Unifying Ranking and Generation in Query Auto-Completion via Retrieval-Augmented Generation and Multi-Objective Alignment

Query auto-completion is traditionally split into a ranking system and a generative system that are trained and tuned separately. We unify the two with retrieval-augmented generation and align the model against multiple objectives at once, so relevance and generation quality are traded off explicitly rather than implicitly.

arXiv preprint arXiv:2602.01023