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