3 ms·
Besides the concept of "best" being "found using math"...I don't really get even the metric here. Posts that have been submitted over the years and have gotten
by danso 11y ago
Besides the concept of "best" being "found using math"...I don't really get even the metric here. Posts that have been submitted over the years and have gotten a lot of points? Why is that by any standard a more reasonably explainable metric than just posts which have garnered the most points, period, barring things like obits (e.g. Steve Jobs)?
This is harder to compute, but I'd argue that the best posts are the ones that garner a shitton of upvotes while also not being overly political ("Drop Dropbox") or tied to a current news event ("G is for Google") or originates from or otherwise involves a major proper noun (e.g. "Why I Hated Working At Google/Facebook/Amazon").
Among recent hits, "Make a programmable mirror" comes to mind [1], as does "I'm learning to code by building 180 websites in 180 days. Today is day 115" [2]. And hell, much of what's been submitted from idlewords.com.
I don't get how "timelessness" is measured. The Brutal Ageism of Tech [3], for example, is from 2014...being popular for 2 years is not necessarily timeless.
This is not to say any of the things on the list aren't great. But the word "best" means something relatively absolute. I didn't expect to see a list of the unquestionable "best"...but I did expect to see a bit more math. If we're talking about HN specifically, as in, things that impacted the HN community...I would expect an assessment to have some heuristic scoring of the discussions generated, including number of upvotes, number of unique participants, and number of longtime participants. Or even a metric based on unusualness of participants...for example, a story and discussion that elicits a "We're investigating this now" from Matt Cutts [4] should have some weight (though obviously, measuring the external impact of a HN discussion, such as the ban on Rapgenius, is outside the purview of an automated classifier)
[1] https://news.ycombinator.com/item?id=10204018 https://news.ycombinator.com/item?id=10204018
[2] https://news.ycombinator.com/item?id=6097155 https://news.ycombinator.com/item?id=6097155
[3] https://newrepublic.com/article/117088/silicons-valleys-brutal-ageism https://newrepublic.com/article/117088/silicons-valleys-brut...
[4] https://news.ycombinator.com/item?id=6956658 https://news.ycombinator.com/item?id=6956658
- minimaxir 11y agoIf you had a heuristic to more accurately rank links than the current time/score based algorithms, you'd have a million dollar startup idea. Which is something that wouldn't be revealed in a blog post. Or OP's random daily email newsletter.
- danso 11y agoWell, I like to think I have lots of million dollar ideas ;) However, in this case, a retrospective algorithm doesn't have to compete in the same category as the one that ranks and applies gravity to the submissions. Ignoring the effect of flags and manual moderation, the ranking algorithm has to work in near real time -- and it seems to, as occasionally a single upvote will send a submission straight up the front page depending on its current gravity -- while not causing the server to shit the bed. A retrospective algorithm doesn't have to be that efficient, and can drastically reduce computation by filtering for past submissions that have received a cutoff of top votes. Running various aggregations/analysis on discussion can likewise be done in relative peace. Furthermore, you have the ability to provide external heuristics...for example, filtering for submissions that were first submitted to HN before getting strong notice on Twitter or Reddit (which can be done by comparing URL submissions and timestamps). This is not to say that coming up with the "best" or even "better" is trivial. It's just that the computational factors are significantly different between the algorithm that has to do real-time ranking and one that performs a historical analysis. That said, it'd be an interesting challenge to create a relatively simple bot that auto-submitted popular links in the same way that the popular Facebook "On this day" plugin works...the heuristic could be very simple: submit only URLs that haven't made the front page within the last 1.5 years and garnered 200+ upvotes...and then some filtering to avoid submitting obituaries or other current-event type posts. If the HN mods give approval to run a spam bot, I'll give it a try :)