Lessons Learned Addressing Dataset Bias in Model-Based Candidate Generation at Twitter

TL;DR

This paper investigates dataset bias issues in model-based candidate generation for recommender systems, demonstrating how random sampling and fine-tuning can improve performance in large-scale applications like Twitter.

Contribution

It identifies the limitations of existing bias correction methods and proposes effective random sampling and fine-tuning techniques for better candidate generation.

Findings

Random sampling mitigates dataset bias effectively.

Fine-tuning improves candidate generation performance.

Proposed methods enhance Twitter's timeline recommendations.

Abstract

Traditionally, heuristic methods are used to generate candidates for large scale recommender systems. Model-based candidate generation promises multiple potential advantages, primarily that we can explicitly optimize the same objective as the downstream ranking model. However, large scale model-based candidate generation approaches suffer from dataset bias problems caused by the infeasibility of obtaining representative data on very irrelevant candidates. Popular techniques to correct dataset bias, such as inverse propensity scoring, do not work well in the context of candidate generation. We first explore the dynamics of the dataset bias problem and then demonstrate how to use random sampling techniques to mitigate it. Finally, in a novel application of fine-tuning, we show performance gains when applying our candidate generation system to Twitter's home timeline.