AI_SENTIMENT

Overview

AI_SENTIMENT is an AI sentiment analysis function provided by Singdata Lakehouse. It analyzes the sentiment of input text and returns one of four labels: positive, negative, neutral, or mixed. Supports Chinese, English, Japanese, and many other languages — one line of SQL is all it takes.

Singdata pushes AI computation down to the storage and execution engine layer. Data is processed intelligently within the platform without leaving the system, ensuring data security while significantly reducing task latency.

Syntax

AI_SENTIMENT( <model>, <text> )

Parameters

Required Parameters

model

Specifies the model for sentiment analysis. Supports two sources:

Source 1: API Gateway Endpoint (Recommended)

A platform administrator pre-configures model services in the API Gateway. Regular users reference them with the endpoint: prefix, without needing to know the underlying connection details.

'endpoint:<endpoint_name>' -- Examples 'endpoint:qwen3-max-preview' 'endpoint:qwen3.5-plus'

Source 2: API Connection Object

Users create their own connection objects via CREATE API CONNECTION, suitable for custom service addresses, authentication keys, or private deployment models.

-- Create a connection object CREATE API CONNECTION conn_bailian TYPE ai_function PROVIDER = 'bailian' BASE_URL = 'https://dashscope.aliyuncs.com/api/v1' API_KEY = 'sk-xxxxxxxxxxxxxxxxxxxxxxxx'; -- Reference using <connection_name>:<model_name> format SELECT AI_SENTIMENT('conn_bailian:qwen3.5-plus', 'This product is great!');

CREATE API CONNECTION field descriptions:

Field	Description
`TYPE`	Fixed as `ai_function`
`PROVIDER`	Model provider identifier, e.g. `'bailian'`, `'openai'`, `'anthropic'`
`BASE_URL`	Base API URL of the model service
`API_KEY`	Authentication key for calling the service

text

The input text to analyze, type STRING. Supports Chinese, English, Japanese, French, German, Spanish, and many other languages — no need to specify the language manually; the model detects it automatically.

Return Value

STRING type, one of the following values:

Value	Meaning
`positive`	Text expresses a positive evaluation or optimistic sentiment
`negative`	Text expresses a negative evaluation or pessimistic sentiment
`neutral`	Text contains no clear sentiment (e.g. factual statements, notifications)
`mixed`	Text contains both positive and negative sentiment

Error Behavior

By default, if the function cannot process the input, it returns NULL without raising an error. In multi-row queries, rows that error return NULL without affecting other rows.

Usage Notes

NULL and empty strings both return NULL: When text is NULL or an empty string '', the function returns NULL without error. Whitespace-only strings (e.g. ' ') are sent to the model and typically return neutral.
Results are non-deterministic: LLM-based sentiment analysis may vary slightly across model versions or call times; the same input may produce different results across executions.
Note the difference between mixed and neutral: mixed means the text contains both positive and negative evaluations (e.g. "food was great but service was terrible"); neutral means the text has no emotional tone (e.g. "meeting at 3pm").
Multilingual requires no extra configuration: The model automatically detects the input language; Chinese, English, Japanese, French, German, Spanish, and others all work directly.

Examples

Basic Usage

-- Positive sentiment SELECT AI_SENTIMENT( 'endpoint:qwen3-max-preview', 'This product is amazing, highly recommended!' ) AS sentiment; -- Returns: positive -- Negative sentiment SELECT AI_SENTIMENT( 'endpoint:qwen3-max-preview', 'The quality is terrible, complete waste of money, very disappointed.' ) AS sentiment; -- Returns: negative -- Neutral text SELECT AI_SENTIMENT( 'endpoint:qwen3-max-preview', 'There is a meeting at 3pm today to discuss next quarter budget planning.' ) AS sentiment; -- Returns: neutral -- Mixed sentiment SELECT AI_SENTIMENT( 'endpoint:qwen3-max-preview', 'The food was delicious, but the service was terrible — waited an hour for our order.' ) AS sentiment; -- Returns: mixed

Multilingual

-- English SELECT AI_SENTIMENT('endpoint:qwen3-max-preview', 'This is the best purchase I have ever made!') AS sentiment; -- Returns: positive -- Japanese SELECT AI_SENTIMENT('endpoint:qwen3-max-preview', 'この製品は素晴らしいです。品質が非常に高いです。') AS sentiment; -- Returns: positive -- French SELECT AI_SENTIMENT('endpoint:qwen3-max-preview', 'Ce restaurant est absolument terrible.') AS sentiment; -- Returns: negative

Semantic Understanding Capabilities

The model handles the following without extra configuration:

-- Sarcasm/irony detection SELECT AI_SENTIMENT('endpoint:qwen3-max-preview', 'Oh great, working until midnight again, I am just so thrilled.') AS sentiment; -- Returns: negative (correctly identifies sarcasm, not the literal positive meaning) -- Double negation SELECT AI_SENTIMENT('endpoint:qwen3-max-preview', 'This product is not bad at all.') AS sentiment; -- Returns: positive -- Emoji sentiment SELECT AI_SENTIMENT('endpoint:qwen3-max-preview', 'Tonight''s dinner 🤮🤮🤮') AS sentiment; -- Returns: negative SELECT AI_SENTIMENT('endpoint:qwen3-max-preview', 'Got a gift! 🎉❤️😍') AS sentiment; -- Returns: positive

Batch Analysis on Table Data

SELECT id, review_content, AI_SENTIMENT('endpoint:qwen3-max-preview', review_content) AS sentiment FROM product_reviews WHERE review_content IS NOT NULL;

Sentiment Distribution Statistics

SELECT sentiment, COUNT(*) AS cnt, ROUND(COUNT(*) * 100.0 / SUM(COUNT(*)) OVER(), 1) AS pct FROM ( SELECT AI_SENTIMENT('endpoint:qwen3-max-preview', content) AS sentiment FROM customer_feedback WHERE content IS NOT NULL ) GROUP BY sentiment ORDER BY cnt DESC;

Prioritize Negative Support Tickets

SELECT ticket_id, description, created_at FROM support_tickets WHERE AI_SENTIMENT('endpoint:qwen3-max-preview', description) = 'negative' ORDER BY created_at DESC LIMIT 50;

Limitations

model parameter is required: Omitting it causes the error AI function must have at least two arguments.
Invalid model format causes an error: model must use 'endpoint:<name>' or '<connection_name>:<model_name>' format; incorrect format causes Invalid model coordinates.
Input length is model-limited: Input text length is limited by the underlying model's context window; overly long text is automatically truncated without error.
No dimension-level sentiment analysis: Only overall sentiment is returned; per-dimension analysis (e.g. "price", "service") is not supported.