Skip to content

[EMNLP2019] Rethinking Attribute Representation and Injection for Sentiment Classification

License

Notifications You must be signed in to change notification settings

rktamplayo/CHIM

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CHIM

[EMNLP2019] Rethinking Attribute Representation and Injection for Sentiment Classification

This PyTorch code was used in the experiments of the research paper

Reinald Kim Amplayo. Rethinking Attribute Representation and Injection for Sentiment Classification. EMNLP, 2019.

Data

IMDB, Yelp2013, and Yelp2014 datasets are originally from here. I did some changes, which I cannot unfortunately recall everything, with the format (not the content) of the file (e.g., changed the ordering of the input and output, etc.).

I am therefore sharing my version of the above datasets, as well as the Amazon datasets in the paper, in this link. The link will direct you to a directory named data. You will need to download GloVe vectors as well and save the file as glove.txt inside the data directory.

If you are using any of the above three datasets, please also cite the original paper. The BibTeX is shown at the end.

Train and Evaluate CHIM

To train the model, simply run using the format:

python src/train.py <dataset> weight.chunk.imp <inject_location> 300 <chunk_ratio> <gpu_device>

where:

  • dataset is the name of the dataset directory dataset folder (e.g., yelp2013).
  • inject_location is the location to inject the attributes. Choose from the following: embed, encode, pool, classify. Multiple locations can also be used by separating them with a comma (e.g., embed,encode)
  • chunk_ratio is the chunk size factors discussed in the paper. The one used in the paper is 15.
  • gpu_device is the GPU device number.

Evaluation is done similarly, but with another file:

python src/evaluate.py <dataset> weight.chunk.imp <inject_location> 300 <chunk_ratio> <gpu_device>

Cite the Necessary Papers

To cite the paper/code/data splits, please use this BibTeX:

@inproceedings{amplayo2019rethinking,
	Author = {Reinald Kim Amplayo},
	Booktitle = {EMNLP},
	Location = {Hong Kong, China},
	Year = {2019},
	Title = {Rethinking Attribute Representation and Injection for Sentiment Classification},
}

If using either of the IMDB/Yelp2013/Yelp2014 datasets, please also cite the original authors of the datasets:

@inproceedings{tang2015learning,
	Author = {Duyu Tang and Bing Qin and Ting Liu},
	Booktitle = {ACL},
	Location = {Beijing, China},
	Year = {2015},
	Title = {Learning Semantic Representations of Users and Products for Document Level Sentiment Classification},
}

If there are any questions, please send me an email: reinald.kim at ed dot ac dot uk