Generative Query Suggestion via Intent Coverage and Query-Level Credit Assignment

The paper introduces a generative query suggestion framework that focuses on covering distinct user intents while ensuring each suggested query is useful. It employs a two‑stage optimization: first, intent‑aware diversity modeling creates supervised fine‑tuning data and a reward that encourages intent coverage; second, query‑level credit assignment directs quality signals to individual query tokens while sharing a slate‑level diversity signal. Experiments on a large production dataset, including online A/B tests, demonstrate gains in click‑through rate, query quality, and intent coverage.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

• fine-tuning