The correct option is Unsupervised Learning.

This approach is designed for situations where you have large volumes of unlabeled text and you want the algorithm to discover structure on its own. With millions of chat transcripts and reviews, techniques like clustering and topic modeling can surface natural groupings and recurring themes without any predefined labels or guidance. This aligns directly with the goal of mining untagged data to uncover patterns.

Deep Learning is a class of model architectures rather than a learning paradigm. It can be used in supervised or unsupervised contexts, so naming it alone does not address the requirement to find patterns in unlabeled data.

Reinforcement Learning focuses on training an agent to take actions in an environment to maximize cumulative reward. It is not suited to discovering latent structure in a static corpus of text.

Supervised Learning requires labeled examples that map inputs to known targets. The scenario explicitly lacks labels and seeks emergent themes, so this paradigm does not fit.

Full AWS Practitioner Certification Question