IBruceWayne / gittest

Unified Feature and Instance Based Domain Adaptation for Aspect-Based Sentiment Analysis

This repository contains code for our EMNLP 2020 paper:

Unified Feature and Instance Based Domain Adaptation for Aspect-Based Sentiment Analysis

Dependencies

-Python 3 -torch 1.5.1 -tqdm 4.42.1 -pytorch-transformers 1.2.0

Usage

1. First step, run: bash ./scripts/run_feature_learning.sh to prepare the unlabel dataset for auxiliary tasks and do feature-learning for 10 domain transfer pairs.

2. Second, run: bash ./scripts/run_uda.sh to get the BERT-uda results for 10 domain transfer pairs.

3. For BERT-based baseline, run: bash ./scripts/run_base.sh to get the BERT-base results for 10 domian transfer pairs.