An ML engineer must extract meaningful, unique keywords from documents using AWS services with the least operational management. Which approach meets this requirement?

The correct answer is to use Amazon Comprehend custom entity recognition and key-phrase extraction. Amazon Comprehend is a fully managed natural language processing (NLP) service that can identify key phrases and entities in text, requiring minimal operational overhead. Its key-phrase extraction feature directly addresses the need for extracting meaningful keywords. Custom entity recognition allows for identifying unique, domain-specific terms. Running NLTK on EC2 instances or using Lambda functions with custom Python scripts for stemming, stop-word removal, and n-grams (options 1 and 3) involves significant operational management, including server provisioning, patching, and code maintenance, which goes against the "least operational management" requirement. Using Amazon SageMaker with BlazingText and TF-IDF (option 2) also requires more hands-on management for model training, deployment, and custom preprocessing logic compared to the fully managed Comprehend service.