I am trying to create a naive bayes spam filter for social networks and i need a spam dataset (containing posts/tweets etc.) of popular social networks like twitter/facebook. Is there any publicly available dataset which i can use to train the classifier?
From the Machine Learning Repository at UCI there are some datasets that may get you started.
For blogs marked as spam, check out this site.
Twitter data are not allowed to be publically shared. You can share the analysis or the visualization, but not the raw data of tweets. I imagine Facebook has a similar policy.
But, you can collect more than you ever dreamed of by using the public live stream. Here is an example python code to collect many, many tweets as they are sent. There should be plenty of spam if you let it run for a couple minutes.
More than a decade ago, when I worked for an ISP, I subscribed to the spamtools mailing list. Unfortunately, their archives link doesn't work, so I have no idea if they're still active or not.
They also were more focused on e-mail spam, but at the time social network spam didn't really exist (as social networks didn't really exist in the state they do today).
I would suspect that if someone had a training set, it'd be someone on that list. I'd also assume that twitter spam would be significantly different from blog or e-mail spam, due to the character limitations.