This repository holds Pytorch (near)-implementations of the following papers:
- Hierarchical Attention Networks for Document Classification paper
- Neural Sentiment Classification with User & Product Attention paper
Implementations are single-gpu batched.
-
Pytorch 0.4
-
TQDM
-
Gensim (Only for
BuildW2VEmb.pyto build word embeddings with word2vec) -
Spacy 1.9 (Only for
prepare_data.pyto tokenize reviews and split in sentences)
chmod +x minimal_ex_cuda.sh./minimal_ex_cuda.sh
han.py-> Hierarchical Attention Networks for Document Classification model training script.nscupa.py-> Neural Sentiment Classification with User & Product Attention model training script.
han.py and nscupa.py expect a pickled dictionnary with the following keys as input:
input = {"data":data,"splits":splits,"rows":("user_id","item_id","review","rating")}
Where:
- The
"data"field holds is a list of("user_id","item_id","review","rating")tuples where id's and ratings arestrand review is tokenized and sentence-splitted textlist(list(str)). - The
"splits"key is a list integerslist(int). It's each data point split. - The
"rows"is used as helper for tuple indexing in the former scripts.
=> A helper script prepare_data.py is provided to create such input pickle file.
han.py and nscupa.py can use pre-trained embedding. It expects a .txt where each line is a word followed by its vector. The first line of this file provides the number of words and the size of each vectors. BuildW2VEmb.py is provided to build word embeddings from data using the word2vec-skipgram algorithm.
If no pre-trained embeddings are provided han.py and nscupa.py build embedding dictionnaries and vectors on the fly (---max-feat arg).
prepare_data.pytransforms gzip files as found on Julian McAuley Amazon product data page to a list of(user,item,review,rating)tuples.minimal_ex(_cuda).shDoes everything and start learning (justchmod +xthem).fmtl.pyholds data managing objects.Nets.pyholds neural network models.beer2json.pyis an helper script to convert ratebeer/beeradvocate datasets.BuildW2VEmb.pycan help you build word embeddings from data.
This code has been written for python-3.6. If you are only using python-2.7, switch.