Apppublic

ai-conferences/conference-paper-search

sourceHugging Faceupdated 17h agoView on Hugging Face

5likes

App README

Conference Paper Search

Semantic and keyword search over accepted papers from ML conferences, built on `ai-conferences/all-papers` (one row per accepted paper across ICLR, NeurIPS, ICML, CVPR, ICCV, ECCV, ACL, EMNLP, NAACL, COLM, ICRA, SIGGRAPH, and SIGGRAPH Asia).

How it works:

—Semantic mode encodes the query with BAAI/bge-base-en-v1.5 on CPU and runs a brute-force cosine search over the precomputed paper embeddings.
—Keyword mode is a substring match over titles, authors, and abstracts.
—HF paper-page stats (upvotes, code and project links) are joined from `cfahlgren1/hub-stats` at startup.
—GitHub star counts are joined from `hysts-bot-data/paper-github-stars`, which a scheduled HF Job refreshes daily.

The frontend is a custom HTML/JS page served by gradio.Server; the search API is exposed as Gradio endpoints (/facets, /search), which are also served as MCP tools at /gradio_api/mcp/.

Dependencies are managed with uv (pyproject.toml + uv.lock) and exported:

bash

uv export --no-hashes --no-dev --emit-index-url --no-emit-package typer-slim --no-emit-package spaces -o requirements.txt