Using Stanford CoreNLP and SVM-Rank in a Supervised Approach to Text Difficulty Ranking
Dylan Mann, Pia Kochar
mannd, pkochar
Instructions to run the code:
To run our code, you need to first convert your excerpts into one excerpt per file (this can be done using a variation of the function convert_to_files(), and then put them into the proper location and edit the output_dir variable in run_corenlp.sh to use the proper directories. Then you run that script from anywhere on eniac or the biglab machines. Then you must run the two functions.
generate_svm_rank_train() and
generate_svm_rank_test()
To run these, you need to edit a few parameters:
the location of your xml files (default “xml/train” and “xml/test”)
the name of the output files (default “train.dat” and “test.dat”)
Then you can run the python program using python3 to produce the test and train data files.
Afterwards you need to download or build a copy of svm_rank here:
https://www.cs.cornell.edu/people/tj/svm_light/svm_rank.html
Once you have the executable for your system, you train it using the commandsvm_rank_learn -c 10 train.dat model.dat
and classify the test data with the command:svm_rank_classify test.dat model.dat predictions.txt
and your 50 predicted values will be in the proper format in predictions.txt