Abstract—Text classification is one of the important problems being solved in information retrieval. However,traditional single-label classifiers are no longer sufficient and multi-label approaches are becoming more relevant. There have been a lot of proposals for multi-label learners and in our work, we tried applying ensemble techniques, which have proven to be effective in solving other multi-label classification problems, to combine them. We implemented seven ensemble techniques presented in previous works and evaluated their performance. We have found that some of the ensemble classifiers out perform all of the individual classifiers, namely mean and top 3 techniques. We have also found Calibrated Label Ranking to be a very useful multi-label learner for text classification with a small amount of labels. Ensemble techniques have thus proven themselves to be applicable and beneficial to the domain of text classification.

Index Terms—Text classification, multi-label classification,ensemble techniques

The authors are from Shanghai Jiao Tong University, 800 Dong chuanRoad, 200240 Shanghai(e-mail: boros@sjtu.edu.cn)

Cite: Martin Boroš, Franky and Jiří Maršík, "Multi-Label Text Classification via Ensemble Techniques," International Journal of Computer and Communication Engineering vol. 1, no. 1, pp. 62-65, 2012.