I vaguely remember several years back Reddit added the option for users to allow their subreddit and votes data to be used for research purposes with the hope of building a recommendation engine similar to this. Does anyone know if anything came from that? It would be great if the dataset was publicly available.
The option is opt-in (which is good, the fickle reddit community would revolt otherwise) which means almost nobody uses it. If Reddit would remind their users frequently (e.g. at the end of popular Reddit Blog posts as an aside) or reward people for enabling the option (free Reddit Gold for a week, etc.) I'm sure many more people would sign up.
I'm not so sure about that. I believe the problem has been solved, the solution just isn't widely known yet. I read a paper on arxiv probably a year ago that describes a method that seems pretty straightforward and secure, but I've never seen anything about it since. It involved, essentially, throwing out any records which could actually contribute to a change in a statistical measure. You basically end up finding what aspects of the data are actually identifiable, and throw out any records that contain that. It's guaranteed not to screw up your observations because, by definition, if something is statistically significant it has to show up often enough that it CANT be used to single out a source.
Given the dismal history of anonymization, a paper on arvix is roughly up there with a blogger saying 'I've proven p!=np'...
> It's guaranteed not to screw up your observations because, by definition, if something is statistically significant it has to show up often enough that it CANT be used to single out a source.
What's 'statistically significant' here? The usual p<0.05 convention? You realize that there can be multiple measurements or pieces of data all of which individually have p>0.05 but together have p<<0.05... Information leakage should be measured in bits, not p-values.
(This kind of aggregation is one of the benefits of approaches like meta-analysis.)
I didn't look very far into that, but I think that the issue was that subreddit data wasn't included in the vote dump they did. I actually only heard that secondhand - I could be mistaken.
I just used post and comment histories, which suited my purposes fairly well because the larger project was looking into how memes spread.
Comments
I vaguely remember several years back Reddit added the option for users to allow their subreddit and votes data to be used for research purposes with the hope of building a recommendation engine similar to this. Does anyone know if anything came from that? It would be great if the dataset was publicly available.
Edit: Here are the original threads, I don't think the project got very far. http://www.reddit.com/r/announcements/comments/ddz0s/reddit_...
http://www.reddit.com/r/redditdev/comments/dtg4j/want_to_hel...
The option is opt-in (which is good, the fickle reddit community would revolt otherwise) which means almost nobody uses it. If Reddit would remind their users frequently (e.g. at the end of popular Reddit Blog posts as an aside) or reward people for enabling the option (free Reddit Gold for a week, etc.) I'm sure many more people would sign up.
EDIT: (Sorry for all the parentheticals.)
Why not just anonymize the user data part?
There are startups selling health data this way, I don't think it would be so bad for subreddit subscription data.
I don't think Reddit users would go for that. Also, it's surprisingly difficult to anonymize data effectively without removing nearly all of it.
I'm not so sure about that. I believe the problem has been solved, the solution just isn't widely known yet. I read a paper on arxiv probably a year ago that describes a method that seems pretty straightforward and secure, but I've never seen anything about it since. It involved, essentially, throwing out any records which could actually contribute to a change in a statistical measure. You basically end up finding what aspects of the data are actually identifiable, and throw out any records that contain that. It's guaranteed not to screw up your observations because, by definition, if something is statistically significant it has to show up often enough that it CANT be used to single out a source.
Given the dismal history of anonymization, a paper on arvix is roughly up there with a blogger saying 'I've proven p!=np'...
> It's guaranteed not to screw up your observations because, by definition, if something is statistically significant it has to show up often enough that it CANT be used to single out a source.
What's 'statistically significant' here? The usual p<0.05 convention? You realize that there can be multiple measurements or pieces of data all of which individually have p>0.05 but together have p<<0.05... Information leakage should be measured in bits, not p-values.
(This kind of aggregation is one of the benefits of approaches like meta-analysis.)
I didn't look very far into that, but I think that the issue was that subreddit data wasn't included in the vote dump they did. I actually only heard that secondhand - I could be mistaken.
I just used post and comment histories, which suited my purposes fairly well because the larger project was looking into how memes spread.
Took me a while to find it:
http://www.reddit.com/r/redditdev/comments/dtg4j/want_to_hel...
Here's another version:
https://archive.org/details/2010-reddit-research