Enable DynamoDB global tables in multiple Regions and use on-demand capacity to absorb unpredictable traffic is the correct choice because it provides active multi Region replication and local reads and writes with low latency while scaling automatically to match unpredictable demand.

Global tables deliver fully active multi Region replication so users get local read and write performance in each Region and the system continues operating during a Regional disruption. On-demand capacity charges per request and removes the need to predict traffic so it is cost effective for spiky and hard to forecast workloads and it reduces operational overhead from capacity planning.

Persist user state in Amazon S3 with Cross-Region Replication and invoke AWS Lambda to update data in real time is unsuitable because S3 is object storage and it does not provide millisecond transactional access patterns for user graphs and interactions and using Lambda to emulate database semantics adds latency and complexity.

Configure DynamoDB global tables across multiple Regions using provisioned capacity with Auto Scaling for traffic bursts offers the right active multi Region topology but provisioned capacity can require careful tuning and it can be more expensive or lead to throttling when traffic is highly unpredictable compared to on demand.

Run a single-Region DynamoDB table in provisioned capacity mode and replicate asynchronously to a secondary Region with DynamoDB Streams for disaster recovery increases operational complexity and it does not provide active multi Region writes so users far from the primary Region face higher latency and failover is more complex.

Full AWS Practitioner Certification Question