The code implementation for the paper "Unsupervised User Identity Linkage via Factoid Embedding (ICDM'18)".
To cite:
@inproceedings{xie2018unsupervised,
title={Unsupervised user identity linkage via factoid embedding},
author={Xie, Wei and Mu, Xin and Lee, Roy Ka-Wei and Zhu, Feida and Lim, Ee-Peng},
booktitle={2018 IEEE International Conference on Data Mining (ICDM)},
pages={1338--1343},
year={2018},
organization={IEEE}
}
python main.py --path ../data_for_experiment/facebook_twitter/ \
--source_prefix fb --target_prefix tw --source_col 0 --target_col 1 \
--name_method tfidf --image_exist True --user_dim 1024+256 \
--name_concatenate True \
--image_dim 256 --n_iter 52 --supervised False\
--image_method vgg16 --skip_network False;
Or see example.sh
usage: main.py [-h] [--path PATH] [--skip_network SKIP_NETWORK]
[--source_prefix SOURCE_PREFIX] [--target_prefix TARGET_PREFIX]
[--source_col SOURCE_COL] [--target_col TARGET_COL]
[--name_dim NAME_DIM] [--name_concatenate NAME_CONCATENATE]
[--name_preprocess NAME_PREPROCESS] [--name_method NAME_METHOD]
[--screen_name_exist SCREEN_NAME_EXIST]
[--image_exist IMAGE_EXIST] [--image_method IMAGE_METHOD]
[--image_identical_threshold IMAGE_IDENTICAL_THRESHOLD]
[--image_dim IMAGE_DIM]
[--cosine_embedding_batch_size COSINE_EMBEDDING_BATCH_SIZE]
[--cosine_embedding_learning_rate COSINE_EMBEDDING_LEARNING_RATE]
[--supervised SUPERVISED] [--snapshot SNAPSHOT]
[--snapshot_gap SNAPSHOT_GAP] [--n_iter N_ITER]
[--warm_up_iter WARM_UP_ITER] [--user_dim USER_DIM]
[--nce_sampling NCE_SAMPLING]
[--triplet_embedding_batch_size TRIPLET_EMBEDDING_BATCH_SIZE]
[--triplet_embedding_learning_rate_f TRIPLET_EMBEDDING_LEARNING_RATE_F]
[--triplet_embedding_learning_rate_a TRIPLET_EMBEDDING_LEARNING_RATE_A]
[--stratified_attribute STRATIFIED_ATTRIBUTE]