Dataset Viewer
Auto-converted to Parquet Duplicate
domain
stringclasses
2 values
messages
listlengths
3
3
paper_id
stringlengths
10
18
ml
[ { "content": "You are a research scientist skilled at synthesizing ideas from existing literature into novel research proposals. You are given a set of related research papers (with titles and abstracts). Your task is to analyze these papers, identify research gaps and opportunities, and propose a novel researc...
03OkC0LKDD
ml
[ { "content": "You are a research scientist skilled at synthesizing ideas from existing literature into novel research proposals. You are given a set of related research papers (with titles and abstracts). Your task is to analyze these papers, identify research gaps and opportunities, and propose a novel researc...
01ahsMovBx
ml
[ { "content": "You are a research scientist skilled at synthesizing ideas from existing literature into novel research proposals. You are given a set of related research papers (with titles and abstracts). Your task is to analyze these papers, identify research gaps and opportunities, and propose a novel researc...
0Az25lvdT2
ml
[ { "content": "You are a research scientist skilled at synthesizing ideas from existing literature into novel research proposals. You are given a set of related research papers (with titles and abstracts). Your task is to analyze these papers, identify research gaps and opportunities, and propose a novel researc...
0BS36re3Cx
ml
[ { "content": "You are a research scientist skilled at synthesizing ideas from existing literature into novel research proposals. You are given a set of related research papers (with titles and abstracts). Your task is to analyze these papers, identify research gaps and opportunities, and propose a novel researc...
0A9f2jZDGW
ml
[ { "content": "You are a research scientist skilled at synthesizing ideas from existing literature into novel research proposals. You are given a set of related research papers (with titles and abstracts). Your task is to analyze these papers, identify research gaps and opportunities, and propose a novel researc...
0TD3eO46gk
ml
[ { "content": "You are a research scientist skilled at synthesizing ideas from existing literature into novel research proposals. You are given a set of related research papers (with titles and abstracts). Your task is to analyze these papers, identify research gaps and opportunities, and propose a novel researc...
0LZRtvK871
ml
[ { "content": "You are a research scientist skilled at synthesizing ideas from existing literature into novel research proposals. You are given a set of related research papers (with titles and abstracts). Your task is to analyze these papers, identify research gaps and opportunities, and propose a novel researc...
00SnKBGTsz
ml
[ { "content": "You are a research scientist skilled at synthesizing ideas from existing literature into novel research proposals. You are given a set of related research papers (with titles and abstracts). Your task is to analyze these papers, identify research gaps and opportunities, and propose a novel researc...
004uTlSufe
ml
[ { "content": "You are a research scientist skilled at synthesizing ideas from existing literature into novel research proposals. You are given a set of related research papers (with titles and abstracts). Your task is to analyze these papers, identify research gaps and opportunities, and propose a novel researc...
06lrITXVAx
ml
[ { "content": "You are a research scientist skilled at synthesizing ideas from existing literature into novel research proposals. You are given a set of related research papers (with titles and abstracts). Your task is to analyze these papers, identify research gaps and opportunities, and propose a novel researc...
0ctvBgKFgc
End of preview. Expand in Data Studio

IdeaAnchor

IdeaAnchor is a dataset for research idea generation, with training data from machine learning (ml) and Nature Communications (nc), and a shared evaluation set of 924 ML papers.

Each source record includes a proposal-style formulation of the paper's idea, prior works, and evaluation rubrics.

Configurations

  • raw: Source records with reference proposals and rubrics.
  • sft: Chat-format training examples derived from reference proposals.
  • rl: Prompts and rubrics for reinforcement learning.
  • ssd: Chat-format training examples generated by Qwen3-8B.
  • eval: Evaluation prompts and grading metadata.
  • eval_reference: Full evaluation reference records.

The first four configurations use the train split; the last two use eval.

Fields

  • domain: Source domain, ml or nc.
  • paper_id: Identifier for the source paper.
  • paper_url: Original paper PDF URL in raw and eval_reference.
  • messages: Chat messages; SFT and SSD include an assistant target.
  • prior_works: Related papers with titles and abstracts.
  • proposal / proposal_gt: Reference motivation and method text.
  • rubrics: Evaluation criteria, stored as a JSON string in rl.
  • source_model: Model used to generate SSD targets.

Fields vary by configuration.

Usage

from datasets import load_dataset

train = load_dataset("idealand/IdeaAnchor", "sft", split="train")
eval_data = load_dataset("idealand/IdeaAnchor", "eval", split="eval")
Downloads last month
5