SASRec
Self-Attentive Sequential Recommendation
Wang-Cheng Kang and Julian McAuley · ICDM 2018
Read only observed history and predict the next item using position-wise positive/negative training targets.
What you will implement
Causal/query/key padding visibility, stable sampled binary loss and tiny synthetic sequence training; no full recommendation benchmark.
- 67. SASRec: Causal Histories and Sampled Next-Item Loss
Predict the next item from history while isolating future information and excluding padded training targets.
Implement causal/query/key padding visibility and stable position-wise binary loss; encoder and tiny training loop are provided.