Domain_Adaptation_ACL2018

The repository of domain adaptation project for the Examining Temporality in Document Classification in ACL 2018. The slides will come soon.

Table of Contents

• Installation
• Data
• Usage
• Contact and Citation

Installation

1. Platform:

• Ubuntu 16.04
• Anaconda, Python 3.6

2. Run the followings to create environment:

• conda env create -f environment.yml
• python -m nltk.downloader punkt stopwords
• source activate domain

Data

1. Amazon CDs and Vinyl
2. Yelp reviews of Hotel and Restaurant
3. Political Platforms: Political Parties -> United States -> * Party Platform
4. Economical News
5. Vaccine Data

Usage

1. Data extraction and sample:

• Extraction: python extract_data.py within each data folder to extract data.
• Sample: go to the utils folder, run python under_sample.py

2. Run cross domain classification, under the project root folder: python domain_clf_analysis.py
3. Generate the feature vectors: python build_feas.py
4. Run grid search to find the optimal parameters: python grid_search.py
5. Run domain adaptation (section 4.1 and 4.2 in the paper): python run_exps.py
6. Combine both seasonal and non-seasonal information: python build_sgd_base.py

Contact

xiaolei.huang@colorado.edu