So I started using Neeva.com recently and they also have accounts and subscriptions and allow you to rank your search results but overall I am not a fan in comparison to Kagi. Their search quality isn't as good and Kagi is far better in letting the user refine and customize their search results. What I did like about their search engine however is that when you boost and down boost a website, they use those as guides to help further personalize your search results based on what you've told Neeva you don't want to see and what you do want to see more of. It'd be cool for Kagi to have a toggle to allow for further personlization based on this concept.
8
Kagi should have a toggle to allow for further personlization based upon what you've already boosted and lowered
Vlad added the Clarification tag .
No idea what Neeva does, but from what NoGoogle said I'm guessing it's some sort of AI thing?
So (as an example) if you upvote StackOverflow, and many people who've upvoted StackOverflow also upvote GitHub, then it can automatically start showing you more GitHub results as well. (Looking at it one way, this is effectively some kind of collective preference/decision-making rather than an individual one. I think the intention here is different though; it's to have the search engine guess your preferences without you having to tell it).
Alternatively it might be done through some sort of "similarity" ranking, so if the algorithm finds out (through some other means) that Medium and Newsbreak are similar, downvoting Medium will also show you less Newsbreak posts.
Personally, I'm a little concerned that this would again introduce some sort of algorithmic black-box/filter-bubble which the user has no control over. But perhaps it'll work out if you adjust the strengths properly (eg. upvoting StackOverflow also shows more of GitHub, but I can override that by downvoting GitHub. The catch is to make sure downvoting GitHub doesn't also downvote StackOverflow by the same amount, because then there'd be no easy way to disentangle one from the other!).
I'm totally against that, I don't want to see the search engine acting like a social media. Adding it to a bookmark service (like Pocket, Tinygem, etc) is acceptable for me, though.
- Edited
Personally, I'm a little concerned that this would again introduce some sort of algorithmic black-box/filter-bubble which the user has no control over
This is exactly what will it create.
Imagine Kagi becomes popular with a certain community, say flat-earthers (nothing personally against). They decide to collectively promote certain types of sites and block other. If we take this as a signal and propagate to others ... well you can see the problem.
This is why Kagi does not want to be 'the arbiter of truth' type of search engine, but instead create powerful tools for users to be in the position to be that for themselves.
Vlad Imagine Kagi becomes popular with a certain community, say flat-earthers (nothing personally against). They decide to collectively promote certain types of sites and block other. If we take this as a signal and propagate to others ... well you can see the problem.
Oh wow, once you put it that way, it's definitely not desirable! I suppose that's the problem with any crowdsourced model. Although, regarding the other point,
Browsing6853 I'm totally against that, I don't want to see the search engine acting like a social media. Adding it to a bookmark service (like Pocket, Tinygem, etc) is acceptable for me, though.
Just playing the devil's advocate here but isn't search already acting like social media? Definitely in other search-engines, like Google's PageRank algorithm which takes into account how many people click on a certain link (among other things) opening the door to exploitation of "search voids". (I'm guessing Kagi avoids much of this somehow, but I'm curious to know how—probably off-topic though so I'm saving that discussion for Discord 
).
I'm still against this particular idea, but would it hurt to enable it for people who do want such personalisation? Or are we taking the stance that, since people come to Kagi to get away from some stuff, we shouldn't be giving them more of the same when they come here? If it's the latter, perhaps something to add not to the codebase but to the FAQ—a general question about whether we plan to/why we don't use more AI and crowdsourced data to optimise results!
3 months later
Kagi cannot know what the authoritative site for certain searches is, e.g. government websites or ISO standards, etc.
When I rank a site highly in kagi, it appears to only affect my rankings.
Kagi could make use of aggregated rankings to reorder sites with a tag like 'ranked highly by community members like you', so that the human intelligence of the users overrides the machine intelligence of the algorithm.
- Edited
What if Kagi becomes suddenly very popular with say flat earthers, and large number of them rank highly sites that carry this version of the 'truth' and then as a result of that we also start ranking high those sites in your results (collective community intelligence).
Hopefully you see a problem with that, and this is why Kagi does not impose any non-seaerch quality biases on the users, but give them tools to build what their own preferred ranking is.
Merged 2 posts from Community Site Ranking.
Perhaps the solution to crowdsourcing is to have trusted 'followed' users instead.
- Edited
I think Kagi is already on a great path with providing the user direct control over her filter bubble. But it isn't clear to me that having trusted followers or some other function to delegate control over aspects of this bubble to others is actually a feature. If I wanted to be a bit cheeky, I might even classify this kind of functionality as Depersonalization...
2 years later
Perhaps I'm missing something, but it seems the original idea got buried in a broader application. If the personalization is based only on my own set of raise/lower feedback, it doesn't appear to me that most of the concerns above really apply. Yes, there is an increased degree of opaqueness and complexity but if it is genuinely based on my own input rather than "people like me" that doesn't seem like a huge deal and could be toggled off.
I can certainly see things get complicated and challenging when you start using feedback from other users in an automated fashion and would agree with the generally conservative approach Kagi seems to be taking on this.
No one is typing