融合序列增广与表征增强的会话推荐OA

The performance of session-based recommendation is inherently limited by the high sparsity of anonymous interaction behavior sequences.Although existing methods have continuously evolved in modeling architectures,they still struggle to accurately learn user preferences from limited contextual information.To address this,SeaNoe(Sequence interaction augmentation and Node representation enhancement)framework is proposed to improve session-based recommendation,which alleviates session sparsity jointly by improving session density and node richness.Regarding session density,virtual interaction nodes are inserted into sparse sequences under the guidance of an association rule graph coupled with an intervention effect feedback mechanism,thereby directly mitigating the problem of insufficient interactions.Regarding node richness,semantic features of item texts are extracted via a pre-trained language model and then aligned and fused with identification features carrying collaborative information,so as to enrich node information to indirectly compensate for the information scarcity caused by interaction sparsity.Experiments on real-world datasets demonstrate that SeaNoe,as a plug-and-play framework,effectively enhances the recommendation performance of multiple base models,with an average improvement ratio of 9.89%and a maximum improvement ratio of 29.31%.