NeRank

@Author: Zeyu Li, Jyun-Yu Jiang, Yizhou Sun, Wei Wang

Introduction

This is the repo of "Personalized Question Routing via Heterogeneous Network Embedding". Bibtex citation format:

@inproceedings{li2019personalized,
  title={Personlaized Question Routing via Heterogeneous Network Embedding},
  author={Li, Zeyu and Jiang, Jyun-Yu and Sun, Yizhou and Wang, Wei},
  year={2019},
  organization={AAAI Conference on Artificial Intelligence (AAAI)}
}

Install Dependencies

NeRank is prototyped in PyTorch 0.4.0a0. Other package dependencies are listed in "requirements.txt". You can run

$ pip install -r requirements.txt

to install the packages that are needed.

Preparing Data

The archive of the dataset is available here. Download the dataset and unzip the 7z files into ./raw/. Then run

$ python src/preprocessing.py [name of dataset] [threshold] [prop of test] [test sample size]

to preprocess the XML to json.

Parameters:

• name of dataset: name of the dataset, such as "Biology", "English", and "3dpringting".
• threshold: #. of entities to be selected as for the training.
• prop of test: float. proportion of questions for testing.
• test sample size: number of candidate questions in each test samples. The produced files will appear in ./data/.

Generate Metapath and Input Pairs

Generate the input pairs by generate_walk.py using the following command.

$ python src/generate_walk.py [name of dataset] [length] [num_walk] [window_size]

Parameters:

• name of dataset: Similar to above.
• length: the length of the generated walks.
• num_walk: the number of times each node is covered by walks.
• window_size: the size of Skip-gram sampling window.

Run

Run NeRank by the following command:

$ ./run.sh 

But before running run.sh, please look into it and tweak the settings.

Parameter:

• ID: the identifier of a certain training/testing, will be used in output file name. You would see the performance in ./performance/.