colall / ione

Source Code and data for IJCAI 2016 paper "Aligning Users Across Social Networks Using Network Embedding"

IONE

Source Code and anonymous twitter_foursquare data for IJCAI 2016 paper "Aligning Users Across Social Networks Using Network Embedding"

With the permission of sharing anonymous twitter_foursquare data from Prof.Jiawei Zhang, we share the anonymous data and code of IONE here. If you use the data and code of this project, please cite the following three papers. Note that the data and code only can be used for research purposes

@inproceedings{zhang2015integrated,
title={Integrated Anchor and Social Link Predictions across Social Networks.},
author={Zhang, Jiawei and Philip, S Yu},
booktitle={IJCAI},
pages={2125--2132},
year={2015}
}

@article{DBLP:journals/corr/abs-1804-09874,
author = {Jiawei Zhang},
title = {Social Network Fusion and Mining: {A} Survey},
journal = {CoRR},
volume = {abs/1804.09874},
year = {2018},
url = {http://arxiv.org/abs/1804.09874},
archivePrefix = {arXiv},
eprint = {1804.09874},
timestamp = {Mon, 13 Aug 2018 16:47:39 +0200},
biburl = {https://dblp.org/rec/bib/journals/corr/abs-1804-09874},
bibsource = {dblp computer science bibliography, https://dblp.org}
}

@inproceedings{liu2016aligning,
title={Aligning Users across Social Networks Using Network Embedding.},
author={Liu, Li and Cheung, William K and Li, Xin and Liao, Lejian},
booktitle={IJCAI},
pages={1774--1780},
year={2016}
}

When running the code, firstly check the Vars.java to make sure that the path of the data is correct. New two folds which are named as "embeddings" in foursquare fold and twitter fold.

Description of Data directory

AcrossNetworkEmbeddingData

	foursquare:
	
		following: the relation file, "1  2" means user 1 is the follower of user 2.  			   			
		following.reverse: the reverse relation file, for model which considers only one direction context. ONE model.
		
	twitter:
	
		the same as the foursquare fold
		
	twitter_foursquare_groundtruth:
	
		groundtruth: the groundtruth for our experiment, the anchor users between twitter and foursquare. 
		Note that pls make the anchors as the *same* id during the pre-preparation, 
		although the testing anchors will have the same id, 
		they will *not* take part in the training progress as they are not contained in the groundtruth.x.foldtrain.train file.	
		
		groundtruth.x.foldtrain.train, the traning anchors, which are the 0.x of all the anchors.
		
		groundtruth.x.foldtrain.test,  the testing anchors, which are the 1-0.x of all the anchors.

There are four models of our paper, INE, ONE , IONE and IONE-S

For the INE and ONE, run the INE.java. If you run the INE model, use "test.Train(10000000, "", 100);". For ONE model, reverse the relationship between users for a new following file, then run the INE.java.

For the IONE, run the IONE.java.

We will add the IONES model with anonymous data as soon as possible.

All the embeddings in the embedding directory of foursquare and twitter.

The getPrecision.java is used for p@1-p@30 calculation of our model。 Note that in the evaluation, we use the |UnalignedUsers|=(|UnalignedAnchors|+|UnalignedNonAnchors|) as the candidate list. The Precison may be lower than metrics which only use |UnalignedAnchors| as the candidate list

Feel free to contact me (Liu Li [email protected]) when you have any problems about the paper or the code.

ione's Issues

锚链接文件groundtruth里只有一列，是因为有层间链接的两层网络对应节点标号相同吗？

比如5，就是源网络的5号和目标网络的5号之间有层间连接，是这个意思吗？谢谢~

I want to ask you the something about the data

I view the code and want to konw where is the anchor nodes, 0-1609 is the anchor nodes?

Question about the ground-truth data structure

Dear Mr./Ms. Liu,

Your work in the IJCAI 2016 paper "Aligning Users Across Social Networks Using Network Embedding" is impressive, and I’m trying to test the code you provided.

But I have a question about the structure of the twitter_foursquare_groundtruth data, which you introduced as “the groundtruth for our experiment, the anchor users between twitter and foursquare.”
For example, in the file groundtruth.9.foldtrain.train.number, the data contains 1451 lines single-column ids like following:

0
1
2
4
5
...

I wonder how it could indicate the anchor links between twitter and foursquare data, and to users in which dataset, Twitter or Foursquare, do these indexes represent to? In my opinion, it should be in the form that could indicate which user in twitter has the same account in foursquare (like an edge):

twitter anchors(id)<->foursquare anchors(id)
0<->4
1<->5
...

So could you further explain the structure of how the data indicate the ground-truth relationship?

Another point I would like to confirm is that in your paper, the number of both twitter and foursquare user node is more than 5000, but the number of anchor users is only 1609. Does that mean that the training/test data could only be sampled from these 1609 anchors, as the relationship of the rest of the users in these two network (which is |UnalignedNonAnchors| if I am correct) remain unknown, thus could not be used to test the prediction result is correct or wrong?

Thank you very much for the time.

source图和target图的节点可以打乱吗

了解到IONE的输入source和target网络对节点的要求是对应的节点名称必须一致，在2018年的综述A comparative study on network alignment techniques里，HT等人对您的模型做出了修改，使模型支持打乱的节点输入。我用相同的数据集和相同的配置，同时用您的IONE模型和HT等人修改的IONE模型跑了几组实验，发现结果有较大的区别，不知道老师是否有官方支持打乱节点的IONE版本呢😄