The Content Optimizer executes a two-stage acquisition process that balances exploration across Rank Factor Classes via a freshness weight before selecting specific claims using Thompson sampling, Upper Confidence Bound (UCB), or epsilon-greedy selection.
flowchart LR
A[All Active Claims] --> B["Stage 1: Sample Factor Class<br/>Freshness ~ 1 / (1 + n)^w"]
B --> C[Filtered Claims in Class]
C --> D["Stage 2: Acquisition Policy<br/>Thompson / UCB / eps-Greedy"]
D --> E[Selected Claim & Theta]Referenced by