A retail company collects customer comments from social media, its website, and call logs. Each product mention can belong to multiple company-defined categories (labels) that are not mutually exclusive. For example, a mention of “Sample Yogurt” should be labeled as "yogurt", "snack", and "dairy product". Using Amazon Comprehend and needing the fastest delivery, which feature should the team use to train their model?

The correct choice is Custom classification in multi-label mode because the problem explicitly states that each product mention can belong to multiple company-defined categories that are not mutually exclusive. Multi-label classification is designed for scenarios where a single item can be assigned several labels simultaneously. Custom classification in multi-class mode is incorrect because it assumes that each item belongs to exactly one class out of several mutually exclusive classes. Custom entity recognition is used to identify specific entities (like product names or people) within text, not to categorize the text itself. Pre-built (built-in) Comprehend models offer general-purpose classification (e.g., sentiment, topics) but cannot be trained on custom, domain-specific labels like "yogurt" or "dairy product.