Improving Automated Email Tagging with Implicit Feedback
summary
An Outlook add-in plus a Java backend (EP2) records reads, replies, moves and tag edits and uses three algorithms to turn long email interaction into training signals; a lab study with 15 participants and 330 messages (60 train / 270 test) finds only 16% tag corrections yet implicit signals add much more labeled data, with an implicit feedback threshold of seven working best, about 64% of confirmed labels are correct at low EF and the system cuts error toward the online upper bound and saves 15–25 minutes per day for real users.
AI-written summary · about
chapters (11)
- 0:00Start
- 0:12OUTLINE
- 0:44MOTIVATION
- 3:17THE EMAIL PREDICTOR (UI)
- 4:31EP INSTRUMENTATION
- 4:55ALGORITHMS
- 7:17THE USER STUDY
- 7:59THE EMAIL DATA SET
- 8:51POST-STUDY SIMULATION
- 13:12DEMO/DEPLOYMENT
- 13:39SUMMARY
related talks
12mIncreasing the Speed and Accuracy of Data Labeling Through an AI Assisted Interface
26mBuild AI Systems for Discernment, Not Approval
18mSession: Player Value At Scale: Automating Replies That Actually Help
21mA Human-in-the-Loop System for Sound Event Detection and Annotation
01h 26mShipping AI That Works: An Evaluation Framework for PMs
21mBoosting Data Efficiency with AI: Insights from a Case Study
44mRAG vs Fine tuning which one is better for your AI project
20mEmploying AI to extract actionable insights from qualitative user...