Design metrics to detect harmful content and fraud

## Analytics case: harmful content / fraud detection You are on a social platform that must reduce **harmful content** (e.g., scams, misinformation, harassment) and **fraudulent activity**. ### Your task 1. Propose a **measurement framework** (metrics) to understand the prevalence of harmful content and the effectiveness of mitigation. 2. Describe how you would **detect** harmful content/fraud using data (rules, ML, human review), and how you would evaluate the system. 3. Call out key **data issues** (label delay, bias, false positives) and how you would address them. Assume you have: - User events (views, clicks, messages, reports, blocks) - Moderation actions (removed, downranked, warning) - Human review labels for a sampled set (possibly delayed)