Hacker News Reader

Comment by jerjerjer | Hacker News Reader

> In this case, I included the post title, author, date, and content. All of those factors could be relevant to the chance a story gets voted up.

> Even if the model gets extremely good at predicting final_score_if_it_hits_front_page, there’s still the inherent randomness of probability_of_hitting_front_page that is fundamentally unpredictable.

In addition to date, you might want to include three fields:

- day of week (categorical)

- is weekend/holiday (boolean)

- hour or time of the day (categorical, you can have 24 of them or morning/afternoon/etc.).

The probability of a post hitting the front page is usually affected by these things so it can really help the model.

View on HN

I find that the best stories get posted by folks in EU time zones as well as the weekend (more of hacker ethos). The flame bait startup drama is M-F Pacific.

I haven't run the data, but anecdotally I can tell you that those things probably don't affect hitting the front page. They do affect the total score, but that is not what is being optimized here.

It's counterintuitive, but if you post at a really popular time, you're competing with a lot of other submissions. If you post at a really slow time, you'll get fewer votes, but it will take fewer to reach the front page and you'll have less competition.

In the end, it kinda evens out. The number of votes it takes to get to the front page and the number of competing submissions are both correlated to your fields above.

I think that this assumes a uniform distribution of "interestingness" in the competing posts across all of those dimensions and I wouldn't be surprised if that isn't the case

It may not be even, but I don't think interestingness is correlated with time of day. But I could be wrong!

Interestingness is subjective, and I would imagine different timezone people have different preferences. Interesting thing to ponder for a bit

Popular time for voting vs posting are not the same

> is weekend/holiday

Somehow this reminded me of someone datamining spiegel.de (german news site) and using the timestamps of the posted articles to extrapolate the writers religion (holidays) and relationships (shared vacations) among dozens of other data points from several years of publicly available data. I think no AI was involved back then.

For anyone interested, it was this CCC talk by David Kriesel (sadly german only).

https://media.ccc.de/v/33c3-7912-spiegelmining_reverse_engin...

There is an english translated audio track, actually. (sound quality is not fantastic though)

I wonder if hour of day would benefit from being combined with HN's visitors location data to be truly relevant? I think the location is embedded in the time somehow if the visitors' origins are stable over time. If 9am PT is a popular time and most of the visitors are on the PT timezone then even if this 9am PT is encoded as UTC the model will pick it up (I think). Now, if over time visitors get more diverse and a big chunk is now coming from Europe, this original 9am will make less sense to the model. Adding visitors origin stats at time of the post would probably even help surface region trends. But I guess this historical data isn't public.

Yep that makes sense. Would be interesting to do a follow-up that explicitly includes these variables and see if it meaningfully improves the results.

The data is massively interconnected too: if Apple releases new m chip, people flood here to see if there's a thread on it, while browsing they may be more or less likely to see other threads given that first case.

I would replace author with a boolean of if the author's account is new or not (the green marker that HN has for new users' posts and comments).

> might want to include three fields:

This has been studied multiple times on HN posts, most seem to have link-rotted. Web Archive them if looking for insights - https://hn.algolia.com/?q=best+time+to+post

Reply on news.ycombinator.com