AAGenRec: A Novel Approach for Mitigating Inter-task Interference in Multi-task Optimization of Sequential Behavior Modeling

Multi-task optimization is an emerging research field in recommender systems that aims to enhance the recommendation performance across multiple tasks. Various methodologies have been introduced to address aspects like balancing task weights, handling gradient conflicts, and achieving Pareto optimality. These approaches have shown promise in specific contexts, but are not well-suited for real-world scenarios that involve user sequential behaviors. To address this gap, we present AAGenRec, a novel and effective solution for sequential behavior modeling within multi-task recommender systems, inspired by concepts from acoustic attenuation and genetic differences. Specifically, AAGenRec leverages an established genetic distance method to quantify the dissimilarity between tasks, then introduces an impact attenuation mechanism to mitigate the uncertain task interference in multi-task optimization. Extensive experiments conducted on public e-commerce datasets demonstrate the effectiveness of AAGenRec.