Question 160
At fictional Hillside Cooperative Bank, a Snowflake data and AI engineering team is conducting a source-system modernization and AI search review using only first-party public technical references and independently written facts. The separately composed technical incident is: A service team interprets every model-generated negative sentiment as evidence of dissatisfaction, but its dataset contains sarcastic or mixed-language customer messages that produce misleading labels. Assume only these documented facts and use safe, authorized production-change procedures; exact current Snowflake feature syntax and region or account availability require current review. Which repeatable assurance or control is most relevant to preventing this situation?
Explanation
Writer-proposed answer: D (UNVERIFIED)
Preliminary individual option explanations — NOT expert validated
Correct Answer
Use periodic labeled audits, language coverage tests and human verification for consequential routing decisions. Sentiment inference is probabilistic and context sensitive; a numerical or categorical label is not a direct observation of a person's intent.
Incorrect Answers
This response concerns Verified query repository for business metric questions, which is different from the stated failure. A service team interprets every model-generated negative sentiment as evidence of dissatisfaction, but its dataset contains sarcastic or mixed-language customer messages that produce misleading labels.
Sensitive translation with regional constraints is insufficient to demonstrate this result: Reviewed benchmark samples show lower misinterpretation rates and no automatic adverse customer action solely from uncertain sentiment classification.
Similarity threshold excludes rare queries is not the cause of the reported failure: A service team interprets every model-generated negative sentiment as evidence of dissatisfaction, but its dataset contains sarcastic or mixed-language customer messages that produce misleading labels.