Neural collaborative filtering recommendation model for de-exposure bias based on fused rewards
In recommendation systems,strong exposure bias caused by sparse interaction data and uneven exposure tends to concentrate recommendations on highly exposed items,neglecting the potential value of low-exposure items,thus limiting user choices and diminishing user experience.To address this issue,this paper proposed a model that integrated neural collabora-tive filtering and the linear upper confidence bound(LinUCB)algorithm to mitigate exposure bias.Firstly,the model used neural collaborative filtering to analyze interaction data between users and items,learning their features and capturing latent preferences.Secondly,it introduced the LinUCB algorithm,embedding its generated reward feature into the neural collabora-tive filtering model to enhance the exploration capabilities for low-exposure items.Finally,experiments conducted on the Mo-vieLens-100K and MovieLens-1M datasets demonstrated that this model increased exposure by approximately 60％compared to traditional neural collaborative filtering models.This enhancement suggests that the proposed method effectively mitigates expo-sure bias and improves both the accuracy and fairness of recommendations,thereby validating the effectiveness of the model.
万方数据
维普
