Hacker News new | past | comments | ask | show | jobs | submit login
Reverse Engineering the HN Ranking Algorithm (sangaline.com)
70 points by foob on March 10, 2017 | hide | past | favorite | 7 comments

Welp. I'm glad I managed to catch this in whatever window of time it was visible on the site. Great analysis, reasonably readable.

While I might disagree with the methods used to suppress it, I understand why the powers that be might not want this sort of article to be widely spread, but what can you say. Thanks OP.

Very cool!

And the IPython notebook is direct available here: https://github.com/sangaline/reverse-engineering-the-hacker-...

Unironically, this was apparently censored (no realistic human interaction would result in this post, which was gaining popularity quickly, disappearing from the first 6 pages within moments).

Meanwhile, a post about YCombinator itself (the US company) going to Canada to skirt US visa regulation changes so that it can continue its access to foreign founders whose exploitable economics are desirable to US founders... remains on the first page.

This, IMHO.

This article was excellent, btw, despite my lack of basic calculus knowledge and, thus, my inability to understand the notated equations.

This article was flagged by users for some reason. Moderators never saw it (edit: wrong, see below). If we had, we would have turned off the flags.

The post about YC and visas was untouched by moderators except that we eventually downweighted it because it had spent such a long time on the front page.

So what these two examples illustrate is not what we do, but how easily people can find examples to fit any belief. We're always happy to answer questions and correct mistakes though.

Edit: ok, I think I know what happened now. It's not true that a moderator never saw the post: a moderator put 2010 on the title by mistake. I didn't think to look for title changes when I checked the logs. Later, a user pointed out the mistake and we corrected it.

In this case although we didn't penalize the post, it's possible that the inaccurate '2010' annotation acted as a penalty without us intending it that way. Obviously when we add years to submission titles, which we do every day, the intention is to get it right, not wrong.

My comment above has just received a down-vote, despite being 2 days old, well outside of the down vote time limit (unless there is a comment flag feature without a time limit that I'm unaware of), at the same time @dang left a reply.

So, another example of moderator tampering...? Or was there a recent up-vote that was removed?

I guess we are all just conspiracy theorists.

(re: your too little, too late edit)

How a public apology for implying that we are conspiracy theorists[1], since you are the one with the ability to "fact check" us, and you failed to do so, most notably, and conveniently, after it was too late. Your attempt to debunk us, combined with lowering the weight of today's article, effectively censored the entire conversation.

While we're at it, how did somebody accidentally add a 2010 tag to a clearly brand-new article that was just published on the day of the post? Here's an actual conspiracy theory: this was the first attempt to censor the article, which failed when people immediately recognized that it was not from 2010. It's one thing to originally post something with an incorrect date (e.g., a typo, etc.), but this was a moderator adding a date. That is a deliberate action, which means more than just a passing glance would have been given.

1. I have arrived at this conclusion by logical deduction: A) Confirmation bias is the cognitive bias being described by your quote, "...how easily people can find examples that fit any belief...". B) One does not need to confirm something other than a theory. C) There are many moderators, therefore making it a conspiracy, if it was deliberately done. D) It would therefore need to be a conspiracy theory in order for cognitive bias to be relevant.

Mods: please remove the 2010 in the title. The publication date is today and the post discusses data from 2017.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact