Text classification

Introduction to Generative AI in Snowflake

James Cha-Earley

Senior Developer Advocate, Snowflake

Challenges of classifying text

The problem

Reviews can contain any information

Manual categorizing is time-consuming and not scalable

The solution

Automate using Snowflake Cortex!

classify_text()

Defining categories

# Define categories
categories = ["overall_experience", "location", "staff", "food_beverages",
              "facilities"]

Classifying text

from snowflake.cortex import classify_text


category = classify_text(
    str_input="The check-in was smooth and the staff were very friendly.",
    categories=categories
)


print(category)

{  
  "label": "staff"  
}

Converting outputs to the dictionary

print(type(category))

<class 'str'>

import json

category_dict = json.loads(category)
print(type(category_dict))

<class 'dict'>

Scaling the workflow

Classification pipeline categorizing reviews for a given month

# Python code
month = 5

-- SQL query
SELECT *
FROM HOTELS.REVIEWS
WHERE EXTRACT(month FROM date) = '{{month}}'

Applying classify_text()

df = cell2.to_pandas()


def classification(text):
    result = classify_text(
        str_input=text,
        categories=labels
    )
    result_dict = json.loads(result)
    return result_dict["label"]

Applying classify_text()

# Apply classification to the DataFrame
df["category"] = reviews["DESCRIPTION"].apply(classification)


# Print the first row's predicted category
print(df["category"].head(1))

0    overall_experience  
Name: category, dtype: object

Sentiment analysis

Only interested about reviews sentiment

AI_SENTIMENT in a SQL cell

Sentiment analysis

AI_SENTIMENT - extracting a sentiment

Available sentiments: positive, negative, neutral, mixed, and unknown

Let's practice!

Introduction to Generative AI in Snowflake