Twitter on Drugs: Pharmaceutical Spam in Tweets

Twitter presents a new forum for spammers to facilitate illegal pharmaceutical scams. We present a classification scheme using decision strategy and data mining techniques taking into account the unbalanced nature of the data set. Four classifiers are used to identify pharmaceutical spam tweets. Classifiers J48 and Random Tree (RT) are generated by Weka tools, and classifiers DL(J48) and DL(RT) are based on the combination of J48 and RT with the decision matrix. The classifiers were tested using manually labeled data sets collected at different time spans. Experimental results suggest that the combination of RT with the decision matrix provides a stable performance improvement over using standalone tree-based classifiers. Keywords-data mining; text mining; spam; pharmaceuticals; social networking; Twitter;