This paper proposes FairGENRec, a new sequential recommendation model based on the Diffusion Model (DM), to solve the fairness problem of the recommendation system. Existing recommendation systems can excessively learn correlations with sensitive features (such as gender and age) in the process of learning the user's behavioral patterns, which can lead to unfair results. FairGENRec learns by injecting random noise into the original distribution through a model that recognizes sensitive features and reconstructing items using a sequential noise removal model. In addition, it models the fairness of recommendations by injecting various interest information that removes the bias of sensitive user features into the generated results. In the inference stage, noise is added using the user's past interactions and the target item representation is reconstructed through derepresentation. The experimental results using three datasets show that FairGENRec is effective in improving both accuracy and fairness, and the degree of fairness improvement is visualized through case analysis.
