/Unsupervised-meta-review-generation

Primary LanguagePythonMIT LicenseMIT

Unsupervised-meta-review-generation

Preprocessing

To run our framework, you need to preprocess your dataset and extract opinion phrases from the input sentences.

Running

The workflow has following 4 steps.

  • Step 1. Data preparation
  • Step 2. Training
  • Step 3. Aggregation
  • Step 4. Generation

You can skip Steps 1-3 by downloading our pre-trained model and dataset.

Step 1. Data preparation

Opinion Extraction

python src/aspect_extraction/aspect_extraction/aspect_generator.py

Data preparation for generation

python src/prepare_new.py \
  config/prepare_default.json

Step 2. Training

pip install -r requirements.txt

python src/train_new.py \
  config/prepare_default.json \
  config/train_default.json

Step 3. Aggregation

python src/aggregate_final.py \
  config/aggregate_default.json \

Step 4. Generation

python src/evaluate_final.py \
  config/aggregate_default.json \