This is a suite of Perl scripts with man pages for preprocessing text collections to create dictionaries and bag/list files for use by topic modelling software. Output is in various sparse vector formats (ldac, Matlab Topic Toolbox, libsvm, ...). Scripts are also given for extracting data from WEX files (Wikipedia dumps), PubMed XML and Reuters RCV1 files. The input format is a simple text file which can be created with XSL from standard XML formats. The input format has fielded entries for things like text, links and categories. The system uses buffering so input files can be 30Gb. Example use: (1) Generate PMI matrices from Wikipedia for coherence evaluation ala Newman, Lau, Grieser and Baldwin, NAACL 2010. (2) Process RCV1, Pubmed, NewsML or WEX files into sparse matrix "bags". (3) Generate bags in LDAC or sparse matrix format.
User Guide (read this first) and other information at: http://www.nicta.com.au/people/buntinew/softwareanddata#dcabags
Sample prepared input data sets and XSL files available at http://www.nicta.com.au/people/buntinew/softwareanddata#data
Help and discussion forums at https://forge.nicta.com.au/forum/?group_id=129
- Changes to previous version:
Cleaned up man pages and created user guide.
No one has posted any comments yet. Perhaps you'd like to be the first?
Leave a comment
You must be logged in to post comments.