Curriculum Learning for Debiased Recommendation with Explicit and Implicit Feedback
Tianren Fu, Jiawei Chen · Journal of Physics Conference Series · 2023
Abstract The recommender system (RS) has played an increasingly important role in Internet applications. Recent literature on RS mainly focused on better fitting the user behavior data. However, user behavior data is observational, not experimental. This makes for a wide range of biases in the data. In this paper, we introduce a novel framework to combine the advantages of both multi-task and curriculum learning for the debiased recommendation. Unlike existing methods that are limited to specific feedback, our method follows multi-task learning to unify both explicit and implicit feedback. And these two feedbacks are learned in a curriculum learning manner by shifting from explicit to implicit. In this way, our method not only makes better use of the available information in user behavior data but also overcomes the task-balancing problem in multi-task learning. Extensive experiments have been conducted on two real-world datasets and prove that our method delivers state-of-the-art performance and significantly improves the debiasing ability of the recommendation model.