Why it took a long time to build that tiny link preview on Wikipedia | Hacker News Reader

Why it took a long time to build that tiny link preview on Wikipedia

diff.wikimedia.org

48 pointsgodelmachine27 comments

This is one of the most useful additions to WP in recent years. If you run into a word/phrase you don't know or are fuzzy on, it encourages satisfying your curiousity, saves having to load that page.

In the past few months I've also noticed a lot of innovation happening in the maps department. E.g. Ran into a great interactive Mars map the other day.

Reading this finally made me figure out how to turn this feature off (to be fair, it was quite straightforward). I find it extremely annoying when there are always things popping up over the text you are trying to read.

Do you move your mouse cursor along the text you're reading? Or do you move your cursor out of the way and often end up hovering a link?

I (not OP) highlight text as I'm reading it. Sometimes paragraph at a time, sometimes with frantic mouse movements and clicking highlighting a few words or a sentence at a time.

I will admit it's very weird to watch, but it helps my ADHD concentrate.

I just leave the cursor wherever it happens to be. With link-dense pages, as Wikipedia tends to have, that very often causes a fairly large box to pop up.

Also, when scanning a page looking for info and links, I seem to both read and keep the mouse pointer somewhere close to the center of the window, increasing the risk of annoying pop-ups. (Moving the mouse too far away often causes the wrong thing to scroll whey you use the wheel, perhaps not on Wikipedia but on many other sites.)

>We had to choose a thumbnail

All examples I tried just take the main image from the article (if it exists).

>We had to generate a summary

Similarly, all I see is, literally, the first sentence of the linked article with some extremely trivial and minor adaptations.

>Design design design

It is literally a photo and a paragraph of text in the default sans-serif font.

I know it's PR but damn.

Nope, this article isn't exaggerating about these being difficult decisions. I watched it unfold over the years -- because wikipedia is somewhat of a democracy, the amount of bikeshedding behind the scenes over the years was insane. Every little detail of these preview boxes had to be perfected or else someone would come out with objections. It is actually a miracle that this feature was able to be shipped at all.

I'd bet if there was a top-down leadership hierarchy from day 1 of opening the website, rolling out this feature would take less than a month.

But Wikipedia would be Microsoft Encarta if it had top-down leadership from day 1.

Wow, even after your comment I didn't expect the degree of contention. One user fears that this will give a new excuse to "North American Exceptionalists" to use different images on North American Ice Hockey pages compared to Commonwealth ones. This person's signature is "abolish rule #2 like slavery", where rule #2 is

> All North American hockey pages should have names without diacritics, except where their use is likewise customary

I can't imagine taking a job where my work was subject to review by this sort of person.

Welcome to humanity. These people exist around us every day :D

In the best case scenario for top-down Wikipedia, it would have been Google Knol. That was where they were going when it was nominally a startup founded by a guy whose previous venture had been an online men’s magazine. But then the community revolted.

We all know what a successful startup looks like. But I feel that the internet lacks criteria for a community like Wikipedia. Or for that matter, the chan sites. The tech can be almost ludicrously out of date, the design can be bland, the programmers sub-par, changes roll out over years rather than days, and yet it can have an unstoppable and highly influential community.

Wikipedia should be data driven.

Let the bikeshedders all try out their design in AB tests, and pick whichever the user metrics say is best.

You're going to get much better results that way than spending 4 years writing specifications and having meetings.

> Let the bikeshedders all try out their design in AB tests, and pick whichever the user metrics say is best.

This is essentially exactly what happened. But after this was tried, contributors disagreed on which of the metrics are important. Is it more important that a Wikipedia user stays on the site to discover more tangential information, or is it more important that they leave the site as quickly as possible after finding the relevant information? User retention doesn't matter because it's Wikipedia, but which philosophy provides more social utility?

How do you measure "best"?

Highest likelyhood of returning to Wikipedia within 30 days, not including the day they saw the info card.

If people find a site useful, they will voluntarily return again later. That's a very hard metric to game/abuse. Do not count a user returning from a promo email or browser notification or any of those dark patterns.

>extremely trivial and minor adaptations

The article goes into why this wasn't as simple as it seems.

Edges represent the bulk of work/stress in most projects, and there is no reason this should differ.

The article doesn't mention Navigation Popups[1], a Wikipedia extension that you can activate from you profile that does basically the same thing as the new link previews.

I've been using it since 2007, and to me it was always a mystery why it wasn't more discoverable.

[1] https://en.wikipedia.org/wiki/Wikipedia:Tools/Navigation_pop...

"We realised HTML was very important"

[blank stare]

This sentence makes a lot more sense in context. They were using a text-only preview, but it didn't work well for a lot of pages.

Yep, I feel a lot of us would have tried to use text or some kind of simple Markdown flavor for simplicity! And that isn’t sufficient.

https://www.mediawiki.org/wiki/Extension:PageImages

They developed an algorithm to choose the best fit image which would appear in the page preview

Wikipedia developed something many though was impossible by assembling thousands (millions?) of volunteers.

Why did the solution here need to be algorithmic? Couldn't they scale useful summaries in the same way they scaled the actual article content?

You have to actually get those contributors to show up, which is an issue because wikipedia has been slowly bleeding active contributors for years.

Perhaps a great way to get them to show up, would be to give them the opportunity to be the first to write a summary and select/upload a picture for an article in their field? :)

Wholeheartedly agree.

Every now and then I will naively take on some kind of difficult programming challenge requiring lots of hacks and out of the box thinking. These can be quite fun projects, and very rewarding if you make it to the end. But sometimes, as you work your way through the project, and the number of exceptions and edge cases grows, you have to step back and ask yourself is it really worth it? Sunk cost fallacy usually kicks in and you hate yourself for a couple of days, but ultimately you realise it was the right decision.

"They were too occupied thinking about how they could do it, that nobody stopped to ask whether they should."

I don't know what drives Wikipedia contributors to do what they do, but I imagine there would be a large number of them who would want to go out and be the first to write a summary and select/upload a picture for as many articles as they possibly could. If the motivation simply isn't there - gamify it somehow.

Yes, use a default basic implementation that just grabs the main image and first paragraph with some basic stripping. Then allow authors to later modify the card content specifically. I'm sure it would take a page editor all of 2 minutes to do something like that. This really seemed like perfect is the enemy of good.

I also get the feeling that wikipedia is afraid to stand up to the mob of editors.

Ed: title should be (2018)

Reply on news.ycombinator.com