MUDA for Sentiment Analysis

This repo contains the source code for our paper:

Adversarial Training Based Multi-Source Unsupervised Domain Adaptation for Sentiment Analysis

Requirements:

• Python 3.6
• PyTorch 0.4
• PyTorchNet
• scipy
• tqdm (for progress bar)

Model Overview

Brief Introduction

In this paper, we focus on the multi-source unsupervised domain adaptation for sentiment analysis and desire to combine the hypotheses of multiple labeled source domains to derive a good hypothesis for an unlabeled target domain. For this purpose, we introduce two transfer learning frameworks. The first framework is Weighting Scheme based Unsupervised Domain Adaptation (WS-UDA), in which we integrate the source classifiers to annotate pseudo labels for target instances directly. Our second framework is a Two-Stage Training based Unsupervised Domain Adaptation method (2ST-UDA), which further utilize pseudo labels to train a target-specific extractor.

Our model is divided into two parts. The first part is to get the pre-trained model of each module, and the second part is to use our pre-trained model to get the results of our two transfer learning frameworks.

Training Process

Training data downlowd

链接: https://pan.baidu.com/s/1U5GUq99KORj2qYI74oxVZA 提取码: snc4

Training Tips

Using Microsoft's open source tuning tool nni, the final result has a fluctuation of ±0.5%

Before Running

Before starting to run the program, you must set the values of base_save_dir and exp2_model_save_file (exp3_model_save_file) to store the model and parameter files during the training process.

Get The Pre-trained Model For Amazon review dataset

cd code/
python3 get_pre-trained_model_exp2.py

Exp 1: WS-UDA on the Amazon review dataset

cd code/
python3 WS-UDA_exp2.py

Exp 2: 2ST-UDA on the Amazon review dataset

cd code/
python3 2ST-UDA_exp2.py

Get The Pre-trained Model For Amazon review dataset

cd code/
python3 get_pre-trained_model_exp3.py

Exp 3: WS-UDA on the FDU-MTL dataset

cd code/
python3 WS-UDA_exp3.py

Exp 4: 2ST-UDA on the FDU-MTL dataset

cd code/
python3 2ST-UDA_exp3.py