最后更新:11/16/2026最后核验:2026-10-27
PaperScorer 解决投稿论文"值不值得投"的参考问题:把匿名化后的投稿 PDF 上传给系统,它按目标会议的评审模型阅读前 15 页,给出参照该会议真实录用率的校准接收概率,结果分为 Leans Reject、Leans Accept 和 Likely Accept 三档,通过邮件和你自己的结果页链接交付。支持 NeurIPS、ICML、ICLR、CVPR、ACL 等主流会议,单篇最多 20 MB,同一邮箱每天最多提交 5 篇(按 UTC 计)。结果不是论文评审意见:系统只做概率预测,不提供修改建议;若页面或匿名化检查不通过,邮件会说明需要修正的地方。官网用 LoRA、Segment Anything、Mamba 等知名论文的匿名版演示了校准效果,其中部分确实被拒的论文也被正确预测为拒绝,说明打分并非简单跟随论文名气。预测基于自动模型而非官方评审结果,最终录用以会议决定为准。本文依据官方站点核验,未上传论文实测。
-
上传匿名版 PDF,选择目标会议后预测录用概率,以邮件和结果页链接交付
-
概率按该会议真实录用率校准,分 Leans Reject / Leans Accept / Likely Accept 三档
-
覆盖 NeurIPS、ICML、ICLR、CVPR、ICCV、ECCV、ACL、EMNLP、NAACL、AAAI、COLM、AISTATS、CoRL
-
单篇最多 20 MB、只读前 15 页,同一邮箱每天最多提交 5 篇
详情
平台
网页版
功能特性
- Uploads an anonymized paper PDF and predicts its chance of acceptance at the chosen venue
- Calibrates predictions against each venue’s real acceptance rate, reported as Leans Reject, Leans Accept, or Likely Accept
- Supports NeurIPS, ICML, ICLR, CVPR, ICCV, ECCV, ACL, EMNLP, NAACL, AAAI, COLM, AISTATS, and CoRL with venue-specific models
- Emails the prediction plus a bookmarkable result page after re-checking pages and anonymization
支持语言
en
适合人群
Researchers gauging submission risk before polishing and submitting to a top conferenceGraduate students and labs choosing between venues for a finished manuscriptAdvisors triaging a queue of draft papers by predicted acceptance likelihood
已知限制
- Only the first 15 pages of a paper are read, PDFs are capped at 20 MB, and each email may submit up to 5 papers per day (UTC)
- The result is an AI-estimated chance of acceptance, not a peer review, and does not guarantee the real committee decision
- Results arrive by email and the site warns it is running behind, so scoring can take longer than usual
- Requires the anonymized submission version; authors must confirm ownership before scoring
替代工具
Deep Review — Google’s AI reviewer that returns detailed technical feedback instead of an acceptance probability








