Code and data accompanying paper "Twitter Homophily: Network Based Prediction of User’s Occupation": ACL 2019.
- Twi_data folder contains the training/development/test sets split
- Twi_data folder contains the processed dataset used in GCN model and Deep Walk model, extracted in around February 2018.
- The dataset is processed from collected Twitter ego-network for a sample of Twitter users whose occupational classes are labeled. Please refer to the paper for collection and processing details.
- Due to privacy concerns, we are not releasing raw Twitter network with Bio.
- Total number of edges: 586303
- Total number of main users (with real labels): 4557
- Total number of users (including main users): 34603
- src folder contains code for running GCN model on the processed dataset.
- To execute, please
cd src
to navigate to src folder and thenpython train_model.py
.
We include the jupyter notebook that processes the raw network into a densenly connected network which is then used with GCN and DeepWalk.
- 'python=3.6'
- 'torch>=1.0'
- 'scipy'
- 'numpy'
- 'pandas'
The code for GCN model is forked and modified from https://github.com/tkipf/gcn
If you find the resource useful to you, please cite:
@InProceedings{pan-19-homophily,
author = {Pan, Jiaqi and Bhardwaj, Rishabh and Lu, Wei and Chieu, Hai Leong and Pan, Xinghao and Puay, Ni Yi},
title = {Twitter Homophily: Network Based Prediction of User’s Occupation},
booktitle = {Proceedings of ACL},
year = {2019}
}
If you have any questions, please feel free to contact jiaqi.pan1019@gmail.com
or rishabhbhardwaj15@gmail.com
.