Hacker News Reader

Comment by amelius | Hacker News Reader

Yes, give me a downvote button on Google results.

I worked on a search engine at a startup that did exactly this, you could up and downvote each result. The main feature was that we essentially "sharded" the search engine so it could be embedded on different sites and give different results based on each community. So a search for "casting" on a fishing website would give different results than the same search on a metallurgy site, as voted by each community. We could also learn passively by watching which links users clicked on and where they "disappeared" from the search engine - presumably the last clicked link was the result they were looking for.

Google did copy the voting feature on their results page briefly but abandoned it. [0] This was back in 2006-7. We learned the hard way that it's pretty much impossible to compete with Google in search even when you're innovating. They either copy you, or can just blackhole you out of existence.

[0] https://techcrunch.com/2007/11/28/straight-out-of-left-field...

> presumably the last clicked link was the result they were looking for.

This is a risky assumption. I tend to click all the links first, and only then check if they're the results I'm looking for.

You can work around that by using javascript (checking for tab losing focus) and checking how quickly the links are being clicked. If you middle-click a bunch of links, then don't click any more links, then it can be inferred that one of the links that you clicked was "good".

How can that be inferred? Maybe the results were all bad and I'm refining my search query. Or trying a different search engine.

>Maybe the results were all bad and I'm refining my search query

that can be detected, by checking further searches are made using the same session.

>Or trying a different search engine.

Maybe if your audience is mainly HN users it might cause an issue, but I think most people don't bother with that so you can count it as noise.

The lack of an ability to vote on search results seems like a baffling omission. Normally Google loves to crowdsource their work, but for this one area where we would actually want it, they decide that they know better when they clearly don't.

It would be abused into the ground by the same people who set up these scam sites in the first place. For most things on the internet, the crappiness of the average person is the limiting factor.

They don't have to make it a public vote. I'm fine to do this work myself for a week or so, and then enjoy crap-free searches. Or link my votes account to someone who I trust and use their opinion.

Idk why everyone is praising public anything, because local communities and knowledge webs worked fine pre-internet. Most of the bullshit came with globalization.

I wish there was some standardized way to say "I trust you" on the internet, and share part of their information bubble (e.g. reviews, up/downvoted websites, youtube recommendations, etc.), and this bubble could have some transitive properties (if I trust X and X trusts Y, then to some extent I trust Y too).

But it's probably too privacy sensitive (if I see which sites you upvoted, then I have some information about you). Hence, that's probably why this has to be either completely private or completely public.

Well, every social media network tries to capture and monetize this, of course.

But yes, it would be nice to have this in a less proprietary way. But I fear it's either going to be a privacy issue (because the person you trust has to publish every page they like and dislike), or it's going to be anonymous and therefore easily gamed by bad actors.

If the group of trusted people in a bubble is large enough (say 1000s of people), then it wouldn't be a problem, I suppose.

For example, I wouldn't mind sharing my upvoted websites/videos/products with everyone on HN, as long as it is anonymized. Bad actors can be distrusted by the community, I suppose (though moderation seems to be largely an unsolved problem still).

Anonymization is actually really, really hard.

Click up/downvote.

Click/dismiss "make it private/public" bubble.

Set default vote visibility in settings.

Enable/disable the bubble.

List/add/remove sites in a privacy list.

But the common notion is that it's probably too hard for "regular" users, so we have the internet of nonsense instead. Few more years and we won't find anything good.

What if they took only votes from people that pay and have account for X time? Also what’s the problem with adding revert per user? With some events architekture it should be trivial. They could even train AI on this data to automatically sugest suspects.

Here's an easy thought experiment: for any propose of the type "what if they had a system that did X?", imagine that someone tells you they'll give you 10 million dollars if you can figure out a way to game that system that can't be detected by Google.

If you spend 5 minutes on it, can you think of a way? If you can, congratulations, now imagine millions of other people thought about the same tricks and you get the reason why Google can't ever really win against SEO.

Google might not be able to 'win', but they can stay ahead.

Eg if they can get to a place where SEO efforts point in the same direction as making your website genuinely more useful to people, then that's good enough for Google.

It's totally fine if they only use my own votes to modify my results. In fact, if we're going to do that, I'd really like to block some sites from my search results entirely. Though ideally, I would also like to include the votes from people I trust.

People sell good accounts for lots of money to the spammers.

And when those accounts suddenly start behaving like bad accounts, their votes get treated as such. There will be a window of confusion, but it might not be that big. And there would be an economic incentive for people to establish trusted accounts with high-quality voting behaviour.

Their entire business (advertising) is clustering. They've already done the hard bit.

Just cluster people by downvotes and whatever other thousand metrics are already being tracked, and allow them to see results given by other clusters by showing which areas are dense on a PCA or something or saying 'I want my results to only be influenced by people who have downvoted github.org' or whatever.

Wouldn't that provide more data on who are the scammers?

Not really. In these areas there's a general principle that the moment you start using a signal to enforce on bad actors that signal's value decays. It's sad, I actually think web3 could help curb abuse at scale by making it too expensive to scale horizontally. Imagine paying a "stamp" to send an email on the blockchain, spam email would dramatically drop. Maybe similar mechanics could be in play for other things but then you lose democratization of information.

