Analyze customer reviews using AI Functions

This article illustrates how to use AI Functions to examine customer reviews and determine if a response needs to be generated. The AI Functions used in this example are built-in Databricks SQL functions, powered by generative AI models made available by Databricks Foundation Model APIs. See AI Functions on Azure Databricks.

This example performs the following on a test dataset called reviews using AI Functions:

• Determines the sentiment of a review.
• For negative reviews, extracts information from the review to classify the cause.
• Identifies whether a response is required back to the customer.
• Generates a response mentioning alternative products that may satisfy the customer.

Requirements

• A workspace in a Foundation Model APIs pay-per-token supported region.
• These functions are not available on Azure Databricks SQL Classic.
• During the preview, these functions have restrictions on their performance. Reach out to your Databricks account team if you require a higher quota for your use cases.

Analyze sentiment of reviews

You can use the ai_analyze_sentiment() to help you understand how customers feel from their reviews. In the following example, the sentiment can be positive, negative, neutral, or mixed.

SELECT
  review,
  ai_analyze_sentiment(review) AS sentiment
FROM
  product_reviews;

From the following results, you see that the function returns the sentiment for each review without any prompt engineering or parsing results.

Classify reviews

In this example, after identifying negative reviews you can use ai_classify() to gain more insights into customer reviews, like whether the negative review is due to poor logistics, product quality, or other factors.

SELECT
  review,
  ai_classify(
    review,
    ARRAY(
      "Arrives too late",
      "Wrong size",
      "Wrong color",
      "Dislike the style"
    )
  ) AS reason
FROM
  product_reviews
WHERE
  ai_analyze_sentiment(review) = "negative"

In this case, ai_classify() is able to correctly categorize the negative reviews based on custom labels to allow for further analysis.

Extract information from reviews

You might want to improve your product description based on the reasons customers had for their negative reviews. You can find key information from a blob of text using ai_extract(). The following example extracts information and classifies if the negative review was based on sizing issues with the product:

SELECT
  review,
  ai_extract(review, array("usual size")) AS usual_size,
  ai_classify(review, array("Size is wrong", "Size is right")) AS fit
FROM
  product_reviews

The following are a sample of results:

Generate responses with recommendations

After reviewing the customer responses, you can use the ai_gen() function to generate a response to a customer based on their complaint and strengthen customer relationships with prompt replies to their feedback.

SELECT
  review,
  ai_gen(
    "Generate a reply in 60 words to address the customer's review.
    Mention their opinions are valued and a 30% discount coupon code has been sent to their email.
    Customer's review: " || review
  ) AS reply
FROM
  product_reviews
WHERE
  ai_analyze_sentiment(review) = "negative"

The following are a sample of results:

Additional resources

• AI Functions on Azure Databricks
• Databricks Foundation Model APIs