Research map / Label-efficient and robust learning / Recognition and long-tail
Classifying Emails into Human vs Machine Category
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Combine sender, action, and salutation models with the content model | Content CNN · Editorial test data | Adj-R@P=96%: 70.4% -> 78.8% | Helped |
| Combine sender, action, and salutation models with the content model | Content CNN · Editorial test data | Adj-R@P=90%: 81.1% -> 83.6% | Helped |
| Combine the four models into the full model | Old production logistic regression · Editorial test data | Adjusted recall: 70.5% -> 78.8% | Helped |
| Combine the four models into the full model | Old production logistic regression · Editorial test data | Adjusted precision: 94.7% -> 96.0% | Helped |
| Increase content sequence length from 512 to 2000 (ablation) | Content CNN · Editorial test data | Adj-R@P=90%: 77.8% -> 81.1% | Helped |
| Increase content sequence length from 512 to 2000 (ablation) | Content CNN · Editorial test data | Adj-R@P=96%: 69.5% -> 70.4% | Helped |
| Use Bert (pretrained) instead of Content CNN with s_content=512 | Human/machine email classifier · Editorial test data | Adj-R@P=90%: 77.8% -> 78.5% | Helped |
| Use Bert (pretrained) instead of Content CNN with s_content=2000 | Human/machine email classifier · Editorial test data | Adj-R@P=90%: 81.1% -> 78.5% | Hurt |
| Train one model on all raw features at once instead of training sub-models separately (ablation) | Content CNN with sender address and name inputs · Human/machine email classification | Prediction performance | Hurt |
| Use final sub-model scores with rectification rather than uncorrected scores (ablation) | Full Model · Human/machine email classification | Prediction performance | Helped |
| Use pretrained GloVe embeddings instead of the paper's own dictionary embeddings (ablation) | Content CNN · Human/machine email classification | Prediction performance | No clear effect |
| Pre-train Bert on mail data for 200K steps before task fine-tuning (ablation) | Bert · Human/machine email classification | Prediction performance | Helped |
Closest papers by meaning
- EmFORE: Learning Email Folder Classification Rules by Demonstration (AAAI 2024 · similarity 0.80)
- Extracting Training Data from Diffusion Language Models via Infilling (NeurIPS 2026 · similarity 0.79)
- Human Texts Are Outliers: Detecting LLM-generated Texts via Out-of-distribution Detection (NeurIPS 2025 · similarity 0.79)
- Few-Shot Detection of Machine-Generated Text using Style Representations (ICLR 2024 · similarity 0.79)
- Maximizing Signal in Human-Model Preference Alignment (AAAI 2025 · similarity 0.79)
- Data Shunt: Collaboration of Small and Large Models for Lower Costs and Better Performance (AAAI 2024 · similarity 0.79)
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.