Scientific information extraction (SciIE) is a key bottleneck for turning unstructured papers into computable knowledge bases, yet most existing systems still follow a “local extraction then global assembly” paradigm. This workflow is inherently lossy: by extracting fields in isolation, it breaks global correlations and discards high-confidence signals that could otherwise be reused as internal supervision, forcing systems to repeatedly restart from scratch, especially in long, multimodal scientific documents. In this paper, We propose a different view: SciIE should be solved as a progressive filling problem, similar to solving a Sudoku,once a field is filled with high confidence, it should act as a constraint that guides the remaining uncertain fields. Based on this idea, we introduce SudokuFill, a multi-agent framework that maintains a Global Filling State and performs priority scheduling to establish reliable anchors first, then reuses them as internal supervision for iterative deliberation over harder fields. Evaluated on a specialized document-level adjuvant dataset, our framework achieves a SOTA score of 51.83% on our benchmark. Crucially, SudokuFill enables a 7B model to outperform the vanilla GPT-4o, proving that structured architectural reasoning can effectively compensate for parameter scale.

Empowering machines to understand scientific literature is crucial for accelerating scientific discovery and advancing the AI for Science (AI4S) paradigm. In this paper, we present a comprehensive survey of datasets serving this domain. We propose a systematic taxonomy that organizes resources spanning structural understanding, text understanding, multimodal understanding and pre-training/instruction fine-tuning. Beyond a structured overview, we discuss the evolution of the field, elucidating how the emergence of Large Language Models (LLMs) has reshaped research priorities of dataset construction. By synthesizing existing datasets and identifying critical future directions, this work provides a roadmap for advancing intelligent scientific research systems.

Automatic scientific survey generation requires a reference set that can support broad topic coverage and citation-worthy reference selection. However, existing systems often approximate this step as one-shot relevance retrieval from a target survey topic, abstract, or generated queries, which may overemphasize papers close to the initial topic wording and miss references needed for broader survey construction. We propose Berry-LR, a literature retrieval framework that models survey reference collection as a berrypicking process. Given a target survey topic, Berry-LR performs move-based search broadening, uses high-confidence retrieved papers as anchors for multi-round retrieval, and records retrieval trajectories as graph evidence for Trajectory-aware Reference Selection. Experiments on SurGE and SurveyBench show that Berry-LR improves reference retrieval, final reference selection, and downstream reference coverage over retrieval and automatic scientific survey generation baselines. Ablations and further analyses suggest that these gains come from move-based search broadening, anchor feedback, and trajectory-aware selection, while many remaining misses are tied to section-local writing contexts.