一种基于因果推断的序列推荐模型

A Sequence Recommendation Model Based on Causal Inference

针对推荐模型由于训练数据与测试数据分布不一致(即分布外情况)而导致的性能下降问题,提出了一种基于因果推断的优化后序列推荐模型─CCSRec(Counterfactual Context Sequential Recommendation)。利用后门调整和虚事实方法,去除用户与物品之间的虚假相关性,降低由于分布外情况对推荐性能的不良影响。CCSRec通过场景数据替换,生成虚事实环境提升模型对场景数据的学习能力,有效减缓原模型在分布外情况下性能下降的问题。CCSRec模型考虑了用户信息的参数化,将用户信息纳入计算范围,进一步提升模型的推荐效果。在多个公开推荐数据集上的实验结果表明,CCSRec模型通过场景数据替换进行虚事实序列生成方法进一步降低了分布外情况对性能的不良影响,性能得到明显提升。

In response to the performance degradation issue in recommendation models caused by the inconsistency between training and testing data distributions,i.e.,out-of-distribution(OOD)scenarios,we propose a novel optimized sequential recommendation model called CCSRec(Counterfactual Context Sequential Recommendation)based on causal inference.By leveraging backdoor adjustment and counterfactual methods,CCSRec effectively removes spurious correlations between users and items,thus reducing the adverse influence of OOD on recommendation performance.CCSRec generates counterfactual environments by replacing scenario data to enhance the model’s learning ability for contextual data,effectively mitigating the performance degradation problem under OOD scenarios.Additionally,CCSRec incorporates user information in its parameterization to further enhance recommendation effectiveness.Experimental results on multiple publicly recommended datasets indicate that the CCSRec model further reduces the adverse effects of out-of-distribution situations on performance by use of scenario data replacement for counterfactual sequence generation,and performance is significantly improved.