A Sequence Recommendation Model Based on Causal Inference
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 im-proved.
国家科技期刊平台
NSTL
万方数据
