Semantic Textual Similarity task 2012/2013/2014
-
numpy
-
sklearn (scikit-learn)
-
nltk (may require X11 under OS X)
-
Google n-gram word counts and Takelab LSA models under directory _data
Make sure you have lib/python in your PYTHONPATH, e.g. in Bash use
$ export PYTHONPATH=$PYTHONPATH:~/Projects/SemTextSim/github/STS13/lib/python
-
Make sure you have Google n-gram word counts and Takelab LSA models under directory _data
-
Add 2014 trial data files under new directory data/STS2014-trial
-
Create lib/python/sts/sts14.py defining dirs, ids and filenames for STS14 trial data
-
Create lib/python/ntnu/sts14.py defining dirs and filenames of features for STS14 trial data
-
Create Takelab features by adding function to ntnu/make-takelab-feat.py:
make_feats(sts.sts14.trial_input_fnames, ntnu.sts14.test_dir, with_lsa)
Temporary comment out calls to make_feats for other STS datasets
-
Change to dir ./ntnu and run ./make-takelab-feat.py
New features appear in files out/STS2014-trial//.txt
Suppose we have a new feature called "my_feat" that we want to try on the MSRpar dataset from STS12.
-
Add the feature to the training and test data as files
out/STS2012-train/MSRpar/my_feat.txt out/STS2012-test/MSRpar/my_feat.txt
-
Run a script ntnu/my_feat.py
Check comments in the script