Future Gradient Descent for Adapting the Temporal Shifting Data Distribution in Online Recommendation Systems

TL;DR

This paper introduces a meta future gradient generator that predicts future data gradients to adapt online recommendation models to temporal distribution shifts, reducing domain generalization error.

Contribution

It proposes a novel meta learning approach for forecasting future gradients, improving adaptation to temporal shifts in online recommendation systems.

Findings

The method achieves lower temporal domain generalization error.

Empirical results show superiority over baseline methods.

Theoretically, it reduces gradient variation in local regret.

Abstract

One of the key challenges of learning an online recommendation model is the temporal domain shift, which causes the mismatch between the training and testing data distribution and hence domain generalization error. To overcome, we propose to learn a meta future gradient generator that forecasts the gradient information of the future data distribution for training so that the recommendation model can be trained as if we were able to look ahead at the future of its deployment. Compared with Batch Update, a widely used paradigm, our theory suggests that the proposed algorithm achieves smaller temporal domain generalization error measured by a gradient variation term in a local regret. We demonstrate the empirical advantage by comparing with various representative baselines.