SnowPro Specialty: Gen AI Question 142
Single answer2.2 Perform data analysis given a use case.A retail company stores product descriptions, customer reviews, and support tickets in Snowflake. The Gen AI team wants to identify which support topics are driving negative customer sentiment so they can prioritize improvements. They need an approach that can be run directly on data in Snowflake and will scale across millions of text records with minimal data movement. Which approach best fits this use case?
- A
Use Snowflake Cortex AISENTIMENT on reviews and tickets, then aggregate the sentiment results by support topic or category in SQL to identify the most negatively rated themes.
- B
Export all text data to a local Python notebook, manually label sentiment for a sample, and use spreadsheet pivot tables to estimate the most common negative themes.
- C
Use a standard SQL AVG() calculation directly on the text columns containing reviews and tickets to determine which topics have the lowest sentiment values.
- D
Create vector embeddings for all documents and assume the topics with the largest vector dimension values are the most negative topics.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use Snowflake Cortex AISENTIMENT and then analyze the resulting scores with SQL. This matches the stated goals: run analysis directly where the data resides, reduce data movement, and scale across large datasets. In a real-world Snowflake Gen AI workflow, sentiment analysis is often combined with structured metadata such as topic, category, region, or product line to support prioritization decisions. Snowflake documentation describes Cortex AI functions such as sentiment analysis as in-platform capabilities for deriving insights from unstructured text. Best practice is to keep processing close to governed enterprise data in Snowflake and use SQL for downstream aggregation, filtering, and reporting.
- A. Correct.
Correct. AISENTIMENT in Snowflake Cortex is designed to analyze sentiment from text directly in Snowflake. In this scenario, the team can score reviews and support tickets, then use standard SQL grouping and aggregation by topic, product, or ticket category to determine which themes are associated with the most negative sentiment. This approach minimizes data movement and is practical for large-scale analysis.
- B. Incorrect.
Incorrect. Exporting data to a local notebook introduces unnecessary data movement, governance risk, and operational overhead. Manual labeling and spreadsheet analysis do not scale well to millions of records. While sampling can help in exploratory analysis, it is not the best fit when the requirement is to analyze large volumes of data directly in Snowflake.
- C. Incorrect.
Incorrect. AVG() cannot be meaningfully applied directly to raw text columns to derive sentiment. Sentiment must first be inferred through a text analysis function or model. This option reflects a misconception that standard numeric SQL aggregations can replace NLP-based sentiment analysis.
- D. Incorrect.
Incorrect. Embeddings are useful for semantic similarity, clustering, and retrieval tasks, but the raw vector values themselves do not indicate sentiment polarity. Larger or smaller embedding dimensions do not correspond to more negative or positive sentiment. This is a common misunderstanding about how embeddings are used in Gen AI workflows.