Trained email sorter beats simple keyword rules on real messages
The paper compares a trained email sorter, keyword rules and generative AI using real messages from one multilingual inbox.
- Publication
- arXiv
- Stage
- Preprint
- What we read
- Summary of the paper
- Authors
- Federico Lucca
- Universities and research institutions
- Not yet supplied in verified metadata; the Brief does not guess.
What they did and found
In a test of 231 real emails, a basic model trained to recognize word patterns had the best accuracy in most cases. It was faster and cheaper than several AI systems. Information about the sender’s contacts also mattered.
Why it matters
For teams using personal inbox tools, start small and add artificial intelligence only where actions are costly; set clear thresholds to control mistakes.
The trained email sorter performed strongly on real messages from one Italian/English inbox. The separate system using simple keyword rules performed poorly.
Correctly sorting a message does not always mean choosing the right action. The paper also examines information about senders and their contacts.
What remains uncertain
The study uses one personal inbox and two languages. Sorting accuracy alone does not show whether the tool chooses the right actions.
Original sources · 1
- Measure Before You Train: Typed Decision Models vs. a Classical Baseline vs. a Generative LLM for Routing a Personal Multilingual Inbox ↗arXiv · 2026-10-23
Check the original paper for its authors, methods, version and access terms.
What is your take?
Ask a question, add useful context or share a different perspective. Keep the conversation respectful and grounded.