Understanding Customer Reviews using Snowflake Cortex

SQL as IMPORT_DATA_SQL


USE ROLE SNOWFLAKE_LEARNING_ROLE;

-- use the existing database and schema
USE DATABASE SNOWFLAKE_LEARNING_DB;

SET schema_name = CONCAT(current_user(), '_CUSTOMER_REVIEW_ANALYSIS_WITH_AISQL');

CREATE SCHEMA IF NOT EXISTS IDENTIFIER($schema_name);

USE SCHEMA IDENTIFIER($schema_name);

/*--
• file format and stage creation
--*/

CREATE OR REPLACE FILE FORMAT csv_ff

  TYPE = 'csv'
  SKIP_HEADER = 1;

CREATE OR REPLACE STAGE s3load

  COMMENT = 'Quickstarts S3 Stage Connection'
  URL = 's3://sfquickstarts/misc/aisql/ecommerce_customer_review/'
  FILE_FORMAT = csv_ff;


/*-- • raw zone table build --*/


CREATE OR REPLACE TABLE customer_data
(
  CUSTOMER_ID VARCHAR(16777216),
  CUSTOMER_SEGMENT VARCHAR(16777216),
  JOIN_DATE DATE,
  LIFETIME_VALUE NUMBER(38,2),
  PREVIOUS_PURCHASES NUMBER(38,0),
  AGE_RANGE VARCHAR(16777216),
  GENDER VARCHAR(16777216),
  PREFERRED_CATEGORY VARCHAR(16777216)
);

CREATE OR REPLACE TABLE product_catalog
(
  PRODUCT_ID VARCHAR(16777216),
  PRODUCT_NAME VARCHAR(16777216),
  CATEGORY VARCHAR(16777216),
  SUBCATEGORY VARCHAR(16777216),
  MANUFACTURER VARCHAR(16777216),
  PRICE NUMBER(38,2),
  RELEASE_DATE DATE,
  REVIEW_COUNT NUMBER(38,0)
);

CREATE OR REPLACE TABLE product_reviews
(
  REVIEW_ID VARCHAR(16777216),
  PRODUCT_ID VARCHAR(16777216),
  CUSTOMER_ID VARCHAR(16777216),
  REVIEW_TEXT VARCHAR(16777216),
  RATING NUMBER(38,0),
  REVIEW_DATE DATE,
  PURCHASE_DATE DATE,
  VERIFIED_PURCHASE BOOLEAN,
  HELPFUL_VOTES NUMBER(38,0)
);

/*-- • raw zone table load --*/

COPY INTO customer_data
FROM @s3load/customer_data.csv
ON_ERROR = CONTINUE;

COPY INTO product_catalog
FROM @s3load/product_catalog.csv
ON_ERROR = CONTINUE;

COPY INTO product_reviews
FROM @s3load/product_reviews.csv
ON_ERROR = CONTINUE;

-- setup completion note
SELECT 'Setup is complete' AS note;

-- Quick preview of the table
SELECT *
FROM product_reviews
LIMIT 25;

SQL as CHECK_DATA_SQL

-- Quick preview of the table 
SELECT *
FROM product_reviews
LIMIT 25;

SQL as SENTIMENT_CHECK_SQL

WITH EXTRACTED_SENTIMENT AS (
    SELECT 
        RATING,
        SNOWFLAKE.CORTEX.SENTIMENT(REVIEW_TEXT) AS SENTIMENT
    FROM PRODUCT_REVIEWS
)
SELECT CORR(SENTIMENT, RATING) AS SENTIMENT_RATING_CORRELATION
FROM EXTRACTED_SENTIMENT;

SELECT 
  AI_AGG(
    REVIEW_TEXT, 
    'What are the top 3 most common product issues reported in these reviews?'
  ) AS TOP_ISSUES
FROM PRODUCT_REVIEWS pr
JOIN PRODUCT_CATALOG pc ON pr.product_id = pc.product_id
WHERE pc.category = 'Electronics'
AND SNOWFLAKE.CORTEX.SENTIMENT(REVIEW_TEXT) < 0;

# to view the result
df = AGG_TOP_ISSUES_SQL.to_pandas()
print(df['TOP_ISSUES'].iloc)

-- Create temporary table on the filtered result to be re-used in next step analytics.
CREATE OR REPLACE TEMP TABLE filtered_product_reviews AS
SELECT *
FROM product_reviews
WHERE AI_FILTER(PROMPT('This review mentions a product issue or complaint: {0}', review_text));

-- Leverage AI_AGG functions to find the common issues mentioned.
-- The text may not display fully in the SQL cell. Please hover around or double check on the SQL cell to view the full text.
SELECT
  AI_AGG(
    review_text,
    'Analyze these clothing product reviews and provide a comprehensive list of all product issues mentioned. Format your response as a bulleted list of issues with their approximate frequency in percentage.'
  ) as clothing_issues
FROM filtered_product_reviews pr
JOIN product_catalog pc ON pr.product_id = pc.product_id
WHERE pc.category = 'Clothing';

SELECT CLOTHING_ISSUES
FROM COMMON_ISSUE_SQL
LIMIT 1;

WITH clothing_issue_reviews AS (
  SELECT
    pr.review_id,
    pr.review_text
  FROM filtered_product_reviews pr
  JOIN product_catalog pc
    ON pr.product_id = pc.product_id
  WHERE
    pc.category = 'Clothing'
), classified_reviews AS (
  SELECT
    review_id,
    review_text,
    AI_CLASSIFY(
      review_text,
      [
        'Sizing issue',
        'Color issue',
        'Fabric quality issue',
        'Washing problem',
        'Pricing issue'
      ]
    ) AS classification
  FROM clothing_issue_reviews
)
SELECT
  review_id,
  review_text,
  classification:labels::text AS issue_category
FROM classified_reviews;

WITH clothing_issue_reviews AS (

  SELECT 
    pr.review_id,
    pr.review_text
  FROM filtered_product_reviews pr
  JOIN product_catalog pc ON pr.product_id = pc.product_id
  WHERE pc.category = 'Clothing'
)
SELECT
    review_id,
    review_text,
    AI_COMPLETE('llama4-maverick', 'Please draft a concise response to the customer complaints below. Please only include the draft and nothing else: ' || review_text) as response
  FROM clothing_issue_reviews