To prepare rcv1 multi-label sets, run the stript

% gen.sh

It does the following steps:

1. Prepare the following two files

rcv1.topics.txt (from rcv1 web site; "B.1. On-Line Appendix 1")
rcv1-v2.topics.qrels (from rcv1 web site; "B.8. On-Line Appendix 8,")
rcv1.industries.txt (from rcv1 web site; "B.4. On-Line Appendix 4")
rcv1-v2.industries.qrels (from rcv1 web site; "B.9. On-Line Appendix 9,")
rcv1.regions.txt (from rcv1 web site; "B.6. On-Line Appendix 6")
rcv1-v2.regions.qrels (from rcv1 web site; "B.10. On-Line Appendix 10,")

and run 

./gen_id_label.py rcv1.topics.txt rcv1-v2.topics.qrels 
./gen_id_label.py rcv1.industries.txt rcv1-v2.industries.qrels 
./gen_id_label.py rcv1.regions.txt rcv1-v2.regions.qrels 

A file "id_label" is generated.

2. Obtain original rcv1 training/testing file from 
rcv1 web site (B.13. On-Line Appendix 13). 
For example, consider lyrl2004_vectors_train.dat
and run 

./dat2svm.py lyrl2004_vectors_train.dat id_label  > output_file