WYJSJTU / GluonNlp-BERT

Sentiment Analysis on Chinese and English Datasets based on GluonNlp BERT model

GluonNLP-BERT

Sentiment Analysis on Chinese and English Datasets based on GluonNlp BERT model

Requirements

• Python 3.7
• CUDA 10.2
• Cudatoolkit 10.2.89
• MXNet mxnet-cu102-1.7.0.post1
• GluonNLP 0.9.2

Datasets

• waimai_10k: Chinese cutomer review on takeaway app with sentiment annotation of positive and negetive, 2 categories classification task.
• simplifyweibo_4_moods: Chinese users posts on Sina Weibo with sentiment annotation of joy, angry and disgust，3 categories classification task.
• Meituan ASAP: Chinese restaurant review with star annotation from 1 to 5, 5 categories classification task.
• Movie comment_50k: English polar movie review with sentiment annotation of positive and negetive, 2 categories classification task.

Data Preperation

Please download datasets from this link: Link：https://pan.baidu.com/s/1dRe7gwVvZxwBlJloO4ExfA Password：y5d3

And place the folders to the ./data directory

Training and Testing

python3 finetune_classifier.py --gpu 4 --task_name ${TASK} --batch_size 8 --seed 0 --lr 2e-5 --warmup_ratio 0.1 --epochs 5 --log_interval 100 --accumulate 4

Task names:

• waimai_10k: waimai
• simplifyweibo_4_moods: weibo3
• Meituan ASAP: meituan
• Movie comment_50k: movie