224 comments

[ 2.4 ms ] story [ 210 ms ] thread
Are you tired of getting 0 up votes on your post? Wouldn't it be nice to have a tool to test if your title will draw people's attention?

Well, this project doesn't offer this tool, but it tries.

The results for this particular HN post are:

Bad: 0.0598 - Good: 0.9307.

The results for "The results for this particular HN post are:" are:

Bad: 0.9800 - Good: 0.0220

NB. Sorry for that, I just had to do it :)

The results for "The results for "The results for this particular HN post are:" are:" are:

Bad: 0.8083 - Good: 0.1910

Is it turtles all the way down?
That would not be noticeable to HN people apparently

"Scientists discover it is turtles all the way down" Bad: 0.9998 - Good: 0.0002

well, I typed "Court: Suspicionless Searches of Travelers’ Phones and Laptops Unconstitutional"

Bad: 0.9969 - Good: 0.0032

You forgot to mention the time-zone in your analysis.
Yup, that's a big one. A topic where it is very noticeable is when it concerns monopolies. Negative post or comment about monopolies when Americans are the online majority? Expect to get downvoted into oblivion. Posted when Europeans are the online majority? Up you go.

To be clear: this isn't a complaint about downvoting, I'm just pointing out the phenomenon. For example, its possible to go to bed with +7 and wake up to -4 (a rather stark difference) because a certain comment was posted during the European evening, and thus was 'exposed' to American HNers longer than it was to European ones.

Yep seen the same phenomenon. There really is too much meaning packed into that one number.
"Show HN:" gets "Bad: 0.0002 - Good: 0.9998". Interestingly, entering a one-word title will get it stuck on "Bad: 0.4528 - Good: 0.5472".
"Show HN: Computer": Bad: 0.0850 - Good: 0.9141

"Show HN: Internet": Bad: 0.9987 - Good: 0.0013

It would be nice to include an analysis of how good is this with real data. Perhaps pick all the submission from yesterday and show the correlation between the real points and the prediction.

The distribution of points is very 0 heavy. It can be a problem to represent it and to model it.

"The future of emulation in compiler optimization LLVM haskell"

Bad: 0.0004 - Good: 0.9996

"Trump" - Bad: 0.4528 - Good: 0.5472 "Can Trump" - Bad: 0.9862 - Good: 0.0142 "Will Trump" - Bad: 0.9896 - Good: 0.0108 "Show HN: Trump" - Bad: 0.9735 - Good: 0.0225
This comment is going to collect the most votes yet is predicted to be Bad: 0.9320 - Good: 0.0800
I am very torn. I really want to upvote this comment, but am equally wanting you to be proved wrong.
Though a comment is not a post title.
Does this not add up to 1.0 by the way.
I didn’t read the code in the post and don’t have any deep familiarity with machine learning, but I have implemented a naive bayesian classifier to do something similar for tweets. The scores you get from that method don’t add up to 1 either.
This is somewhat interesting. If what is stated in the comment itself is true then that means the poster had to solve to find x and y such that 'This comment is going to collect the most votes yet is predicted to be Bad: x - Good: y' would give Bad: x and Good: y as the output when passed through the neural net. Maybe they also manipulated other parts of the comment to find x and y.
neural networks are fun

the history of X scores:

power: good 0.91

wealth: good 0.99

sex: good 0.99

squirrels: good 0.99

cables: bad 0.97

keys: good 0.99

why cables?

> why cables?

Maybe something to do with the wikileaks leaked cables?

(comment deleted)
More training data required, I think, or else HNers like 4chan memes from yesteryear:

>frosted butts

>Bad: 0.1214 - Good: 0.8659

Interesting, tired with:

I made cheese sandwiches for a week, here's what happened

And got: Bad: 0.0001 - Good: 0.9999

Not sure what to make of that.

wait, what happened? i need to know.
He had lots of cheese sandwiches.
"I made, here's what happened": Bad: 0.0022 - Good: 0.9976
VC licked my balls, here's what happened Bad: 1.0000 - Good: 0.0000
Hahahahahahahahha omg my mahn
Insert "for a week" to get "Bad: 0.0016 - Good: 0.9986"
I've got only 0.9773 good with:

Can you pass the butter?

But exactly the same score with:

Cannot your passed their butter?

Couldn't hill-climb past that.

(comment deleted)
Can a neural net predict Google's stocks if it goes up or down?

t. Google employee knowing what do with my stocks

People have certainly tried. If someone figures out how to do it, they usually don't tell everyone because they will have less of an edge.
"Announcing styled-components v5": 0.9682 Good

"Show HN: Announcing styled-components v5": 0.9973 Bad

So, don't use "Show HN"?

Probably a false positive. Many show hn posts don't receive too many votes.
What are it's results on historical submissions? Presumably a subset was used for training?
Your own title seems to be performing quite badly...

Bad: 0.9917 - Good: 0.0076

Bill gates dead. Bitcoin buy now

Good - 1.0 Bad. - 0.0

What do I win?

This is actually 1.0 bad, apparently. :)

Each word added, "bitcoin", then "buy", then "now", drops the score lower until it hits 1.0 bad.

Implementing an operating system for Risc-v in Rust Bad: 0.0542 - Good: 0.9655
Would have been nice if some keywords were likely to increase or lower the score prediction. For instance, popular things like "rust" would probably increase popularity.

e.g: "Rihanna concert cancelled" Bad: 0.0006 - Good: 0.9992

vs "Rust 1.33 released" Bad: 0.9896 - Good: 0.0108

react-penis-app

Bad: 0.0015 - Good: 0.9986

Not sure what I'm going to build, but it's going to make me a lot of money.

Interesting:

"A mouse killed our network engineer" - Bad: 0.2068 - Good: 0.7895

VS

"A rat killed our network engineer" - Bad: 0.9698 - Good: 0.0322

Perhaps "mouse" is better than "rat" because of computer mice?
ha, didn't think of that, you might actually be right :)
(comment deleted)
"A mouse killed our network engineer" - Bad: 0.2068 - Good: 0.7895

"A network engineer killed our mouse" - Bad: 0.0397 - Good: 0.9529