Some thoughts on the problem, not intended as a complete proposal or argument:

* Indexing is expensive. If there's a shared public index, that'd make it a lot easier for people to try new ranking algorithms. Maybe the index can be built into the way the new internet works, like DNS or routing, so the cost is shared.

* How fast a ranking algorithm is depends on how the indexing is done. Is there some common set of features we could agree on that we'd want to build the shared index on? Any ranking that wants something not in the public index would need either a private index or a slow sequential crawl. Sometimes you could do a rough search using the public index and then re-rank by crawling the top N, so maybe the public index just needs to be good enough that some ranker can get the best result within the top 1000.

* Maybe the indexing servers execute the ranking algorithm? (An equation or SQL-like thing, not something written in a Turing Complete language). Then they might be able to examine the query to figure out where else in the network to look, or where to give up because the score will be too low.

* Maybe the way things are organized and indexed is influenced by the ranking algorithms used. If indexing servers are constantly receiving queries that split a certain way, they can cache / index / shard on that. This might make deciding what goes into a shared index easier.

> Indexing is expensive. If there's a shared public index, that'd make it a lot easier for people to try new ranking algorithms. Maybe the index can be built into the way the new internet works, like DNS or routing, so the cost is shared.

But what are you storing in your index? The content that is considered in your ranking will vary wildly by your ranking methods. (example - early indexes cared only for the presence of words. Then we started to care about the count of words, then the relationships between words and the context. Then about figuring out if the site was scammy, or slow.

The only way to store an index of all content (to cover all the options) is to...store the internet.

I'm not trying to be negative - I feel very poorly served by the rankings that are out there, as I feel on 99% of issues I'm on the longtail rather than what they target. But I can't see how a "shared index" would be practical for all the kinds of ranking algorithms both present and future.

> The only way to store an index of all content (to cover all the options) is to...store the internet.

An index cannot hope to cover all options, these ideas are antithetical.

this is a pretty killer idea

How about open sourcing the ranking and then allowing people to customize it. I should be able to rank my own search results how I want to without much technical knowledge.

I want to rank my results by what is most popular to my friends (Facebook or otherwise) so I just look for a search engine extension that allows me to do that. This could get complex but can also be simple if novices just use the most popular ranking algorithms.

I think Facebook really missed the boat on building their own "network influenced" search engine. They made some progress in allowing you to search based on friends' posting and recommendations to some degree but it seems to have flatlined in the last few years and is very constricting.

One thing I haven't seen much on these recent threads on search is the ability to create your own Google Custom Search Engine based on domains you trust - https://cse.google.com/cse/all

Also, not many people have mention the use of search operators, which allows you to control the results returned. Such as "Paul Graham inurl:interview -site:ycombinator.com -site:techcrunch.com"

That would bring to an even bigger filter bubble issue, more precisely to a techno élite which is capable, willing and knowledgeable enough to feel the need go through the hassle, and all the rest navigating in such an indexed mess that would pave the way to all sort of new gatekeepers, belonging to the aforementioned tech élite. It’s not a simple issue to tackle, perhaps a public scrutiny on the ranking algorithms would be a good first step.

I disagree. The people who don't know anything and are unwilling to learn wouldn't be any worse off than they are today and everyone else would benefit from an open source "marketplace" of possible ranking algorithms that the so called "techno elite" have developed.

I think the proposed improvement to the web in its intentions should mostly benefit the "ignorants", not those that can already navigate through the biases of today's technological gatekeepers. Please note, ignorants are not at fault for being so. Especially when governments cut funds for public education, and media leverages (and multiplies) ignorance to produce needs and sales, fears and votes. Any solution must work first to make the weak stronger, more conscious. A better and less biased web can help people grow their unbiased knowledge, and therefore exercise their right of vote with a deeper understanding of the complexity. Voting ignorants are an opportunity for the ill intentioned politicians, as much as are a problem for me, you and the whole country.

blekko and bing both implemented ranking by popularity with your Facebook friends, and the data was too sparse to be useful.

If the details of a ranking algorithm are open source, it would be easy to manipulate them.

Open sourcing the ranking... YES!!!

I wonder if indexing and ranking could be decentralized. Lets say we design some data formats and protocols to exchange indexing and ranking information. Then maybe instead of getting a single Google, we could have a hierarchical system of indexers and rankers and some sort of consensus and trust algorithm to aggregate the information between them. Maybe offload indexing to the content providers altogether, i.e. if you want your website found, you need to maintain your own index. Maybe do a market on aggregator trust, if you don't like a particular result, the corresponding aggregator loses a bit of trust and its rankings become a bit less prominent.

Spitballing here, but what if instead of a monolithic page rank algorithm, you could combine individually maintained, open set rankings?

===Edit=== I mean to say you as the user would gain control over the ranking sources, the company operating this search service would perform the aggregation and effectively operate marketplace of ranking providers. ===end edit===

For example, one could be an index of "canonical" sites for a given search term, such that it would return an extremely high ranking for the result "news.ycombinator.com" if someone searches the term "hacker news". Layer on a "fraud" ranking built off lists of sites and pages known for fraud, a basic old-school page rank (simply order by link credit), and some other filters. You could compose the global ranking dynamically based off weighted averages of the different ranked sets, and drill down to see what individual ones recommended.

