CLASSIFY_TEXT (SNOWFLAKE.CORTEX)¶

Syntax¶

SNOWFLAKE.CORTEX.CLASSIFY_TEXT( <input> , <list_of_categories>, [ <options> ] )

Arguments¶

Required:

input

String to classify. The input string is case sensitive. You may get different results for the same string that uses different capitalization.

list_of_categories

Array that represents the categories. Must contain at least two and at most 100 unique categories. Categories are case sensitive. If these requirements are not met, the function returns an error. For a full list of errors, see Error conditions.

Categories may be simple strings or SQL objects; all categories must be the same type. Using objects, you can provide a description and examples of each category, providing context that can help improve classification accuracy. It is not required to provide descriptions or examples for each category; you are free to provide a description, examples, both, or neither for each category.

• label: The name of the category. This key is required.

• description: A description of the category. Descriptions should be no longer than about 25 words (1-2 sentences) long. This key is optional.

• examples: An array of examples that are representative of the category. Typically no more than five examples are needed, but there is a limit of 20 examples per category. The number of examples does not need to be the same for every category. This key is optional.

Note

Descriptions and examples count as input tokens, which increases the cost of the classification operation. Read more in Cost considerations.

Optional:

options

An object that contains optional configuration (as key/value pairs) for the classification operation. Currently, the only available key is:

• task_description: A string containing a short explanation of the text classification task. Task descriptions should be no more than about 50 words (3-4 sentences) long.

Returns¶

Returns a string that contains a JSON object. The JSON object contains the category that the input prompt was classified as. If invalid arguments are given, an error is returned. For details on the errors, see Error conditions.

Access control requirements¶

Users must use a role that has been granted the SNOWFLAKE.CORTEX_USER database role. See Required privileges for more information on this privilege.

Usage notes¶

For optimal performance, follow these guidelines:

• Use plain English text for input and categories.

• Limit the amount of text that is not plain English in the input text. For example, try to limit content such as code snippets or logs in the text input.

• Text shouldn’t contain code or formats that are not open source (company specific languages, proprietary formats, etc.). The function won’t return an error, but the results may not be what you expect.

• Don’t use abbreviations, special characters, or jargon in the category labels.

• Categories should be descriptive. For example using a category such as Xa4s3 or category 1 won’t produce good results.

• Categories should be mutually exclusive.

• Adding a clear task description can improve accuracy when the relationship between the input text and categories is ambiguous or nuanced.

• Adding label descriptions can improve accuracy in cases where the descriptions are ambiguous or when specific logic should be followed when selecting a particular label. When writing descriptions, focus on key aspects that distinguish a particular label from the others.

• Examples can help to improve accuracy.

Examples¶

SELECT SNOWFLAKE.CORTEX.CLASSIFY_TEXT('One day I will see the world', ['travel', 'cooking']);

{
  "label": "travel"
}

CREATE OR REPLACE TEMPORARY TABLE text_classification_table AS
SELECT 'France' AS input, ['North America', 'Europe', 'Asia'] AS classes
UNION ALL
SELECT 'Singapore', ['North America', 'Europe', 'Asia']
UNION ALL
SELECT 'one day I will see the world', ['travel', 'cooking', 'dancing']
UNION ALL
SELECT 'my lobster bisque is second to none', ['travel', 'cooking', 'dancing'];

SELECT input,
       classes,
       SNOWFLAKE.CORTEX.CLASSIFY_TEXT(input, classes)['label'] as classification
FROM text_classification_table;

SELECT SNOWFLAKE.CORTEX.CLASSIFY_TEXT(
  'When I am not at work, I love creating recipes using every day ingredients',
  ['travel', 'cooking', 'fitness'],
  {
    'task_description': 'Return a classification of the Hobby identified in the text'
  }
);

{
  "label": "cooking"
}

SELECT SNOWFLAKE.CORTEX.CLASSIFY_TEXT(
  'I love running every morning before the world wakes up',
  [{
    'label': 'travel',
    'description': 'Hobbies related to going from one place to another',
    'examples': ['I like flying to Europe', 'Every summer we go to Italy' , 'I love traveling to learn new cultures']
  },{
    'label': 'cooking',
    'description': 'Hobbies related to preparing food',
    'examples': ['I like learning about new ingredients', 'You must bring your soul to the recipe' , 'Baking is my therapy']
    },{
    'label': 'fitness',
    'description': 'Hobbies related to being active and healthy',
    'examples': ['I cannot live without my Strava app', 'Running is life' , 'I go to the Gym every day']
    }],
  {'task_description': 'Return a classification of the Hobby identified in the text'})

{
  "label": "fitness"
}

SELECT SNOWFLAKE.CORTEX.CLASSIFY_TEXT(
  'I love running every morning before the world wakes up',
  [{
    'label': 'travel',
    'description': 'Hobbies related to going from one place to another',
    'examples': ['I like flying to Europe']
  },{
    'label': 'cooking',
    'examples': ['I like learning about new ingredients', 'You must bring your soul to the recipe' , 'Baking is my therapy']
    },{
    'label': 'fitness',
    'description': 'Hobbies related to being active and healthy'
    }],
  {'task_description': 'Return a classification of the Hobby identified in the text'})

{
  "label": "fitness"
}