SnowPro Associate: Platform Question 327
Single answer● CLASSIFY_TEXT functionA support operations team stores incoming customer email bodies in a Snowflake table named SUPPORT_EMAILS(email_id NUMBER, email_text STRING). They want to use Snowflake Cortex to automatically assign each email to one of these categories for routing: BILLING, TECHNICAL_SUPPORT, ACCOUNT_ACCESS, or OTHER. The team wants a SQL-only solution that returns one label per row and does not require model training. Which approach best meets this requirement?
- A
Use CLASSIFY_TEXT(email_text, ['BILLING','TECHNICAL_SUPPORT','ACCOUNT_ACCESS','OTHER']) in a SELECT statement to return the best matching label for each email.
- B
Use COMPLETE with a prompt listing the categories, because CLASSIFY_TEXT requires a pre-trained custom model before it can classify text.
- C
Use SENTIMENT on email_text and map positive, neutral, and negative results to the four routing categories.
- D
Use EMBED_TEXT_768 on email_text, because embeddings directly return the final business category label without any additional logic.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use CLASSIFY_TEXT because it is specifically designed to assign text to one of a set of provided labels using SQL, which aligns with the team's need for a SQL-only, no-training solution. In Snowflake Cortex, CLASSIFY_TEXT is appropriate when you already know the target categories and want a straightforward classification result per row. By contrast, COMPLETE is a general-purpose text generation function and would be less deterministic and less purpose-built for this task. SENTIMENT measures emotional polarity rather than business topic, and EMBED_TEXT_768 returns embeddings that require additional processing before a category can be inferred. This reflects Snowflake best practice of choosing the most specialized Cortex function for the task: classification for label assignment, sentiment for tone analysis, completion for generation, and embeddings for semantic representation and similarity workflows.
- A. Correct.
Correct. CLASSIFY_TEXT is intended for classifying input text into one label from a provided set of categories, making it a strong fit for routing support emails into predefined business classes. It is a SQL function and does not require the user to train a custom model. This matches the requirement for a SQL-only approach that produces one label per row.
- B. Incorrect.
Incorrect. COMPLETE can be used for generative prompting, but the statement about CLASSIFY_TEXT requiring a pre-trained custom model is false. CLASSIFY_TEXT is designed to classify text into supplied labels without customer-managed model training. While COMPLETE might be used to attempt similar behavior, it is not the most direct or purpose-built choice for this requirement.
- C. Incorrect.
Incorrect. SENTIMENT determines emotional tone such as positive or negative; it does not classify content into operational categories like BILLING or ACCOUNT_ACCESS. A candidate might choose this if they confuse text analytics tasks, but sentiment analysis does not solve routing into these business-specific labels.
- D. Incorrect.
Incorrect. EMBED_TEXT_768 generates vector embeddings, which are numerical representations of text, not final category labels. To use embeddings for classification, additional workflow steps would be needed, such as similarity search or downstream ML logic. That does not meet the stated requirement as directly as CLASSIFY_TEXT.