README.md

OVRAU

An Overall low-dimensional Vector Representations for Anchor Users from multiplex heterogeneous social network

Overview

This directory contains code necessary to run the OVRAU algorithm. OVRAU can be viewed as a graph convolutional neural network based on aggregator functions incuding mean aggegator, max-pooling aggregator, and LSTM aggregator. It is especially useful for learning latent representations for shared or anchor users across social networks to capture their intra-network and cross-network structural information.

Requirements

• Python >= 3.7
• TensorFlow >= 2.2.0

Installation

Clone this repository:

git clone https://github.com/AnonymizedAccount/OVRAU
cd OVRAU 

Then, you should install other dependencies using the following command:

pip install -r requirements.txt

Dataset

We provide the used dataset under the folder data to help you understand our code and reproduce our experiment.

Input format

You can also use your own multiplex social network dateset, you should prepare the following three files (train.txt, test.txt, and valid.txt), as long as it fits the following template.

edge_type head tail weight
    r1     n1   n2    1
    r2     n2   n3    1
    .
    .
    .

Here, each line represents an edge which contains three tokens edge_type, head, tail, and weight .

Running the code

To train OVRAU model on the example data, you can simply use the following command:

python src/main.py --input data 

You can also replace the name of provided dataset with your own dataset and do not forget to make data files in the format described above.

the proposed model presents three possible variants depending on the used aggregator function and you can also specify the variant to use using --aggregator argument

• --aggregator mean -- OVRAU with mean-based aggregator (the used aggregator by default)
• --aggregator LSTM -- OVRAU with LSTM-based aggregator
• --aggregator max-pooling -- OVRAU with max-pooling aggregator

These aggregators are described in detail in the paper.