Quality Over Clicks: Iterative Reinforcement Learning for Early-Stage E-Commerce Query Suggestion

The article introduces QualEQS, an iterative reinforcement learning framework designed for e-commerce query suggestion, focusing on quality metrics rather than traditional click-through rate (CTR) optimization. It emphasizes three dimensions of actionable suggestion quality: answerability, factuality, and information gain, and employs group-level disagreement to enhance training from sparse feedback. The framework is validated with the EQS-Benchmark dataset, consisting of 16,949 queries, demonstrating a 6.81% improvement in online performance in a real-world conversational shopping assistant, making it relevant for practitioners seeking effective query suggestion methodologies in early-stage deployments.