Scam Detection in Twitter

Twitter is one among the fastest growing social networking services. This growth has led to an increase in Twitter scams (e.g., intentional deception). There is relatively little effort in identifying scams in Twitter. In this paper, we propose a semisupervised Twitter scam detector based on a small labeled data. The scam detector combines selflearning and clustering analysis. A suffix tree data structure is used. Model building based on Akaike and Bayes Information Criteria is investigated and combined with the classification step. Our experiments show that 87% accuracy is achievable with only 9 labeled samples and 4000 unlabeled samples, among other results.