The current implementation is still under development and is only intended for testing and experimentation purposes.
This repository contains the implementation of the L-GGNN-ATT model, which is a variant of the SR-GNN model [1] that uses the Relevant Order Graph Formulation as its connection scheme.
[1] Wu, S., Tang, Y., Zhu, Y., Wang, L., Xie, X., & Tan, T. (2019, July). Session-based recommendation with graph neural networks. In Proceedings of the AAAI conference on artificial intelligence (Vol. 33, No. 01, pp. 346-353).
To train the model, create a directory with the name of your dataset in the datasets folder and include two files train.csv and test.csv in it. These files should contain the training sessions and testing sessions respectively. A session is a sequence of item IDs separated by commas (e.g 5, 12, 52, 1004, 6, 52, 5). Then, run main.py with DATASET_NAME set to the name of your dataset directory and the hyperparameters adjusted to your needs.