• dataset/
  • electronics/
    • uid_voc.pkl: users;
    • mid_voc.pkl: items;
    • cat_voc.pkl: categories;
    • item-info: mapping dict {item:category};
    • reviews-info: interaction records [user, item, rating, timestamp];
    • local_train_splitByUser: train data;
    • local_test_splitByUser: test data;
    • wnd3_alpha_01_theta_09_tri_num_10: triangles data with α=0.1 and θ=0.9;
• triangle_data/: processed triangles data of the public datasets.
• script/: implementations of TGIN.
• triangle_mapreduce.zip: MapReduce implementations of triangle extraction and selection.

We have processed the raw data and upload it to the electronics/ fold. You can use it directly.

Also, you can get the data from the amazon website and process it using the script:

sh prepare_data.sh

You can use the processed triangles data directly, and just skip this step.

python script/gen_wnd_edges.py

We have extracted and selected the triangles of both amazon(books) and amazon(electronics) datasets. You can download and put it into the triangle_data/ folder.

Next, the triangle indexes should be transformed into the input format of the TGIN model.

python process_tridata.py

Also, you can refer to the MapReduce source code in triangle_mapreduce.zip folder to generate triangle indexes.

tar xvf triangle_data/electronics_triangle.tar.gz
tar xvf dataset/electronics.tar.gz 
python script/process_tridata.py

sh run.sh

The code has been tested running under Python 2.7.18, with the following packages installed (along with their dependencies):

• cPickle == 1.17
• numpy == 1.16.6
• keras == 2.0.8
• tensorflow-gpu == 1.5.0