Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: Testing HN titles against a neural network

github.com

Post reply on HN
Print view View original

121–130 of 225 posts

Re: Show HN: Testing HN titles against a neural network

#121

Post by progval »

lifthrasiir wrote: post #112
self_awareness wrote: post #83

"Can a neural network predict if your HN post title will get up votes?" Bad: 0.9917 - Good: 0.0076 I'm sorry guys!

But: "Show HN: Can a neural network predict if your HN post title will get up votes?" Bad: 0.0598 - Good: 0.9307 Just in case, I've tested some more titles to make sure that "Show HN" doesn't (EDIT: typo, ugh) just boost any title when prepended.

> "Show HN" just boosts any title when prepended.

"Internet": Bad: 0.4528 - Good: 0.5472

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

Re: Show HN: Testing HN titles against a neural network

#122

Post by JoeAltmaier »

blauditore wrote: post #68

It's basically a buzzword detector. "this is just a tool for detecting buzzwords" => Bad: 0.9991 - Good: 0.0011 "this is merely a device for detecting artificially sophisticated words" => Bad: 0.0019 - Good: 0.9980

So that's a neural network thing? If the input is 'word space' it trains to detect buzzwords?

I'm honestly interested - do we hobble our neural networks with our choice of training space? If for instance, the HN-NN-input space including 'prepositions and word-frequency statistics', would the network train for sentence sophistication? Just because those stats were in front of it?

Re: Show HN: Testing HN titles against a neural network

#123

Post by thaumasiotes »

lm28469 wrote: post #61

I'd argue that titles, as long as they're neutral and accurate, have nothing to do with the upvotes you'll get. This is a perfect example of using data to try to find something that simply doesn't exist or that bear so little weight in comparison to the other variables that you can safely ignore it.

> I'd argue that titles, as long as they're neutral and accurate, have nothing to do with the upvotes you'll get.

This is nonsense. If the title is accurate, then it is closely related to the content, and the content has a lot to do with the upvotes you get.

Re: Show HN: Testing HN titles against a neural network

#125

Post by Scarblac »

ngrilly wrote: post #85

In response to this article, I was expecting insightful comments on the neural network itself and how to improve it, but instead I'm mostly reading funny attempts to play with the scoring :)

Well, step 1 is finding an answer to the question in the title. Can it?

Re: Show HN: Testing HN titles against a neural network

#128

Post by thaumasiotes »

hanoz wrote: post #74

"Stephen Hawking has died": Bad: 0.9727 - Good: 0.0322 https://news.ycombinator.com/item?id=16582136

This is an interesting question of what context titles should be evaluated in.

If your model is "pick a title, and then predict how it will do", with the title being an independent variable, then the overwhelmingly negative assessment is quite correct. Most of the time, Stephen Hawking (or any other celebrity) hasn't just died, and that title would be a lie or a hoax.

To predict these obituary articles better, you'd need to build in an assumption that the content of the title was true, which would cause a lot of problems in the general case.

Re: Show HN: Testing HN titles against a neural network

#129

Post by vectorEQ »

this neural network is politically biased and possibly racist. it also prefers swedish girls over swedish boys. I do have to agree though, that cheese is better than mouldy cheese. and the fact "comacho for president for ever" nets a score of "Bad: 0.0415 - Good: 0.9614" which i think is fair reflection of voter behaviours and general consensus among humans. 'man' and 'not man' its not so fussy about, 'woman' gets same score as 'man', but 'not woman', oh thats just bad (Bad: 0.9909 - Good: 0.0097)

all in all a great tool to analyse your titles, it will surely help this community grow and mature over time, finally. Thanks a lot for creating this.

Re: Show HN: Testing HN titles against a neural network

#130

Post by kdmccormick »

thaumasiotes wrote: post #107
bencollier49 wrote: post #90

Earlier quoted context omitted.

sohpisticated ?

I think whether the title is spelled correctly is a valid thing to take into consideration when predicting whether it will get upvotes. Put another way, I think misspelled titles are well within the problem domain here.

Is it within the problem domain? Yes.

Is it interesting for this project? Based on my skimming of the README, probably not. I don't think it handles misspellings intelligently.

Post reply on HN