ESim: Meta-Path Guided Embedding for Similarity Search in Large-Scale Heterogeneous Information Networks

Publications

Please cite the following paper if you are using our tool. Thanks!

• Jingbo Shang, Meng Qu, Jialu Liu, Lance M. Kaplan, Jiawei Han, Jian Peng, "ESim: Meta-Path Guided Embedding for Similarity Search in Large-Scale Heterogeneous Information Networks", submitted to SIGKDD 2017, under review. arXiv:1610.09769 [cs.CL]

Deployment

The implementation of ESim relies on two external packages: Eigen and GSL. We use Ubuntu as an example for the deployment.

First, let's install the GSL.

sudo apt-get install libgsl0ldbl

Second, we will download Eigen and build the binaries for ESim using Makefile.

make

In the end, the binaries are in the folder bin/.

Default Run

The default run will download a bibliographic network (dblp_data) from our data mining group server.

./run.sh

After ESim generates the embedding vectors, users may use the search script to do similarity search. The usage is similar to the similar word search in word2vec.

./search.sh

Input Formats

ESim takes three files as input for training, which are the node file, link file and meta-path file.

Node File

The node file contains the nodes in the training network. Each line of the file describes the name (a string) and the type (a char) of a node, which are separated by space. An example is presented below:

Jiawei_Han a
PathSim p
KDD v
...

Link File

The link file contains the links between different nodes. Each line of the file describes the head node and the end node of a link, which are separated by space. An example is presented below:

Jiawei_Han PathSim
PathSim Jiawei_Han
PathSim KDD
KDD PathSim
...

Meta-path File

The meta-path file contains the meta-paths used for training. Users can specify multiple meta-paths, and each line describes a meta-path and its relative weight. An example is presented below:

apa 0.3
apvpa 0.7
...

Example Output

Below is the example output on the DBLP dataset with the query as "Jiawei_Han" and the meta-path as "apvpa".

------------------------------------
Xifeng_Yan		0.931779
Ramin_Afshar	0.930142
Zhenhui_Li		0.930056
Tianyi_Wu		0.923093
Cindy_Xide_Lin	0.923068
Yiwen_Yin		0.919200
Xiaolei_Li		0.918408
Petre_Tzvetkov	0.917137
Chinying_Chaou	0.915410
Yizhou_Sun		0.914187

Contact:

If you have any questions about the codes and data, please feel free to contact us.

Jingbo Shang, [email protected]
Meng Qu, [email protected]