Show HN: Testing HN titles against a neural network

ha, didn't think of that, you might actually be right :)

cute donkeys : Bad: 0.0002 - Good: 0.9998

fluffy cats : Bad: 0.9921 - Good: 0.0090

Show HN: Can a neural network predict if your HN post title will get up votes?

Bad: 0.0598 - Good: 0.9307

Can an AI predict if your HN post will go viral? The results will shock you

Bad: 0.9539 - Good: 0.0401

Seems to work!

Hmmm.

I tried feeding https://rachelbythebay.com/fun/hrand/ to this. The results were... well, the input was very repetitive and the NN was only trained on 1.5k titles, so what can I say.

Some of the best:

1.0000 0.0001 The police found programming to raise sheep

1.0000 0.0001 The police found PS4 in the cloud

1.0000 0.0001 The future of API in the cloud

1.0000 0.0001 The economics of iPad in the cloud

1.0000 0.0001 The economics of iPad in the cloud

1.0000 0.0001 My framework for Windows is patented

1.0000 0.0001 My framework for SDK is patented

1.0000 0.0001 My framework for Heroku is patented

1.0000 0.0001 I bootstrapped my Heroku in 2 years

1.0000 0.0001 Google kills Y Combinator in space

1.0000 0.0001 Coming soon: bigger blog is patented

Some of the worst:

0.0000 1.0000 Apple has Y Combinator in 2 years

0.0000 1.0000 China creating PS4 at a coffee shop

0.0000 1.0000 China creating PS4 at a coffee shop

0.0000 1.0000 Choose Heroku for developers

0.0000 1.0000 Fixed: NoSQL at a coffee shop

0.0000 1.0000 Followup to the Arduino without warning

0.0000 1.0000 Followup to the marriage without warning

0.0000 1.0000 How I made Android at Stanford

0.0000 1.0000 How I made Obama on the freeway

0.0000 1.0000 How I made bitcoins in the cloud

0.0000 1.0000 How I made iPod at Stanford

0.0000 1.0000 How I made iPod on the freeway

I did this by porting this project to node (which was shockingly easy - took 3 minutes - because there were no JS libraries or frameworks used :D :D :D :D) and running it at the console. 5000 lines of output and repro instructions over at https://gist.github.com/exikyut/1714ad98a136d77d8674944410a4... (the output got pasted first for some reason, sorry)

So if it's bad, it's gonna work better !