Hacker News new | past | comments | ask | show | jobs | submit login

Using Naive Bayes for text classification problems such as spam detection is fine for teaching purposes but:

-it doesn't understand context

-if the spammer uses related words, good luck catching those

-false positives can be quite high

As a recruiter receiving such a wonderful automated reply from you, I would ask you why didn't you use a more suited technique to the problem such as transformer-based attention network, or at least deep learning neural network.

If you got his automatic reply, you wouldn't know what architecture he used since he's not specifying in the email

If he fixes the polish bug and grabbed more training emails from friends it would be even better than 95%. Which is why Bayes works well enough http://paulgraham.com/better.html

It's pretty clear he wanted something cheap and fast to implement and the headaches of getting a neutral net working aren't part of that formula. Honestly seeing how he cobbled together a lot of separate parts into something that works is pretty damn impressive and is exactly the type of thing that would fast track him if I was interviewing him

Not even Naive Bayes would be needed if recruiters would use a "more suited technique" to predict exactly who is going to be interested in a job and only bother emailing those.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact