Recursive-Neural-Structural-Correspondence-Network
This is an instruction file for successfully running the code provided.
The code is an implementation of the following paper:
"@InProceedings{P18-1202, author = "Wang, Wenya and Pan, Sinno Jialin", title = "Recursive Neural Structural Correspondence Network for Cross-domain Aspect and Opinion Co-Extraction", booktitle = "Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", year = "2018", publisher = "Association for Computational Linguistics", pages = "2171--2181", location = "Melbourne, Australia", url = "http://aclweb.org/anthology/P18-1202" }"
Please follow the following steps:
-
Go to 'util' folder to produce intermediate files:
- Download Stanford dependency tree parser
- use '10depParse.py' to generate dependency trees
- use '20dtreeLabel_cross_split.py' to build data structures and split data to training and testing
- use '30word_embedding.py' to store pre-trained word embeddings
-
Go to the main folder to conduct training
- run 'train_depnn_cross.py' to pre-train recursive neural network first
- run 'train_joint_cross.py' to train the joint model
Note: When the digital device dataset is used as the source domain, we remove the sentences without any aspect words to train the model.