Offical repository for paper: Knowledge-Guided Semantically Consistent Contrastive Learning for Sequential Recommendation.

In our experiments, the Toys, Grocery, Home and Sports datasets are from http://jmcauley.ucsd.edu/data/amazon/. The interaction data is generated from user review records, and the substitute and complementary relations are extracted from item mata data.

You can run KGSCL with the following code:

python runKGSCL.py --dataset toys --train_batch 512 --lamda1 0.1 --lamda2 1.0 --insert_ratio 0.2 --substitute_ratio 0.7