AI_EXTRACT

AI_EXTRACT is a MaxCompute AI function that calls a model to extract specified information from text based on a list of labels.

Syntax

STRING AI_EXTRACT(
  STRING <model_name>,
  STRING <version_name>,
  STRING <input>,
  ARRAY<STRING> <labels>
  [, STRING <model_parameters>]
);

Parameters

model_name: Required. STRING. The name of the model to use. For more information, see SQL AI function.
version_name: Required. STRING. The name of the model version to use. To use the default version, specify DEFAULT_VERSION.
input: Required. STRING. The text from which to extract information.
labels: Required. Data type: ARRAY<STRING>. A list of the labels that you want to extract. The number of labels can range from 1 to 20.
model_parameters: Optional. STRING. Specifies model parameters such as max_tokens, temperature, and top_p. The format is a JSON string:
'{"max_tokens": 500, "temperature": 0.6, "top_p": 0.95}'.
- max_tokens: The maximum number of tokens to generate in a single model call. For MaxCompute public models, the default value is 4,096.
- temperature: A value between 0 and 1 that controls the randomness of the output. A higher value results in more creative and diverse output, while a lower value produces more deterministic and conservative output.
- top_p: A value between 0 and 1 that limits the range of candidate labels the model can choose from. A higher value allows for a broader range and more diversity, while a lower value narrows the range and produces more focused results.

Return value

Returns a STRING in JSON format that contains all extracted labels and their corresponding values. The function follows these rules:

If input is not a STRING or labels is not an ARRAY<STRING>, the function returns an error.
If a constant array is provided for labels and it contains more than 20 labels, the function returns an error.
If input or labels is NULL, or if input is an empty string (""), the function returns NULL.
If the input text lacks information for a label in the labels array, the value for that label is NULL in the output.

Examples

Example 1: Extract with constant labels

Call the MaxCompute public model qwen3.7-max to extract structured information from the given text for the specified labels.

SET odps.namespace.schema=true;

SELECT AI_EXTRACT(
    bigdata_public_modelset.default.`qwen3.7-max`,
    DEFAULT_VERSION,
    'Zhang Wei is a 35-year-old software engineer who works at the Alibaba Cloud office in Hangzhou. He joined the company in 2020 and specializes in distributed computing.',
    ARRAY('Name', 'Age', 'Occupation', 'Company', 'City', 'Year Joined')
) AS extracted_info;

-- Result
+----------------+
| extracted_info |
+----------------+
| {\n  "Name": "Zhang Wei",\n  "Age": 35,\n  "Occupation": "Software Engineer",\n  "Company": "Alibaba Cloud",\n  "City": "Hangzhou",\n  "Year Joined": 2020\n} |
+----------------+

Example 2: Extract from table data

Call the MaxCompute public model deepseek-v4-pro to extract product, issue, and sentiment information from customer reviews stored in a table.

-- Sample data
CREATE TABLE customer_reviews (
    review STRING
);

INSERT INTO customer_reviews VALUES
    ('The new laptop has great battery life, but the keyboard feels a bit off.'),
    ('Shipping was fast, and the sound quality of the headphones is excellent for the price. Very satisfied.'),
    ('The monitor had dead pixels, so I returned it. The customer service was helpful, though.');

-- Call the model to extract structured information from customer reviews in the table
SET odps.namespace.schema=true;

SELECT
    review,
    AI_EXTRACT(
        bigdata_public_modelset.default.`deepseek-v4-pro`,
        DEFAULT_VERSION,
        review,
        ARRAY('Product', 'Issue', 'Sentiment')
    ) AS extracted_info
FROM customer_reviews;

-- Result
+--------+----------------+
| review | extracted_info |
+--------+----------------+
| The new laptop has great battery life, but the keyboard feels a bit off. | {"Product": "Laptop", "Issue": "Keyboard feels off", "Sentiment": "Negative"} |
| Shipping was fast, and the sound quality of the headphones is excellent for the price. Very satisfied. | {"Product": "Headphones", "Issue": null, "Sentiment": "Positive"} |
| The monitor had dead pixels, so I returned it. The customer service was helpful, though. | {"Product": "Monitor", "Issue": "Dead pixels", "Sentiment": "Negative"} |
+--------+----------------+

FAQ

Troubleshooting public models

Symptom: When you call a public model supported by the Model Computing Service, you receive the following error:

FAILED: ODPS-0130071:[1,8] Semantic analysis exception - inference quota status check failed, error message: region cn-shanghai not found in inference quota

Cause: The Model Computing Service has not been activated in your project's region (for example, cn-shanghai). As a result, MaxCompute cannot access the AI inference resources.

Solution: Go to the Alibaba Cloud console and activate the Model Computing Service for your project's region. For detailed instructions, see Purchase and use the MaxCompute Model Computing Service.