Can Large Language Models Forecast What Researchers Study Next?
Paper • 2609.00747 • Published
arxiv_id stringlengths 10 10 | month stringdate 2023-01-01 00:00:00 2024-11-01 00:00:00 | title stringlengths 0 484 | text stringlengths 3.85k 1.94M |
|---|---|---|---|
2301.00004 | 2023-01 | SESNet: sequence-structure feature-integrated deep learning method for data-efficient protein engineering | # SESNet: sequence-structure feature-integrated deep learning method for data-efficient protein engineering
Mingchen $\mathrm{Li^{1,4^{\dag}}}$ , Liqi Kang1,2†, Yi Xiong5, Yu Guang Wang1, Guisheng
Fan4, Pan Tan1\*, Liang Hong1,2,3\*
1. Shanghai National Center for Applied Mathematics (SJTU Center), & Institute... |
2301.00006 | 2023-01 | Recovering Top-Two Answers and Confusion Probability in Multi-Choice Crowdsourcing | # Recovering Top-Two Answers and Confusion Probability in Multi-Choice Crowdsourcing
Hyeonsu Jeong 1 Hye Won Chung
# Abstract
Crowdsourcing has emerged as an effective platform for labeling large amounts of data in a costand time-efficient manner. Most previous work has focused on designing an efficient algori... |
2301.00007 | 2023-01 | Selected aspects of complex, hypercomplex and fuzzy neural networks | "# Selected aspects of complex, hypercomplex and fuzzy neural networks \n\nedited by Agnieszka Niem(...TRUNCATED) |
2301.00008 | 2023-01 | Effects of Data Geometry in Early Deep Learning | "# Effects of Data Geometry in Early Deep Learning \n\nSaket Tiwari Department of Computer Science (...TRUNCATED) |
2301.00011 | 2023-01 | eVAE: Evolutionary Variational Autoencoder | "# eVAE: Evolutionary Variational Autoencoder \n\nZhangkai Wu,1 Longbing Cao, 1 Lei Qi 2 \n\n1 Uni(...TRUNCATED) |
2301.00012 | 2023-01 | GANExplainer: GAN-based Graph Neural Networks Explainer | "# GANExplainer: GAN-based Graph Neural Networks Explainer \n\nYiqiao Li, Jianlong Zhou, Boyuan Zhe(...TRUNCATED) |
2301.00014 | 2023-01 | Time series Forecasting to detect anomalous behaviours in Multiphase Flow Meters | "# Time series Forecasting to detect anomalous behaviours in Multiphase Flow Meters \n\nT. Barbario(...TRUNCATED) |
2301.00015 | 2023-01 | Self-organization Preserved Graph Structure Learning with Principle of Relevant Information | "# Self-organization Preserved Graph Structure Learning with Principle of Relevant Information \n\n(...TRUNCATED) |
2301.00032 | 2023-01 | Bayesian Learning for Dynamic Inference | "# Bayesian Learning for Dynamic Inference \n\nAolin Xu Peng Guan \n\n# Abstract \n\nThe traditio(...TRUNCATED) |
2301.00036 | 2023-01 | "Modified Query Expansion Through Generative Adversarial Networks for Information Extraction in E-Co(...TRUNCATED) | "# Modified Query Expansion Through Generative Adversarial Networks for Information Extraction in E-(...TRUNCATED) |
End of preview. Expand in Data Studio
IdeaForecastBench corpus
The paper corpus behind IdeaForecastBench (Can Large Language Models Forecast What Researchers Study Next?, arXiv:2609.00747, EMNLP 2026): 108,768 arXiv computer-science papers from 2023-01 to 2025-10, converted from PDF to Markdown with MinerU, one file per paper.
This is exactly the corpus the paper's experiments loaded. The benchmark's 624 rolling episodes use the 2024-04 to 2025-09 slice (twelve monthly cutoffs from 2024-07, three-month horizon); the earlier months feed MDF's training episodes.
Layout
One Parquet file per month, columns arxiv_id, month, title, text (the full
paper as Markdown). manifest.json lists paper count, size and sha256 per file.
from datasets import load_dataset
ds = load_dataset("4R5T/idea-forecast-bench", split="train")
ds[0]["title"], ds[0]["text"][:200]
Use with the benchmark
git clone https://github.com/social-world-model/idea-forecast-bench && cd idea-forecast-bench && poetry install
idea-forecast-bench fetch --from-hf 4R5T/idea-forecast-bench --out-dir data/csml/raw_markdown
idea-forecast-bench baselines --input-dir data/csml/raw_markdown
fetch --from-hf downloads the Parquet files and writes each paper back out as
<YYYY-MM>/<arxiv_id>.md, the layout the benchmark's corpus loader reads.
License
Each paper remains under the license its authors chose on arXiv; see the arXiv license page linked above. The Markdown conversions are provided for research use.
Citation
@misc{li2026ideaforecastbench,
title = {Can Large Language Models Forecast What Researchers Study Next?},
author = {Fenghai Li and Zihan Tang and Haofei Yu and Yining Zhao and Jiaxuan You},
year = {2026},
eprint = {2609.00747},
archivePrefix = {arXiv},
primaryClass = {cs.CL},
url = {https://arxiv.org/abs/2609.00747}
}
- Downloads last month
- 279