What is web3? What exactly is it? And how could it be used in this case?

The idea is you'd have an email client but all data is stored encrypted on chain and you'd have to pay tx costs to send things. By having this sort of stamp tax spammer wouldn't be able to scale mass messaging as it would bankrupt them.

in meatspace though, my real mailbox is mostly filled with spam that the sender paid to have delivered. granted, it would at least act as a limiting factor.

If you have an auditable tx history, then mixers/exchanges/wallets etc can be marked as sources of the money used by spammers

Isn't Google's monetization for Search results the primary reason for not allowing crowd funded quality metrics? If only good results get to the top how do they make money? Google's Search business model appears to be predicated on poor results getting disproportionately prominent listing.

The actual sponsored links end up at the top anyway. I don't see how bad regular results would contribute to their revenue; it would just drive people away.

>I don't see how bad regular results would contribute to their revenue

Let's say you run a good song-lyrics site that has the correct lyrics for everyone's favorite songs. You happen to be on page 2 of google results for common queries; all of page 1 is taken up by spammers, fake pages, etc.

How can you possibly drive traffic to your site? Maybe you can invest in SEO but no promises there. You'd be competing against people whose whole focus is SEO and nothing else. The only option left is to buy ads.

Yeah, but so can the spammers and fake pages.

And it all works together to make Google worse.

I agree that it would work as long as Google has an absolute monopoly on search. Google wouldn't care how bad their search results are, because there's nowhere else to go. But if there are alternatives, users should stop using Google and use the alternatives instead, and then Google has an incentive to improve their search results for users again.

Let's say we have three search results, good, average and bad, displayed on one result per page in that order. The user is happy, Google is unhappy as there is no motivation for the site owner of the best search result to pay Google for their listing. The only way Google gets money is if it enables the result order to be modified, i.e. the owners of the average or poor sites to disproportionately effect their listing prominence by appearing on page 1. Google do this by allowing paid adverts. Bad or average search results thus get disproportionately high prominence in the listing with the knock on effect that the owner of the best search result now has motivation to pay Google to regain their primary listing. And even further, if the site owner of the best search results now has to pay Google for the prominence of their listing why waste money on continuing to provide the best results? The user is unhappy, Google is happy.

But you do vote on search results. Google knows if you stay on a page or leave it quickly for another result. That's an implicit vote that they definitely take into account.

Having that explicit button might not really add any additional value.

Which heavily penalizes sites that make information easy to find.

Especially for use cases like reviews where you are looking for multiple opinions.

The hard to navigate site full of waffle, and SEO duckspeak nonsense gets a positive while the site with clear concise information the user can absorb in 2s gets penalized.

How? I don't use Chrome.

Browserfingerprinting, they see you again on ther resultspage after x seconds -> bad result. They don't see you again -> good result.

That's not how I use the results page, though. I open the top x number of results in separate tabs, and then check those pages. I only go back to the tab with the results page after I've checked them all, so that might be a while, even if they all suck.

So if they look at my behaviour the way you say, then the feedback they'd get from me would be that the top x-1 results are always bad, while the xth result is always good. That sounds like a poor algorithm for them, but it might explain why my Google results always suck.

Thanks to things like Google Analytics, Google knows what site you're viewing at any moment in time. It can see you go through the results sequentially and stop at some point.

Luckily for search, it is how the vast majority of users use things. Less than 1% of users doing things differently makes the stats a little messier, but still works quite well.

It's a huge vector for abuse. The signal:noise ratio wouldn't improve at all.

If I can block the 5 word soup gpt2 fake support sites that make up the top 20 results for obscure debugging messages from my own personal account, then my signal to noise ratio went from 0 to some number above 0. This is an infinite improvement for the first page.

Noone ever wants to go to xypdf.com for any reason unless they want to feel like they just had a stroke, how is (was? it made me stop using google except as a last resort in 2019) it often 3 of the top 5 results?

I wish I could just invert their SEO quality metric (there was a golden window around 2018 where you could just type -best into search engines that still respected subtraction to get only good reviews but sadly quality sites have fallen into line with the duck speak). I feel it's a pretty reliable indicator of garbage.

The signal:noise ratio would certainly improve for anyone voting even if they didn’t use it to adjust other people’s search results.

I’m also skeptical that all of Google’s enormous investment in ML and staffing is completely powerless to identify bad actors with atypical usage patterns. What seems far more plausible is that they’ve decided it isn’t costing them more in ad sales than it brings in. There are individual domains which would improve results by being blocked but they also pay for search ads so … unsolved grand challenge of computer science it is!

> I’m also skeptical that all of Google’s enormous investment in ML and staffing is completely powerless to identify bad actors with atypical usage patterns.

Isn't that exactly what people are complaining about here though? At least part o the problem with search results is that google seems powerless to recognize and remove useless bad actors (stack overflow copies, etc.) from their index.

My position is that they could do far more if they cared - simply blocking spam domains like it’s the previous century would make my experience better – but that they have made a business decision not to devote more resources to the problem.

Yes, Google says they want all our information to "improve our user experience". Except it's not very credible that this is the true reason.

Come try Kagi Search and get both upvote and downvote!

Reply on news.ycombinator.com