Codes for the SIGIR 2022 paper Mutual disentanglement learning for joint fine-grained sentiment classification and controllable text generation

---

Requirements

conda create --name dual-scsg python=3.8
pip install -r requirements.txt

---

Data

Download the datasets and put them on the corresponding folds.

• Fine-grained ABSA
  • laptop15
  • restaurant14
  • restaurant15
  • restaurant16
  • mams
• Coarse-grained ABSA
  • trip_advisor
  • beer_advocate

---

Usage

Step 1. Train language model

python run_lm.py 

Step 2. Train the backbone dual models

python run_dsl.py

Step 3. Mutual disentanglemet learning

python main.py

Note: properly configurate the parameters of each script.

Cite

@inproceedings{sigir22-dual-scsg,
  author       = {Hao Fei and
                  Chenliang Li and
                  Donghong Ji and
                  Fei Li},
  title        = {Mutual Disentanglement Learning for Joint Fine-Grained Sentiment Classification
                  and Controllable Text Generation},
  booktitle    = {Proceedings of the 45th International {ACM} {SIGIR} Conference on Research
                  and Development in Information Retrieval},
  pages        = {1555--1565},
  publisher    = {{ACM}},
  year         = {2022},
  url          = {https://doi.org/10.1145/3477495.3532029}
}

License

The code is released under Apache License 2.0 for Noncommercial use only.