Summary
Finding the best set of documents to answer complex questions is hard because just picking individually relevant documents isn’t enough—they need to work well together. The authors develop AdaTutoRank, a system that learns to better rank whole sets of documents by giving more detailed guidance based on how good previous attempts were. It uses a special training method called Adaptive Tutoring Optimization to provide different levels of hints, helping the system learn which documents truly contribute to good answers versus which are just repetitive. This lets AdaTutoRank outperform existing methods while needing fewer document retrievals.
What this means in practice
- •For search engine developers: Enhance document selection for search results by using adaptive hints to better compose relevant and complementary document sets.
- •For knowledge management teams: Improve deep research tools to provide more complete and non-redundant information sets by learning from detailed setwise feedback.
Authors
Kailin Jiang, Lei Liu, Jian Xi, Yangqi Chen, Hui Xu, Hongwei Zhao, Bin Li, Yu Lu, Haibo Shi
Abstract
Document rerankers determine what evidence reaches the downstream model in RAG and deep research, yet mainstream rerankers select by relevance matching, and individually relevant documents rarely constitute the complete, complementary, non-redundant set a complex information need demands. Prior work rewards a set by its aggregate rubric score, shifting the objective from ranking documents to composing sets. Yet that score is one scalar shared by every document in the set, so the supervision is sparse: a redundant document is rewarded with the rest whenever the set scores well, and a decisive one penalized with the rest whenever it does not; credit assignment leaves contributors indistinguishable from free riders. On-policy distillation could densify this supervision, but existing methods give every rollout the same fixed guidance, too prescriptive for strong rollouts and too abstract for weak ones. We therefore propose AdaTutoRank, a setwise reranker trained with Adaptive Tutoring Optimization (ATO) under a three-level hierarchy of nine rubric dimensions, which supplies silver labels for the cold start, rewards for reinforcement learning, and hints for distillation. ATO draws three hint forms of increasing specificity from the policy's own frozen snapshot: the rubrics alone, a self-selector's sibling-set chosen under rubrics, and a self-reflector's reflection contrasting the rollout with that sibling-set; each rollout receives the form matched to its quality. Re-scoring that rollout under the hint-conditioned frozen teacher and the hint-free snapshot distills the hint's effect into a token-level advantage that complements the group-relative outcome advantage. Across ten benchmarks spanning RAG, deep research, and setwise evaluation, AdaTutoRank attains the best overall performance while issuing fewer retrieval calls.