Research map / Label-efficient and robust learning / Recognition and long-tail
EmFORE: Learning Email Folder Classification Rules by Demonstration
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Demonstrates EMFORE, an online, rule-learning system (propositional DNF rules, incremental updates, suppression of low-confidence predictions) that learns email folder classification rules from user moves, implemented as an Outlook extension.
Problem
Email folder management requires users to manually write classification rules; existing learning methods do not learn interpretable rules incrementally, require retraining or multiple examples after mistakes, and do not distinguish inbox from other folders.
Approach
An iterative, programming-by-example style learner that maintains a DNF rule per folder over generic email propositions, generalizing/specializing rules after each mistake using greedy selection from a ranked candidate state, and suppressing low-confidence predictions via a sigmoid over five features.
Key results
- EMFORE outperforms prior systems (Dehghani et al. 2016; Carmona-Cejudo et al. 2013) by 20%-25% in correct decision rate on Enron and Avocado in both online and offline setups.
- EMFORE updates rules instantly after every email, even for inboxes with hundreds of emails, per the demo description.
- Suppression of low-confidence predictions reduces false positives while the system can abstain by defaulting to the inbox.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use EMFORE instead of previous email classification systems | Email folder classification · Enron | correct decision rate: 20%-25% | Helped |
| Use EMFORE instead of previous email classification systems | Email folder classification · Avocado | correct decision rate: 20%-25% | Helped |
Limitations
- Demo paper; detailed experimental results deferred to the referenced prior work (Singh et al. 2023a)
- Evaluation is summarized only at a high level in the demo text
Experimental setup
Datasets: Enron Avocado
Benchmarks: Enron Avocado
Metrics: correct decision rate
Baselines: Dehghani, Shakery, and Mirian 2016 Carmona-Cejudo et al. 2013
Builds on: Singh et al. 2023a (EMFORE rule learning by example)
Closest papers by meaning
- Actionable Email Intent Modeling With Reparametrized RNNs (AAAI 2018 · similarity 0.80)
- Classifying Emails into Human vs Machine Category (AAAI 2022 · similarity 0.80)
- Principled Fine-tuning of LLMs from User-Edits: A Medley of Preference, Supervision, and Reward (NeurIPS 2025 · similarity 0.79)
- 2-in-1 Phishing Detection via Large LM Distillation and Small LM Perturbation (Student Abstract) (AAAI 2025 · similarity 0.77)
- Reversing the Forget-Retain Objectives: An Efficient LLM Unlearning Framework from Logit Difference (NeurIPS 2024 · similarity 0.76)
- STAMP Your Content: Proving Dataset Membership via Watermarked Rephrasings (ICML 2025 · similarity 0.76)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.