Building an embedding-based arXiv paper ranking tool that cuts review time by 85%

Short summary

The author built Paper List, a tool that ranks arXiv papers by relevance using embeddings and cosine similarity against a project description, reducing weekly paper review from 5 hours to 45 minutes. The system learns from saved and dismissed papers to improve recommendations over time. The tool is part of the OpenNomos ecosystem and the author discloses their affiliation.

• •Paper List uses embeddings and cosine similarity to rank arXiv papers by project relevance
• •Reduced weekly paper review from 5 hours to 45 minutes, surfacing 8 relevant papers from 47
• •System improves over time by learning from user save/dismiss actions

Generated with AI, which can make mistakes.