Presentation Information
[5H2-OS-18a-03]Designing Surrogate Objective Functions for Maximizing Long-term Matching in Two-Sided Recommendation Systems
〇Naoki Nishimura1, Ken Kobayashi2, Kazuhide Nakata2 (1. Recruit Co., Ltd., 2. Institute of Science Tokyo)
Keywords:
Two-sided Recommendation,Surrogate Objective,Shape Constrained Optimization
In two-sided recommendation systems, over-emphasizing immediate matching probabilities often hinders the exploration of promising yet unknown job opportunities, leading to user disengagement and early attrition.
This study aims to maximize matching probabilities over the total service lifespan. We design a surrogate objective function defined as the weighted sum of current matching probability and the expected future increment. To ensure stable estimation from sparse data, we formulate the future increment estimation as a convex quadratic programming problem with shape constraints. Experiments on synthetic data and online A/B tests demonstrate the effectiveness of our proposed approach.
This study aims to maximize matching probabilities over the total service lifespan. We design a surrogate objective function defined as the weighted sum of current matching probability and the expected future increment. To ensure stable estimation from sparse data, we formulate the future increment estimation as a convex quadratic programming problem with shape constraints. Experiments on synthetic data and online A/B tests demonstrate the effectiveness of our proposed approach.