wangzhen38/tisasrec-paddle

模型名称 TiSASRec: Time Interval Aware Self-Attention for Sequential Recommendation

1. 简介

但是大多数序列化推荐模型都有一个简化的假设，即这些模型都将交互历史视为一个有顺序的序列，没有考虑这个序列中交互物品之间的时间间隔（即只是建模了时间顺序没有考虑实际上的时间戳）。 本论文提出的方法TiSASRec (Time Interval Aware Self-Attention for Sequential Recommendation), 不仅考虑物品的绝对位置,还考虑序列中物品之间的时间间隔

2. 复现精度

• 目标精度 (NDCG@10：0.5706，Hit@10：0.8038)
• 复现 (NDCG@10: 0.5712, HR@10: 0.8055)

3. 数据集

ml-1m

4. 环境依赖

paddlepaddle-gpu=2.2.0

6. 训练评估

nohup python main.py --dataset=ml-1m --train_dir=default --seed 6 > train.log &

部分日志

epoch:60, time: 186.444842(s), valid (NDCG@10: 0.5910, HR@10: 0.8227), test (NDCG@10: 0.5642, HR@10: 0.7985)
epoch:80, time: 248.875761(s), valid (NDCG@10: 0.5929, HR@10: 0.8232), test (NDCG@10: 0.5693, HR@10: 0.8018)
epoch:100, time: 310.499432(s), valid (NDCG@10: 0.5938, HR@10: 0.8210), test (NDCG@10: 0.5738, HR@10: 0.8060)
epoch:120, time: 372.769209(s), valid (NDCG@10: 0.5985, HR@10: 0.8222), test (NDCG@10: 0.5732, HR@10: 0.8015)
epoch:140, time: 434.958108(s), valid (NDCG@10: 0.5990, HR@10: 0.8227), test (NDCG@10: 0.5716, HR@10: 0.8022)

7. TIPC测试

cd PaddleRec
bash test_tipc/prepare.sh ./test_tipc/configs/tisas/train_infer_python.txt 'lite_train_lite_infer'
bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/tisas/train_infer_python.txt 'lite_train_lite_infer'

8. 精度对齐

python compare.py

# pos_logits的精度误差为2.504792178115167e-07
# neg_logits的精度误差为2.504792178115167e-07
# loss的精度误差为2.384185791015625e-07

引用原论文

@inproceedings{li2020time,
  title={Time Interval Aware Self-Attention for Sequential Recommendation},
  author={Li, Jiacheng and Wang, Yujie and McAuley, Julian},
  booktitle={Proceedings of the 13th International Conference on Web Search and Data Mining},
  pages={322--330},
  year={2020}
}