Live data from Hacker News

A dataset of every Reddit comment

news.ycombinator.com

71–80 of 96 posts

Re: A dataset of every Reddit comment

#71

I used this[0] query to find the top ten[1] most downvoted comments of all time on Reddit. The most downvoted comment[2] is ironically in iAMA, by a mod of iAMA (ironic because of the recent drama). I'd find the top ten most upvoted, but I ran out of free bandwidth on BigQuery :(. [0] https://gist.github.com/alexggordon/7b56353dcf8044a7a5f9 [1] https://drive.google.com/file/d/0Bzxo-UKxFmN-eWticy1BR2tCRDQ... [2] https…

Here's the top ten most upvoted using that query with desc instead of asc:

http://pastebin.com/Zvg6mdjZ

Re: A dataset of every Reddit comment

#73
I wanted the data in a form I could easily query, so I wrote a program to convert the JSON to an SQLite database: https://gist.github.com/ers35/3b615a75fa0ed5e6d5cc

I have the program running on Amazon EC2 right now converting the whole dataset. I plan to upload the database to the Internet Archive when it completes.

Re: A dataset of every Reddit comment

#74

Earlier quoted context omitted.

According to Reddit's User Agreement user content: "You retain the rights to your copyrighted content or information that you submit to reddit ('user content') except as described below." And the exceptions just state that Reddit has a perpetual irrevocable worldwide license. So it seems like there's no default license and others don't have any automatic rights to use the content. Does this assessment seem correct? I…

Good find, I think I got reddit confused with stackoverflow or wikipedia. If there's no clear assignment of copyright to api users, I would imagine that would be problematic for 3rd party app makers.

Isn't this true only if the 3rd party apps are saving the data and/or redistributing it? Apps should be able to request data from reddit just as a normal user would through their browser, since reddit has permission to use (and serve) that content.

Re: A dataset of every Reddit comment

#75
post #9

Anyone know how big it is yet?

Dataset on BigQuery is 265GB.

I wasn't able to get access to it on BigQuery .. the progress bar just sat and sat .. can you confirm that its usable on BigQuery - I'd sure love to play with this data.

Re: A dataset of every Reddit comment

#76

Earlier quoted context omitted.

If it's open source I'll love to take a look

I pastebined the code. It's terrible. I should rewrite it. But here it is: http://pastebin.com/CM9u17jq

Nice, I see you use neural networks, can you explain a bit how you are training them?

Re: A dataset of every Reddit comment

#77
post #76

Earlier quoted context omitted.

I pastebined the code. It's terrible. I should rewrite it. But here it is: http://pastebin.com/CM9u17jq

Nice, I see you use neural networks, can you explain a bit how you are training them?

It searches reddit for whatever the user queries. Reddit returns up to 100 threads. I then choose a thread and take the top comment.

The neural network predicts which thread is most likely to produce a satisfying answer. The main features are number of n-gram matches with the question, the score, the number of comments, and some other metadata.

It's far from optimal but it does improve it a bit.

Re: A dataset of every Reddit comment

#78

I think it's fascinating that if anyone uses this dataset to train an AI, some tiny piece of my personality might make a contribution.

It's going in my /raid/datasets with about 1.4TB of other comment data. So the answer is going to be "yes" ... well probably by Tuesday morning it will be.

I've used older reddit dumps before ... it's not great data for most things, but there's an awful lot of it!

Re: A dataset of every Reddit comment

#80
post #73

I wanted the data in a form I could easily query, so I wrote a program to convert the JSON to an SQLite database: https://gist.github.com/ers35/3b615a75fa0ed5e6d5cc I have the program running on Amazon EC2 right now converting the whole dataset. I plan to upload the database to the Internet Archive when it completes.

Yes, please do that!
Post reply on HN