These are the related documents used in the generation experiments in:

William Yang Wang and Miaomiao Wen, "I Can Has Cheezburger? A Nonparanormal Approach to Combining Textual and Visual Information for Predicting and Generating Popular Meme Descriptions", to appear in the 2015 Conference of the North American Chapter of the Association for Computational Linguistics – Human Language Technologies (NAACL HLT 2015), long paper, Denver, CO., USA, May 31-June 5, ACL.

cheezburger.txt and memegenerator.txt: 
these are the meme descriptions we used to build our own Lucene meme search engine. Each meme description
is indexed as a "document" in the IR engine. 

queryImageIDs.txt: the names of the 50 test images.

GoogleImageParse.txt: the reverse image search (image parsing) results generated by Google.

IR_candidate_descriptions.txt: the returned meme descriptions from our Lucene meme IR engine. The title of each "document" is the candidate meme.  

eval.feat: the feature file that we used to rerank the candidate descriptions. The format is the same as train.feat in the prediction experiment, except 
that we do not know the label ("0").

reference.csv: the reference that we compared to calculate BLEU scores.
