Details about the multi-domain dataset?

[stevenbedrick]

Hello! Do you happen to have any more information about which version of the Blitzer et al. multi-domain sentiment dataset you used for this? The version posted on Mark's website ( https://www.cs.jhu.edu/~mdredze/datasets/sentiment/ ) does not seem to contain any files that line up with what evaluation.py is looking for (multi-domain-sentiment_indomain_*.txt), and unlike the way that the script handles the SemEval dataset, it does not look like the code for generating the test/train split for the multi-domain data is part of this script. Any pointers would be most welcome!