Title: Multi-Objective Field-Aware Factorization for Action-Specific Conversion Modeling Abstract: We introduce a multi-task factorization model that learns distinct feature-interaction patterns for different consumer actions—such as form submissions, purchases, and sign-ups—within a unified embedding space. By sharing representations across conversion types while learning type-specific interaction weights, our approach improves weighted AUC by 0.50% and boosts performance on under-predicted actions by up to 0.84% AUC. The model naturally captures behavioral differences—user-ad interactions drive leads, while publisher-creative pairs influence purchases—enabling more accurate CPA optimization without maintaining separate models per action type.