Topic Driven Adaptive Network for Cross-Domain Sentiment Classification

Introduction

Code for Topic Driven Adaptive Network for Cross-Domain Sentiment Classification.

Requirements

• python 3.8.8
• pytorch 1.8.1
• gensim 4.0.1
• nltk 3.6.2
• numpy 1.20.2

Environment

• OS: Ubuntu 20.04.2 LTS
• GPU: NVIDIA TITAN 1080Ti * 2
• CUDA: 10.2

File organization

• preProcess.py: the code of prepare data
• parameter.py: the parameter for training
• train.py: run all cross-domain tasks
• model.py: the code of TDAN
• raw_data/: unprocessed data
• processedData/: processed data
• result/: training result
• model/: model file
• wordvec: word embeddings

Running

prepare the data

Download Google Word2Vec. Extract the file and put it under the ./wordvec folder.

Run preProcess to prepare the data. This step involves training of LDA so it might takes a moment. LDA is by default trained with four threads and you can adjust the thread number according to your cpu.

python preProcess.py

Run all cross-domain tasks

./run.sh