Results BOW: bag-of-words Entity: entities extracted by semantic parsing NB: naive bayes SVM: support vector machines LP: label propagation LP+Meta-graph: co-training [ Wan et al. SDM 15 Know Sim: unsupervised ensemble of meta-paths [Wang et al. ICDM 16 NB SVM LP Semi lin Ensemble Settings BOW BOW+ BOW BOW+ BOW+ Meta-Know- Full Co SVM EM Datasets Entity Entity Entity path Sim Graph Graph rain 20 NG-SIM39.0248.4637.3449.6754.53577556.8748.94584652.0454446099 20 NG-DIF43.745724395755.7172.4076.1377.1461.3177.6971.36730880.08 GCAT-SIM71.2471.2473.92746470.9771.0560.5979.1481.02687969.9680.97 GCAT-DIF56.6056.6663.5263.9161.9561.3751.6464.326505574858.1966.95 We show our results of five labeled training data for each class. all the numbers are averaged accuracy(in percentage %)over 50 random trialsResults • BOW: bag-of-words • Entity: entities extracted by semantic parsing • NB: naïve Bayes • SVM: support vector machines • LP: label propagation – LP+Meta-graph: co-training [Wan et al., SDM’15] – KnowSim: unsupervised ensemble of meta-paths [Wang et al., ICDM’16] 28 • We show our results of five labeled training data for each class. All the numbers are averaged accuracy (in percentage %) over 50 random trials