Comprehensive List Generation for Multi-Generator Reranking

TL;DR

This work shows that it can achieve a more efficient and effective list proposal with a multi-generator framework and provides empirical evidence on two public datasets and online A/B tests and indicates that the proposed framework can further improve the multi-generator reranking performance.

Abstract

Reranking models solve the final recommendation lists that best fulfill users' demands. While existing solutions focus on finding parametric models that approximate optimal policies, recent approaches find that it is better to generate multiple lists to compete for a ``pass'' ticket from an evaluator, where the evaluator serves as the supervisor who accurately estimates the performance of the candidate lists. In this work, we show that we can achieve a more efficient and effective list proposal with a multi-generator framework and provide empirical evidence on two public datasets and online A/B tests. More importantly, we verify that the effectiveness of a generator is closely related to how much it complements the views of other generators with sufficiently different rerankings, which derives the metric of list comprehensiveness. With this intuition, we design an automatic complementary generator-finding framework that learns a policy that simultaneously aligns the users' preferences and maximizes the list comprehensiveness metric. The experimental results indicate that the proposed framework can further improve the multi-generator reranking performance.

Figures (6)

• Figure 1: Intuitive example for the benefit of multi-generator design in reranking task and the motivation of list comprehensiveness. $P_1$, $P_2$, and $P_3$ are different generators. $E$ represents the efficient and accurate list reward estimator that select the best list as exposure. $\mathcal{I}$ and $\mathcal{J}^\ast$ corresponds to the initial list and the optimal list. $q$ denotes the user request.
• Figure 2: Comparison of overviews between standard G-E and the proposed MG-E framework with CLIG extension.
• Figure 3: The comparison result of different settings of $\lambda$ for CLIG extension on MG-E base model.
• Figure 4: Online recommendation workflow and the detailed forward model of the newly added generator in CLIG.
• Figure 5: Change of list selection bias when adding CLIG.