SentiX

This is the source code of our COLING 2020 paper <SentiX: A Sentiment-Aware Pre-Trained Model for Cross-domain Sentiment Analysis>

Requirements

The following pre-trained models are available for download from Google Drive:

SentiX: PyTorch version, same setting with BERT-base，loading model with transformers.

To train SentiX, simply run:

python pretrain_multigpu_final.py --dataset data_all --max_seq_len 256 --batch_size 16 --gradient_accumulation_steps 4

To test after setting model path:

run_multidomain_sentiment_analysis.sh

Name		Name	Last commit message	Last commit date
data		data
data_processing		data_processing
models		models
README.md		README.md
aspect_train.py		aspect_train.py
aspect_train_back.py		aspect_train_back.py
data_utils.py		data_utils.py
data_utils_aspect.py		data_utils_aspect.py
data_utils_aspect_back.py		data_utils_aspect_back.py
data_utils_pretrain_final.py		data_utils_pretrain_final.py
data_utils_pretrain_multidomain_final.py		data_utils_pretrain_multidomain_final.py
few_shot_train.py		few_shot_train.py
multidomain_inference.py		multidomain_inference.py
multidomain_train.py		multidomain_train.py
multidomain_train_few_shot.py		multidomain_train_few_shot.py
pretrain_multigpu_final.py		pretrain_multigpu_final.py
pretrain_multigpu_multidomain_final.py		pretrain_multigpu_multidomain_final.py
run_aspect_based_sentiment_analysis.sh		run_aspect_based_sentiment_analysis.sh
run_aspect_based_sentiment_analysis_bert_large.sh		run_aspect_based_sentiment_analysis_bert_large.sh
run_few_shot_multidomain_sentiment_analysis.sh		run_few_shot_multidomain_sentiment_analysis.sh
run_multidomain_sentiment_analysis.sh		run_multidomain_sentiment_analysis.sh
run_sentiment_analysis.sh		run_sentiment_analysis.sh
train.py		train.py
zero-shot_sentiment_classification.py		zero-shot_sentiment_classification.py