Hi,
I took a brief look at the pipeline for multi-label document-mode instance ids. For sequence classification, there is a method, TextClassificationTarget.setSuffix(String aSuffix) that can be used in the Reader, to add a string suffix to the unit id as printed in the id2outcome file. (This method is currently nonfunctional for some of the OutcomeId reports, and needs a bug report, but the functionality exists up until the OutcomeId report.)
I didn't find any similar method for document-mode classification, so I suggest altering getFullId(JCas jcas) in org.dkpro.tc.core.feature.InstanceIdFeature as follows:
private static String getFullId(JCas jcas)
throws TextClassificationException
{
JCasId jcasId = JCasUtil.selectSingle(jcas, JCasId.class);
DocumentMetaData dmd = DocumentMetaData.get(jcas);
String documentName = dmd.getDocumentId();
return "" + jcasId.getId() + "-" + documentName;
}
I ran it on org.dkpro.tc.examples.multi.document.MekaComplexConfigurationMultiDemo, and the id2outcome files show filenames.
Best,
Emily