TL;DR: A tool that retrieves semantically similar papers from selected AI venues, including but not limited to recent ICML, ICLR, NeurIPS, CVPR, and over 17,000 ICLR 2026 submissions.
๐ Hugging Face Hosting: https://huggingface.co/spaces/wenhanacademia/ai-paper-finder
๐ NEW: ICLR 2026 accepted papers are updated.
๐ NEW: You can now download search results with one click! Additionally, using the downloaded results, you can download all available PDF files (if a link to the PDF exists in our dataset) with the script batch_download.py under tools. We are working to update our database to include as many pdf links as possible. Downloaded files will be named with the following pattern: <rank> - <affinity_score> - <venue> - <title>.pdf.
๐ Note: PDF links for ICLR 2026 submissions may become invalid during the rebuttal phase, as each new PDF upload from the authors generates a new link. We will update the database to include only accepted papers and update all links then.
Demo
We're still in beta and expanding the list of supported venues!
- There are already some tools avalible, e.g. Paper Digest. However,
- ICLR 2026 submissions are not included yet.
- Most existing tools rely mainly on keyword-based search, whereas we enable searches using full paper abstracts.
- Why abstracts? They contain much richer information than titles or keywords.
You can ACTUALLY build your own Paper Finder in less than 30 minutes. See Tutorial_Making_Paper_Recommenders.ipynb in this repo.
๐ก Support us:
- โญ Star our GitHub repo
- ๐ Share our tool on LinkedIn and X
- โ Buy us a coffee
โจ What's New Compared to ICLR 2026 Paper Finder?
- ๐งญ Expanded Venue Support
- ๐ Multi-lingual Support (performance is suboptimal compared to English)
- ๐ก Upgraded Local Embedding Model (Gemini removed now)
- โก One-click copying and downloading
We are currently:
- Expanding the list of supported venues
- Designing improved user interfaces
- Researching advanced paper-matching algorithms
- Developing multi-agentโsupported search capabilities
- Enabling batch input and batch output
- Exploring researcher-specific personalization features (researcher persona)
Project Lead: Wenhan Gao
Main Contributors: Wenhan Gao, Jingxiang Qu
We are PhD students doing AI research, and itโs hard to afford servers to host this tool.
Weโre currently seeking affordable server options, contact us if you have any suggestions.
This project started as a finder designed specifically for only the ICLR 2026 submissions. We provide the open source implementation as it serves as a good reference for building your own AI Paper Recommender.
Follow along with Tutorial_Making_Paper_Recommenders.ipynb.
Follow along with ICLR2026_Paper_Finder.ipynb.
- Setup the anaconda (skip this if you already have conda)
mkdir -p ~/miniconda3
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda3/miniconda.sh
bash ~/miniconda3/miniconda.sh -b -u -p ~/miniconda3
rm ~/miniconda3/miniconda.sh
source ~/miniconda3/bin/activate- Create a conda enviroment (skip this if you want to just install on your base enviroment)
conda create -n PaperFinder python=3.12
conda deactivate # Make sure no other conda enviroment is activated
conda activate PaperFinder- Install the required python packages
pip install gdown
pip install chromadb
pip install gradio
pip install markdown
pip install google-generativeai
pip install sentence_transformers- Download processed data
gdown https://drive.google.com/uc?id=1RTKWZ4qY4X2mW5BorZOrWTOb2fCipIhr
unzip ICLR2026.zip- Initiate the application
python app.pyAfter running
app.py, Gradio will display a local URL. Simply open it in your web browser.
If you find any bugs or have suggestions, feel free to contact me. My collaborator Jingxiang Qu, my undergraduate mentee Yichi Zhang, and I (and GPT) are actively working on expanding this system, adding support for more venues, improving search mechanisms with specific models, introducing multi-agent support, and introducing new functionalities.
If you're interested in collaborating or contributing, weโd be very happy to hear from you!
Thanks for your interests! You are the (today/total) visitor of this repo.