qinhsiu / icsrec

[WSDM 2024 Oral] This is our Pytorch implementation for the paper: "Intent Contrastive Learning with Cross Subsequences for Sequential Recommendation".

• Hi, nice to see you here 👋

icsrec's Issues

重新训练后无法发现论文结果

你好，我发现使用您给的模型可以很好的得出论文中的结果，但是我自己重新训练后的结果却复现不出来论文中的结果，请问这是什么原因，显卡使用的是您论文中给出的2080ti,pytorch1.7.0

Unfair experimental setting

From the performance comparison (Table 2 in the paper), it is obvious that only DuoRec and ICSRec use CE loss, while other baseline models employ BCE loss. However, models equipped with BCE loss inherently exhibit inferior performance than ones with CE loss. As a result, this unfair comparison may lead to unreliable experimental results.

对比实验里面的指标和原论文指标不一致

你好我在看对比实验的时候发现有些对比模型的精度和原论文不一致比如CoSeRec，CL4Rec，这样的对比是否合理呢，您是把要比较的那些论文代码按照原始论文参数设计在自己环境下重新跑了一遍然后按照这个结果来对比的吗

训练超参数

你好，这篇工作github上面的训练超参数只有Beauty数据集的，能提供其他数据集的训练超参数吗？非常感谢！还是说每个数据集的训练超参数都一样呢。