Advice techniques in multi-stakeholder environments usually require optimizing for a number of goals concurrently to satisfy provider and shopper calls for. Serving suggestions in these settings depends on effectively combining the goals to handle every stakeholder’s expectations, usually by way of a scalarization perform with pre-determined and glued weights. In follow, choosing these weights turns into a consequent downside. Current work has developed algorithms that adapt these weights primarily based on application-specific wants by utilizing RL to coach a mannequin. Whereas this solves for computerized weight computation, such approaches should not environment friendly for frequent weight adaptation. Additionally they don’t enable for human intervention oftentimes decided by enterprise wants. To bridge this hole, we suggest a novel multi-objective suggestion framework that’s environment friendly for a small variety of goals. It additionally allows enterprise resolution makers to simply tune the optimization by assigning completely different significance to a number of goals. We display the efficacy and effectivity of our framework by way of enhancements in on-line enterprise metrics.