Seems hard to crunch in real time, but not sure. It'd certainly be nicer to have different orgs competing to maintain focused lists, rather than a gargantuan behemoth that doesn't have to respond to anyone.

Maybe you could even channel ad or subscription revenue from the aggregator to the ranking agencies based off which results the user appeared to think were the best.

Well I suppose Google has some way of customizing search for different people. The big issue for me is that google tracks me to do this. Maybe there could be a way to deliver customized search where we securely held the details of our customization. Or we were pooled with similar users. I suppose if a ranking algorithm had all the possible parameters as variables, we could deliver our profile request on demand at the time of search. That would be nice. You could search as a Linux geek or as a music nut or see the results different political groups get.

Building something like this becomes much easier with Xanadu-style bidirectional links. Of course, building those is hard, but eliminating the gatekeeper-censors may finally be the incentive required to get bidi links built. It's also worth noting that such a system will have to have some metrics for trust by multiple communities (e.g. Joe may think say, mercola.com is a good and reliable source of health info, while Jane thinks he's stuck in the past - People should be able to choose whether they value Joe's or Jane's opinion more, affecting the weights they'll see). In addition (and this is hard, too), those metrics should not be substantially game-able by those seeking to either promote or demote sites for their own ends. This requires a very distributed trust network.

I like the idea of local personalized search ranking that evolves based off of a on device neural network. I'm not sure how that would be work though.

Sounds like ad-Blocker repos, nice!

Not to mention all the people who will carefully study whatever new system, looking for their angle to game the ranking.

> When you rank, you pick winners and losers.

...To which people responded with various schemes for fair ranking systems.

...To which people observed that someone will always try to game the ranking systems.

Yep! So long as somebody stands to benefit (profit) from artificially high rankings, they'll aim for that, and try to break the system. Those with more resources will be better able to game the system, and gain more resources... ad nauseam. We'd end up right where we are.

The only way to break that [feedback loop](https://duckduckgo.com/?q=thinking+in+systems+meadows) is to disassociate profit from rank.

Say it with me: we need a global, non-commercial network of networks--an internet, if you will. (Insert Al Gore reference here.)

(Note: I don't have time to read all the comments on this page before my `noprocrast` times out, so please pardon me if somebody already said this.)

This is a bang on distillation of the problem (or at least one way to view the problem, per "who controls the ranking and how it's defined").

That’s a very useful distinction, that brings me to a question: are we sure that automating ranking in 2019, on the basis of publicly scrutinized algorithms, would bring us back to a pre-Google accuracy? Also, ranking on the basis of the sole query instead of the individual, would lead to much more neutral results.

Absolutely spot on... I've been using DDG as my default search engine for a couple months. But, google has a huge profile on me. I find myself falling back to google a few times a day when searching for technical terms/issues.

Couldn't you just randomize result ordering?

You know how google search results can get really useless just a few pages in? And it says it found something crazy like 880,000 results? Imagine randomizing that.

---

Unrelated I searched for "Penguin exhibits in Michigan". Of which we have several. It reports 880,000 results but I can only go to page 12 (after telling it to show omitted results). Interesting...

https://www.google.com/search?q=penguin+exhibits+in+michigan

If you think of it as like an old fashioned library or an old fashioned Blockbuster video store.

Sure you could read any book ever printed in the English language in the local library. They might have to get it in from the national collection or the big library in the city. But you ain't going to see every book in the local library. There is more than you could wish for and you will never read every book in the local library. But all the classics are there, the talked about new books are there (or out on loan, back soon). All the reference books that school kids are there, there is enough to get you started in any hobby.

Google search results are like that. Those 880,000 'titles' are a bit like the Library of Congress boasting how big it is, it is just a number. All they have really got for you is a small selection that is good enough for 99% of people 99% of the time. Only new stuff by people with Page rank (books with publishers) get indexed now and put into the 'main collection'.

Much like how public libraries do have book sales, Google do let a lot of the 880,000 results drop off.

It's a ruse!

I heard that they also filter results by some undisclosed parameters, like they don't show you anything that hasn't been modified in the last ~10 years, no matter how hard you try

Yeah, this is a real problem for research into older things that have no need to change. Google seems to think that information has a half-life. That's really only true in the social space. Truth is eternal.

Sure, but then whoever gets to populate the index chooses the winners and losers, because you could just stuff it with different versions of the content or links you wanted to win and the random ranking would should those more often, because they appear in the pool of possible results more often.

That would make it waaaay less useful to searchers and wayyy easier to game by stuffing results with thousands of your own results

I suspect just randomizing the first 20 or so results would fix most problems. The real issue is people putting effort in to hitting the first page, so if you took the benefit out of doing that people would look for other ways to spend their energy.

If you find nothing useful, just refresh for a new set. It would also help discovery.

Sounds like a great ux