An administrator needs to manage a large catalog of items from various external sellers. The administrator needs to determine if the items should be identified as minimally dangerous, dangerous, or highly dangerous based on their textual descriptions. The administrator already has some items with the danger attribute, but receives hundreds of new item descriptions every day without such classification.

The administrator has a system that captures dangerous goods reports from customer support team or from user feedback.

What is a cost-effective architecture to solve this issue?

A. Build a set of regular expression rules that are based on the existing examples, and run them on the DynamoDB Streams as every new item description is added to the system.

B. Build a Kinesis Streams process that captures and marks the relevant items in the dangerous goods reports using a Lambda function once more than two reports have been filed.

C. Build a machine learning model to properly classify dangerous goods and run it on the DynamoDB Streams as every new item description is added to the system.

D. Build a machine learning model with binary classification for dangerous goods and run it on the DynamoDB Streams as every new item description is added to the system.