Predicting Popularity of Posts on Hacker News (2016) [pdf]
cs229.stanford.edu
cs229.stanford.edu
I would like to believe this is true, and I would guess that it's probably more true on HN than elsewhere, but I would really like to see them dive more into how they came to this conclusion.
I think the moderator led title editing makes this comparison more difficult (not bad necessarily) but it would be worth comparing popularity:content to popularity:title to have a more rigorous understanding.
Overall I applaud the approach they took. It seems however that the conclusion/output doesn't give better insight into the parameters which drive post popularity on HN, over previous approaches which seek to understand the underlying voting mechanism [1]. While different they are trying to basically understand the question "what makes for a popular post on HN."
Modeling the sentiment of a community under a formal mechanism (vote algo) is incredibly difficult to do correctly, so again kudos for the work and it seems like it's something that could be built on.
[1]http://sangaline.com/post/reverse-engineering-the-hacker-new...
SELECTED TOPICS EXTRACTED BY NMF
Topic 1 js javascript node react component python config browser
Topic 2 pple iphone ipad fbi mac cook sir headphone watch steve
Topic 3 invest startup fund venture capital entrepreneur tech raise
Topic 4 game pokemon vr nintendo virtual oculus headset steam
Topic 5 email clinton website inbox gmail slack spam campaign
Topic 7 tesla vehicle musk autopilot elect autonomous battery road
Topic 8 facebook twitter advertise zuckerberg messenger snapchat
Topic 11 docker deploy cloud kubernet swarm cluster microservices
Topic 12 ai intelligence algorithm neurual deep alphago deepmind
Topic 14 uber ride city lyft tax vehicle hail self did transport kalanick