of useless "features", and needlessly difficult to use BUT they were making tons of money. If that were the case, while you might not want to exactly recreate them, you could make the argument that they were at least doing something right. As is (so far as I know) they also don't even make money despite their immense traffic.

All of the value of reddit comes from the users adding, sorting, and moderating content. The people running the site actively make it worse and they aren't profiting by doing so.

Re: Evolving beyond 100B recommendations a day at Reddit

#77
My own experience says that you should not attempt to make a serious change without being able to evaluate whether or not your attempt has worked. That could be as simple as fixing a bug - you need a test that would have caught the bug and it (and all your other tests) needs to pass once your fix is in. That could be as complex as something like this, improving your recommendation system. To evaluate something like this you'd likely need a few key metrics, latency and throughput, sure, but also recommendation quality. Maybe click through rate? Growth in recommended subs? Time that users spend on reddit with and without recommendation? Things like that.

If you can't evaluate your change then you will be stumped by a very simple question. Did you succeed? Are you done? If you fix a bug by thinking "I know what the problem is", then change a few lines of code and say "Done" how do you know you actually fixed it? How do you know you didn't make things much worse? Same thing here. So, you changed the recommendation system - did you make it better or worse?

I was quite interested to see how they were evaluating the system and whether or not it was an improvement. Their section on evaluation says "Scaling this component of the system is something we think about a lot and are actively working on." That seems really, really, backwards to me. You are thinking about evaluating it after you've changed the system? Huh?

Re: Evolving beyond 100B recommendations a day at Reddit

#78
Oops, something went wrong!

"This component maintains a 99.9% uptime and constructs a feed with p99 in the low hundred milliseconds. Which means that this design should hold as we scale to handle trillions of recommendations per day."

Is that why Reddit is the only major website that regularly can't keep up with demand? They are wasting all their resources on sending millions of spam emails filled with these 'recommendations'?

Re: Evolving beyond 100B recommendations a day at Reddit

#79
post #59
post #23

Earlier quoted context omitted.

It’s a bit silly considering that this site is pretty much the same as any Reddit subreddit.

HN is about as different as you can get from Reddit. For example, Reddit encourages herd behavior, minimal-effort glib interactions, and addictive content. HN discourages herd behavior, is quite difficult for most newcomers because the higher effort required, and even includes a built-in anti-procrastination feature.

You can tell yourself that, but it just isn’t true. It really depends entirely on the subreddit itself. There are plenty of Reddit subreddits with high quality content, discussion, etc.

Re: Evolving beyond 100B recommendations a day at Reddit

#80

I didn't realize that Reddit even recommends you content. And I can't tell the difference between before and after these recommendations are coming in. The blog post feels like one that Youtube made about their recommender system and how it improved but, their recommender system doesn't work. Reddit has been sort of this weird place where they face the issues that Digg and Twitter have and want to slowly transition i…

Reddit's recommendations are total garbage. But YouTube is actually even worse. For instance, Eddie Van Halen recently died. So I clicked on a few links from Twitter to check out some concert highlights from his life and career. Now my entire YouTube recommendations feed is nothing but Van Halen and 1980s rock music. The Go-Go's recently got inducted into the Rock and Roll Hall of Fame. I clicked on a few links from…

"Am I prepared to get more of this recommended until the end of time?"

I ask myself this question every time I watch a youtube video. A lot of them I watch with private browsing to avoid it. Though lately it feels like I'm getting stuff recommended despite the private browsing measure.

Post reply on HN