<scp>MobileDiffRec</scp> : Out‐of‐Distribution Mobile Application Recommendation Based on Causal Diffusion Graph Representation Learning

ABSTRACT With the increase in mobile applications, recommendation methods for mobile applications are rising in importance. However, most of the existing mobile application recommendation systems assume these user–application interaction data are independent and identically distributed, overlooking the fact that multiple external factors might affect users' willingness to use a certain type of application, creating an out‐of‐distribution dataset and forming false causal correlations. Therefore, a Causal Diffusion Graph Representation Learning‐based Out‐of‐Distribution Mobile Application Recommendation method is proposed in this paper. This method uses an environmental generator, environment inference, and diffusion for invariant learning and generalization in an out‐of‐distribution scenario. The validity of the method is verified by comparing it with preexisting methods with similar directions on the Shopify dataset, producing results of 6.36% (Recall) and 6.28% (NDCG) over the best baseline (LightGCL). The results show that the proposed method generalizes well to this domain and effectively mitigates performance degradation due to distribution shifts.