phrases

You can use this aggregation to identify common user queries, discover trending topics, extract key phrases from logs, or analyze conversation patterns in AI applications.

Usage#

Syntax#

summarize phrases(column, max_phrases)

Parameters#

Name	Type	Required	Description
`column`	string	Yes	The column containing text data from which to extract phrases.
`max_phrases`	long	Yes	The maximum number of top phrases to return.

Returns#

Returns a dynamic array containing the most common phrases found in the specified column, ordered by frequency.

Example#

Extract common phrases from GenAI conversation text to identify trending topics and patterns.

Query

['otel-demo-genai']
| extend conversation_text = genai_concat_contents(['attributes.gen_ai.input.messages'], ' | ')
| summarize common_phrases = phrases(conversation_text, 20)

Run in Playground

Output

Count	common_phrases
3	first query
3	cover related future queries
3	use title case

This query identifies the most common phrases in GenAI conversations, helping you discover trending topics and user needs.

make_list: Creates an array of all values. Use this when you need all occurrences rather than common phrases.
make_set: Creates an array of unique values. Use this for distinct values without frequency analysis.
topk: Returns top K values by a specific aggregation. Use this for numerical top values rather than phrase extraction.
count: Counts occurrences. Combine with group by for manual phrase counting if you need more control.
dcount: Counts distinct values. Use this to understand the variety of phrases before extracting top ones.

Other query languages#

Splunk SPL users

In Splunk SPL, there’s no built-in phrases function, but you might use the rare or top commands on tokenized text.

Splunk example

| rex field=message "(?<words>\w+)"
| top words

APL equivalent

['sample-http-logs']
| summarize phrases(uri, 10)

ANSI SQL users

In ANSI SQL, you would need complex string manipulation and grouping to extract common phrases.

SQL example

SELECT
  phrase,
  COUNT(*) as frequency
FROM (
  SELECT UNNEST(SPLIT(message, ' ')) as phrase
  FROM logs
)
GROUP BY phrase
ORDER BY frequency DESC
LIMIT 10