/Targeted-aspect-based-sentiment-analysis-using-BERT

Primary LanguageJupyter Notebook

ABSA as a Sentence Pair Classification Task

Requirement

  • pytorch: 1.0.0
  • python: 3.7.1
  • tensorflow: 1.13.1 (only needed for converting BERT-tensorflow-model to pytorch-model)
  • numpy: 1.15.4
  • nltk
  • sklearn

Step 1: prepare datasets

SentiHood

Since the link given in the dataset released paper has failed, we use the dataset mirror listed in NLP-progress and fix some mistakes (there are duplicate aspect data in several sentences). See directory: data/sentihood/.

I changed the sentence pair task to a single sentence task where we add the aspect after the location and then we train the model to predict from {None, positive, negative}

Run following commands to prepare datasets for tasks:

cd generate/
python3 generate_sentihood.py

Step 2: train

We have provided different models for using bert use bert_model.py, for using distilbert use model.py

For example, BERT-pair-NLI_M task on SentiHood dataset:

python model.py --num_epochs 10 --max_seq_length 512


## Step 3: evaluation

Evaluate the results on test set (calculate Acc, F1, etc.).

python evaluation.py --pred_data_dir results/test_ep_4.txt