InfoSeek: Open Data Synthesis For Deep Research

Dataset Information

data/InfoSeek.jsonl Contains the full research tree structures of InfoSeek. Each sample starts from a root node with a research question, its corresponding entity, and process information for sub-questions (stored in root). Also expands into intermediate tree structure during each step of construction (stored in all_tree_list). Totally 52K samples.
data/InfoSeekQA.jsonl A collection of QA pairs derived from InfoSeek. Each entry corresponds to the final question (sample['root']['question']) and its answer entity (sample['root']['entity']) in InfoSeek.jsonl.
data/InfoSeek-Hard-18K.jsonl A challenging subset of InfoSeek (18K samples), which is better to conduct end-to-end RL, identified using an LLM with a dedicated prompt for complex deep research.
data/Trajectory-RFT-17K.jsonl Contains 17K reasoning trajectories generated through the workflow described in our paper. These can be used as training data for supervised fine-tuning (SFT).
data/infoseek_eval.jsonl — InfoSeek Eval The public evaluation split, exposed as the infoseek eval config and test split. It contains 300 high-difficulty, training-isolated Deep Search QA pairs with short, verifiable answers.

InfoSeek Eval

InfoSeek Eval is the first public evaluation release for InfoSeek. It contains 300 English Deep Search Question Answering (DSQA) examples in the following schema:

{"question": "...", "answer": "..."}

InfoSeek Eval contains 300 Deep Search QA examples built from complex Research Trees. Each question connects multiple claims and calls for multi-step verification. We screened the release against the public InfoSeek training data for duplicate questions, answers, and source trees. The final split has 300 distinct question-answer pairs, and no question gives away its answer verbatim.

Use this set only for evaluation. Do not include it in SFT, RFT, RL, or any other training mixture. Models should return the final short answer in <answer>...</answer> tags, following the DSQA evaluation protocol.

from datasets import load_dataset

eval_set = load_dataset("Lk123/InfoSeek", "infoseek eval", split="test")

Release date: 2026-07-13. SHA-256 (data/infoseek_eval.jsonl): ed4010c6d6a2f078c933d6f1f9ac0435a8068713d24056422b35c8aa272ca6f6.

Abstract

Large language models (LLMs) are increasingly expected to go beyond simple factual queries toward Deep Research-tasks that require decomposing questions into sub-problems, coordinating multi-step reasoning, and synthesizing evidence from diverse sources. We formalize Deep Research tasks with verifiable answers as Hierarchical Constraint Satisfaction Problems (HCSPs), which are fundamentally different from single-constraint, multi-hop, or flat CSP formulations. However, existing benchmarks (e.g., Natural Questions, HotpotQA) fail to capture this complexity, while recent synthetic datasets often introduce shortcut reasoning, knowledge leakage, or lack sufficient structural depth. To address this gap, we introduce InfoSeek, a scalable framework for synthesizing complex Deep Research tasks. InfoSeek uses a dual-agent system to recursively build a Research Tree from large-scale webpages, blurring intermediate nodes into valid sub-problems, and converting these trees into natural language questions that require traversing the full hierarchy. It also enables rapid scaling,

Lk123/InfoSeek 已完整同步

InfoSeek: Open Data Synthesis For Deep Research

Dataset Information

InfoSeek Eval

Abstract

7 个文件

数据集版权信息