This is great. I had a similar idea recently as I was reading Python documentation: suggest to me random technical things I would like to read or would benefit from reading. We could even have this for all text on the internet (and books), or for all webpages really.<p>I read a few dozen articles on my phone on Safari, here's my feedback:<p>- implement some visual feedback like a swipe animation (the next/previous article pulling in from the side or something that lets me quickly evaluate the next article like a minimap or its Contents) and increase the cancel zone a bit (it should be relative to phone width), it feels brittle, I feel like I have to be careful when scrolling.<p>- I wondered whether the swipe direction is a like/dislike or if the direction doesn't matter. I assumed it didn't matter but turns out right goes back, but after you go back, left swipe gives you a new article instead of going back to the article you went back from, this is not good because I don't mind swiping a few extra times as much as I mind losing an interesting article, so doing it this way means I need to consider anytime I want to go back whether I'm okay with losing the current article instead of having "back" be a safe operation. Also, maybe make it scroll the same way as TikTok, down. So I can scroll down for the next article after I scroll through the entire article or up to go back to the previous one, that way left/right scrolls can be used for something else. This would make it hard to skip long articles, maybe make both left swipe and down swipe go to the next article? I'd have to test to see if it was a good idea.<p>- going back/forward should preserve scroll position<p>- the back button doesn't always work, I often lost articles I wasn't done reading because<p>- scrolling up can refresh the page and lose my current article<p>- scrolling sideways on a table or long formula triggers a swipe<p>- I experienced some jank where I sometimes get sent to another page, I'm not sure why it happens maybe I refreshed and it sent me to a new recommendation<p>- I like that you removed citations and other "cruft". I saw an empty Footnotes section header on "Sub-pixel resolution" that should've also been removed. Also, the articles have extra padding at the end from the last section that you should shorten<p>- sometimes when I click on an article while reading an article, I'll get the article I clicked on recommended soon after, sometimes more than once actually<p>- some articles have lowercase titles but on WikTok they are capitalized<p>- image previews are lower resolution than they are when I open the page on Wikipedia, you might need 2x srcset directives to the image elements<p>- tapping on an image should open the max resolution image file<p>- I was recommended some genetics article and clicked through a couple articles and the algorithm didn't start only suggesting those articles, which was good<p>- I got a place disambiguation page of some villages in Poland and then one of the villages, not very interesting<p>- improve the top text: drop the ".org", just "WikTok" (it's cleaner), make it the same sans serif font as the article or the serif of the headers and use an actual minimal shuffle icon and don't make it part of the hyperlink (if you keep it a hyperlink at all, it should really be a button)<p>- some articles are more fundamental than others and should be recommended first. some articles (especially for math) require prerequisite knowledge so it's less efficient to show me the more specific one first, I might waste time trying to understand regardless but it would be better to show me simpler articles first, I guess an advanced algorithm would have to try to intuit how much I know (simple heuristic: the average person doesn't know what Lie groups are) and have some kind of graph of knowledge dependencies that Wikipedia articles form, probably from how they link to each other. At that point it's more like teaching, you estimate what I know to guide me through the graph of concepts more efficiently<p>- preload the next page<p>- it seems like when I tab out you might not be tracking that and considering that as me still reading the article<p>- does the algorithm track how I scrolled the page, which words I dwelled on etc.? You could probably determine the level of my understanding of an article from how I scroll it and then re-suggest articles I didn't understand or read fully<p>- external links in infoboxes aren't clickable links, which is bad when the article is about a website and I can't click to go to it<p>- some long, important articles I want to read but not right now<p>You're solving a bit of a different problem than TikTok because there's only 6 million Wikipedia articles (much less useful ones) so your "power users" would be people trying to read it (or some section of it) exhaustively over a longer period of time, so you might want to cater to that by letting people see more of what you track, like what articles they've read and show them long lists of related articles for the current article, so that you can see that you're making progress instead of feeling that it's an endless pit of information. You could paint links purple manually because I bet the browser starts to lose that data, or just make a global percentage progress bar... I think Wikipedia has some metadata for which project an article belongs to that you could use here as well.<p>Will you open source this?