Hacker News Reader

Comment by Houshalter | Hacker News Reader

A while back I created a toy IRC bot that answers questions like this just by searching reddit and taking the top comment. It works surprisingly well for questions that are likely to have been asked before. And when it does work, the result is much better than a dry wikipedia excerpt.

I then added some simple machine learning to filter search results for the most relevant threads which improved it quite a bit.

View on HN

This is called the "Take the First" heuristic. If you want to learn more about "fast and frugal" heuristics (which work surprisingly well in a lot of cases), read "Simple Heuristics that Make Us Smart". It explains how to get programs to give good answers when time and processing power are limited (similar to your case).

http://www.amazon.com/Simple-Heuristics-That-Make-Smart/dp/0...

Interesting! Care to share some sample question that is likely to be answered by reddit? I tried it manually and can't find answers to any question I can think of right now (looks like I could be failing a Turing test any time soon)

You can try it at https://kiwiirc.com/client/snoonet.org##bottest, if it's online just do "!ask question". Examples:

>!ask what is the largest prime number?

For primes of the form 2^n - 1 (known as [Mersenne primes](http://en.wikipedia.org/wiki/Mersenne_prime)), a very fast primality test known as the [Lucas-Lehmer test](http://en.wikipedia.org/wiki/Lucas%E2%80%93Lehmer_primality_...) is available. The ten largest currently known prime numbers are all Mersenne primes.

>!ask what's the airspeed velocity of an unladen swallow?

African or European?

> !ask tell me a joke

I was eating chicken tonight last night, and got a little bone in one of the fillets. So I said "Fillet? More like fill-it with bones! Haha!" Then I looked around at the empty table and quietly sobbed into my bowl.

That might actually make a new kind of search engine. You should put it on the web?

That would be huge--after all these years a Reddit search engine that works!

It's currently running on a few IRC channels on snoonet. People seem to have lost interest in it though.

As I said, it only works for questions that are likely to have been posted on reddit before. Not a general search engine.

Sounds interesting. Do you mind sharing how did you trained your machine learning algorithm ? It seems like an awfully hard task, since in a sense you are competing with the Google algorithm and the order it offered you.

The reddit search engine isn't very good. It's just logistic regression with a number of features, like the percent of 1, 2, and 3-gram matches, it's score, number of comments, and if it's a selfpost.

Reply on news.ycombinator.com