Hacker News Reader

Comment by natch | Hacker News Reader

TL;DR: Lamentably, Google has facilitated a transition to organic search results being strictly a money game for most categories, removing the joy of discovery and disproportionally impacting small businesses.

What's missing is a suggestion of how to fix it. Not an easy problem.

View on HN

I was wondering about this the other day, and considering another approach to search, which I'm sure others have thought of too as it's fairly obvious. To restate the problem:

For any given search term there are millions of hits

Of those hits, only the first 20 or so in a given ranking will ever be seen

This renders a huge amount of the internet invisible and a huge amount of indexing/linking moot

At present google uses what sounds like a very complex mix of signals about linking, reputation, novelty etc which it constantly tweaks, and delivers a relatively stable ranking which guarantees that those with deep pockets will win the game of ranking highest simply by buying top position or buying authority in various ways (sponsored blogs, buying authoritative domains, content farms etc).

Another approach would be to constantly inject random search results for a given term into results, and use solely reader reactions to results to rank them (registered user voting, do they come back to search results or not, sites chosen as high quality by this user, that sort of signal). So always have some randomly chosen results to guarantee mixing, and always include some degree of user input to give better and more relevant results chosen by humans. I think you could build good results like this, while completely ignoring any sort of complex ranking algorithm which is both too stable, too prone to gaming, and in conflict with the revenue stream of ads. Obviously you'd have to be careful about which users you listened to, but that is a solvable problem if you can persuade people to register to get better results.

It's definitely well past time for a few serious non-corporate competitors to google to emerge, if only to keep Google honest.

Which results users click on, and especially which results they click on and then don't come back is a huge input into Google's ranking algorithm.

In fact, this is pretty big barrier to entry when creating a competing search engine. Google's scale enables them to get a much better signal from this input than anyone else can.

Yes I've read about that elsewhere, but they do still use a lot of other signals AFAIK. The more complex their algorithm becomes, the less inclined they will be to change it, and the beauty of only using user input is that you get an ever changing signal and people start to be less obsessive about their exact ranking, because it changes constantly. Of course it does introduce problems of its own with trust.

Re competitors, I think that's more about eyeballs than quality - Google is such a behemoth now and is entrenched with revenue from advertising - they can use their other properties and interests like mobile to feed search revenue. I really don't think they're at the top solely because they return the best results any more, because they've been without significant competition for so long.

I was actually more interested in the idea of injecting some random results (say from the top 10 pages), as that might significantly increase the relevance of their results for many queries, give them even better data from user input, and would make it very hard to game the system.

The idea that you could only use user input is not realistic. 15% of Google's incoming queries have never been seen before by the search engine. There's not enough user input available.

how would you track the user inputs in order to create then rank the results further? via browsers? or...?

while I like your idea, I'm not sure if it is feasible and second, if there is an 'acceptable' method which most people will not object to or try to game.

for example, imagine 1000's of mechanical turks being paid cents to click and browse to a specific link in order to fool the algorithm you've set up

Tracking users and combatting fraud would definitely be an issue (just as it is for google), and you might have to set up trusted rings of users or some sort of web of trust rather than just trusting any user. It'd be interesting to try - tracking would have to be by browser cookie and actions like returning to results or upvoting/downvoting on the results page and you'd have to be very strict about fraud prevention from the start.

Perhaps more interesting than a specific ranking algorithm though is the idea of injecting random results into the mix. Perhaps Google has tried all these things already, it wouldn't surprise me and I know they look at user reactions like returning to results/time on page etc, but it'd be nice to see some search engine which is not owned and controlled by a massive corporation like MS, Google or Apple though - is anyone doing something interesting with common crawl?

There's a German metasearch engine funded by an NGO & University, which also funds the Yacy P2P engine: https://www.metager.de/en/ and http://yacy.net

Scholar Ninja is trying to use WebRTC for distributed indexing and searching from a Chrome extension, but there's a Chrome/WebRTC bug which prevents long-running tasks: http://juretriglav.si/an-open-distributed-search-engine-for-...

Thanks

The problem with injecting search results is that most of the time, it is going to give users a worse experience, not a better one.

I'm not convinced that it would - is there a qualitative difference between page 2-3 or even pages 1-10 of Google results and the first results on page 1? A random mix of the first 10 pages of results say I'd say would be more interesting than their current mostly static and gamed results, which are basically pay to play.

The most valuable thing about this idea is that it is immune to gaming because of the mixing. Just as an example relevant to the point of the original article, here is a search for hotels in Devon on google:

https://www.google.co.uk/search?hl=en&q=hotels%20in%20devon

First half of the page are links to Google places entries - so advertising for Google. Are these paid for? I'm not sure what the situation is there but this is partly a manual intervention from Google to try to save their results... After that on page 1 we have lots of aggregator/review sites like laterooms, tripadvisor, stayindevon etc (your results may vary). by the second page of results we have some real hotels, by the third, we have more real hotels, etc, up to page 9, which is still real hotels in devon.

So I'd say the first 10 pages are pretty much equally useful, and some mixing of results would do wonders for ranking and allow good results to surface instead of languishing unless they make an entry on a google property, or on one of the portals.

I like your idea. A small amount of randomisation can be very effective at (1) defeating gaming, (2) finding unexpected signals in complex systems (3) Preventing over-fitting of results when optimising.

All three of these advantages come into play here.

Without going into too much details, just randomly mixing the first ten result in the first page will totally destroy Google's results: anyone will notice in a matter of days, if not hours. And everyone will find Bing so much objectively better that it wouldn't even be funny.

For example,

"rosetta spacecraft": the first two results are from ESA's homepage, the tenth is an article from The Guardian.

"mauna kea weather": after a gigantic weather box, the first result is from Mauna Kea Weather Center, the tenth is a page from some generic website that deals with all mountains.

And on and on.

Without going into too much details, just randomly mixing the first ten result in the first page will totally destroy Google's results

I didn't suggest doing that.

Good analysis

Not a solution, but a thought: it's easier to game a single search engine that owns the market than it would be to game 3 or 4 competing and different products with a significant market share each.

Google's success has its downsides.

Reply on news.ycombinator.com