search

SCIR-MSA-Team / LC-ACSA

The code repository for NLPCC2021 paper "Locate and Combine: A Two-Stage Framework for Aspect-Category Sentiment Analysis".
4 stars 0 forks source link

readme

Python 3.6

LC-ACSA

The code repository for NLPCC2021 paper "Locate and Combine: A Two-Stage Framework for Aspect-Category Sentiment Analysis".

Data preparation

Download the glove.840B.300d.txt,bert-base-uncased,restaurant post-training BERT to the corresponding paths in your computer.Remenber to modify the dataset path according to your setting.

Set up the environment

pip install -r requirement.txt

Running the code

Run the following commands for restaurant 2014 dataset. For the restaurant 2014 hard, restaurant large and restaurant large hard datasets, change rest14DevSplit in the following commands to rest14_hard, rest_large and rest_large_hard respectively.

Train (for LC-LSTM)

  1. python train_ATE.py --dataset allRest --lr 3e-5 --dropout 0.2
  2. python train_ACD.py --dataset rest14DevSplit --epoch 10
  3. sh scripts_ACSC/CDT_categoryEmbedding_gcnCat_extractTerm.sh cuda:0 train rest14DevSplit 40

Test (for LC-LSTM)

  1. sh scripts_ACSC/CDT_categoryEmbedding_gcnCat_extractTerm.sh cuda:0 test rest14DevSplit

Trained models

Trained models can be found here. For Chinese, trained models can also be found here and the key is u6m0.