Then we applied the same training and testing data sets as above to an advanced SVM variation named Max-Margin Multi-Label classifier. M3L is a state-of-the-art multi-label classifier [59]. Different from the one-versus-all heuristic, which assumes label independence, this classifier takes label correlation into consideration. We used the executable file of this algorithm provided by the authors [60]. The performance is better than the simplistic one-versus-all SVM classifier, but still not as good as the Naive Bayes classifier. Table 4 and Fig. 3 show the evaluation measures using M3L. Because SVM is not a probabilistic model, so Table 4 does not have probability threshold values as Table 2 does.