resistzzz / HGCAN

HGCAN: The source code of a paper for TWEB under review

We have used three public datasets in our paper as follows:

You can download the raw datasets from following url:

• DIG: http://cikm2016.cs.iupui.edu/cikm-cup

• YC: http://2015.recsyschallenge.com/challenge.html

• JD: https://jdata.jd.com/html/detail.html?id=8

We have provided the pre-processed datasets to support running our codes, you can find them in datasets directory.

How to run this code

Firstly, we run the file build_graph.py to generate the graph files.

(We have provided the generated graph files in datasets directory).

Then we run the file main.py to train the model and simulate the conversations to obtain the final performance.

(You need to manually create a log folder ./log to store the log files, if you are prompted for a missing log file. )

Take DIG dataset as example:

python build_graph.py --dataset DIG
python main.py --dataset DIG --batch_size 100 --cate_select 5 --dice_threshold 1.0

Requirements

• Python 3.6
• pytorch==1.9.0+cu111

After you have configured your environment, you can run this code directly to obtain experimental results.