SASRec的实现问题

[zhaijianyang]

你好，关于SASrec的实现，最后的输出是BxH，也就是只用了最后一个item计算loss吗，不应该是seq2seq吗？ output = trm_output[-1] output = self.gather_indexes(output, item_seq_len - 1) return output # [B H]

[BoXiaohe]

Thanks for your attention to RecBole! When calculating loss, sasrec only considers the prediction of the last item in the interaction sequence. You can construct subsequences of the original interaction sequences as additional training data to make it similar to seq2seq. Hope this could help you!