This code is used to reproduce the main experiment of our paper.
- Python 3.6.8
- Pytorch 1.2.0
- entmax (pip install entmax)
- Jupyter Notebbok
- DIGINETICA: http://cikm2016.cs.iupui.edu/cikm-cup or https://competitions.codalab.org/competitions/11161
- RETAILROCKET: https://www.kaggle.com/retailrocket/ecommerce-dataset
- preprocess_rr: for RETAILROCKET dataset to generate session.
- Preprocess: generate train and test set(for RETAILROCKET dataset, you need run
preprocess_rr .py
first) - Metric: HR and MRR
- DualAdaptiveTrain: the model of DN dataset
- DualAdaRR3: the model of RR dataset
This folder contains the model that we have trained. Loading this model could directly check results.
This folder contains all the baselines we compared in the paper.
For SKNN, STAN, STAMP, Bert4Rec, GC-SAN and CoSAN we implement them by ourselves referring to the original paper and open source implementation.
For GRU4Rec, SR-GNN, we use the author's source code and for FPMC we use the open source implementation.