All texts have been prepossessed and all intermediate models/supporting data have been set up.
-
To extract metadata (i.e. Court names, plaintiffs and defendants, etc.) with regular expressions, run
step0. -
To check tfidf method for extracting patent terms and key sentences, run
step2. -
To check logistic regression for extracting patent terms, run
step4&step5. If you wanna change the training set and validation set, do following:- open
step4.py, then modifytrain_idsandvalid_ids; - open
step5.py, then modify the range of for statement beneath the lineif __name__ == "__main__:".
- open
-
To check logistic regression for extracting key sentences, run
step10&step11. If you wanna change the training set and validation set, do following:- open
step10.py, then modifytrain_idsandvalid_ids; - open
step11.py, then modify the range of for statement beneath the lineif __name__ == "__main__:".
- open
-
To check CNN for extracting key sentences, run
step7&step8. If you wanna change the training set and validation set, do following:- open
step7.py, then modifytrain_idsandvalid_ids; - open
step8.py, then modify the range of for statement beneath the lineif __name__ == "__main__:".
- open
If you want to start from the very beginning (i.e. start from preprocessing the texts), run steps [0, 1, 2, 4, 5, 7, 8, 10, 11].
WARNING: THIS WILL CLEAR OUT ALL MANUALLY ANNOTATED KEY SENTENCES LABELS IN preprocessed/sents/*.txt, YOU MAY HAVE TO ENTER THIS DIRECTORY AND ANNOTATE THE SENTENCES AGAIN!
If you are informed that glove word embedding file is missing, you can download glove.42B.300d.50K.w2v from the glove's website and save it as glove.42B.300d.50K.w2v.txt under the directory ivan_data
ACKNOWLEDGEMENTS: Most of our code for CNN and calculating tf-idf are downloaded from the github repo of anlp19 (https://github.com/dbamman/anlp19) and modified to fit our work. This saves us plenty of time. Also, the patent case data are downloaded from Westlaw (http://www.westlaw.com).