LIBSVM Data: Multi-label Classification

Recently multi-label classification has been an important topic. Currently there are very few publicly available data sets. We tried hard to collect the following sets. Labels are in the beginning of each line and separated by commas.

• Source: [GT08a]
• Preprocessing: The original set was at Mulan: A Java Library for Multi-Label Learning. Here we slightly modify the sparse form at The Extreme Classification Repository so feature indices start from 1 instead of 0.
• # of classes: 159
• # of data: 7,395
• # of features: 1,836
• Files:
  • bibtex.bz2

• Source: [GT08a]
• Preprocessing: The original set was at Mulan: A Java Library for Multi-Label Learning. Here we slightly modify the sparse form at The Extreme Classification Repository so feature indices start from 1 instead of 0.
• # of classes: 983
• # of data: 16,105
• # of features: 500
• Files:
  • delicious.bz2

• Source: Mediamill / The Mediamill Challenge Problem
• Preprocessing: We combine all binary classification problems into a multi-label one.
• # of classes: 101
• # of data: 30,993 / 12,914 (testing)
• # of features: 120
• Files:
  • train-exp1.svm.bz2
  • test-exp1.svm.bz2 (testing)

• Source: [DL04b]
• # of classes: 101
• # of data: 3,000 / 3,000 (testing)
• # of features: 47,236
• Files:
  • rcv1subset_topics_train_1.svm.bz2
  • rcv1subset_topics_train_2.svm.bz2
  • rcv1subset_topics_train_3.svm.bz2
  • rcv1subset_topics_train_4.svm.bz2
  • rcv1subset_topics_train_5.svm.bz2
  • rcv1subset_topics_test_1.svm.bz2 (testing)
  • rcv1subset_topics_test_2.svm.bz2 (testing)
  • rcv1subset_topics_test_3.svm.bz2 (testing)
  • rcv1subset_topics_test_4.svm.bz2 (testing)
  • rcv1subset_topics_test_5.svm.bz2 (testing)

• Source: [DL04b]
• Preprocessing: The four testing sets correspond to the four testing files from the RCV1 site (appendix B.13). In the testing set, the number of classes is 103. Tokenized texts are from B.12 of RCV1 site
• # of classes: 101
• # of data: 23,149 / 781,265 (testing)
• # of features: 47,236
• Files:
  • rcv1_topics_train.svm.bz2
  • rcv1_topics_test_0.svm.bz2 (testing)
  • rcv1_topics_test_1.svm.bz2 (testing)
  • rcv1_topics_test_2.svm.bz2 (testing)
  • rcv1_topics_test_3.svm.bz2 (testing)
  • train_labels.txt.bz2 (original labels)
  • train_labels.txt.bz2 (raw tokenized texts)
  • train_labels.txt.bz2 (original labels (testing))
  • test_labels.txt.bz2 (raw tokenized texts (testing))

• Source: [DL04b]
• Preprocessing: The four testing sets correspond to the four testing files from the RCV1 site. In the testing set, the number of classes is 350.
• # of classes: 313
• # of data: 23,149 / 781,265 (testing)
• # of features: 47,236
• Files:
  • rcv1_industries_train.svm.bz2
  • rcv1_industries_test_0.svm.bz2 (testing)
  • rcv1_industries_test_1.svm.bz2 (testing)
  • rcv1_industries_test_2.svm.bz2 (testing)
  • rcv1_industries_test_3.svm.bz2 (testing)

• Source: [DL04b]
• Preprocessing: The four testing sets correspond to the four testing files from the RCV1 site. In the testing set, the number of classes is 296.
• # of classes: 228
• # of data: 23,149 / 781,265 (testing)
• # of features: 47,236
• Files:
  • rcv1_regions_train.svm.bz2
  • rcv1_regions_test_0.svm.bz2 (testing)
  • rcv1_regions_test_1.svm.bz2 (testing)
  • rcv1_regions_test_2.svm.bz2 (testing)
  • rcv1_regions_test_3.svm.bz2 (testing)

• Source: [MB04a]
• # of classes: 6
• # of data: 1,211 / 1,196 (testing)
• # of features: 294
• Files:
  • scene_train.bz2
  • scene_test.bz2 (testing)

• Source: SIAM Text Mining Competition 2007 / SIAM Text Mining Competition 2007
• Preprocessing: We remove "." before transforming data to vectors. We use binary term frequencies and normalize each instance to unit length.
• # of classes: 22
• # of data: 21,519 / 7,077 (testing)
• # of features: 30,438
• Files:
  • tmc2007_train.svm.bz2
  • tmc2007_test.svm.bz2 (testing)

• Source: [AE02a]
• # of classes: 14
• # of data: 1,500 / 917 (testing)
• # of features: 103
• Files:
  • yeast_train.svm.bz2
  • yeast_test.svm.bz2 (testing)