Top
Enterprise Postgres 18 Knowledge DataManagement FeatureUser's Guide

3.13.4 Tables/Views Created by Semantic Text Search and Automatic Vectorization Feature

For information about the tables and views that pgai creates, refer to the pgai documentation.

pgx_vectorizer creates the following tables.

pgx_vectorizer.worker_setting_table table

Stores information about the parameters used by the embedded provider.

Column

Type

Constraint

Description

type

text

PRIMARY KEY

The type of parameter to set.

The name of the embedded provider or user

parameter

text

PRIMARY KEY

Parameter name

value

text

NOT NULL

Value to set for the parameter

pgx_vectorizer.pgx_fulltext_index table

The pgx_fulltext_index table stores the definition information of the full-text search index created by the pgx_create_vectorizer function.

Column

Type

Reference destination

Description

vectorizerid

integer

ai.vectorizer.id

Vectorization identifier

config

jsonb

Definition information of full-text search index.

Configurable values are the return values of the pgx_fulltext_indexing_gin function or the pgx_fulltext_indexing_gist function.

pgx_vectorizer.pgx_trace_query table

Provides information on hybrid search where trace information is recorded.

When a hybrid search is performed with tracing enabled, one record is inserted.

Deleting records from this table also deletes records from the pgx_trace_subquery table, pgx_trace_results table, and pgx_context_metrics table with the same queryid.

Column

Type

Description

queryid

bigint

Hybrid search ID

retrieval_time

timestamp with timezone

Time when hybrid search was conducted

retrieval_user

oid

Database user who performed the hybrid search

application_name

text

Application name that issued the hybrid search. The value of application_name specified

query

jsonb

Search criteria for hybrid search

pgx_vectorizer.pgx_trace_subquery table

Provides information about the subqueries for hybrid searches where trace information is recorded.

When a hybrid search is performed with tracing enabled, multiple records are inserted for each subquery.

Column

Type

Description

queryid

bigint

Hybrid search ID

subquery_seq

smallint

Serial number of subquery conducted within a hybrid search(A serial number closed in a certain hybrid search process)

subquery_type

text

Types of subqueries

'-' : Overall hybrid search (main query)

'semantic' : Semantic text search

'fulltext' : Full-text search

subquery_text

text

Content of the executed subquery (SQL statement)

pgx_vectorizer.pgx_trace_results table

Provides information about the text chunks returned by each search process of hybrid search and subquery.

When performing a hybrid search with trace information enabled, multiple records corresponding to the text chunks of the search results of each subquery are inserted.

Column

Type

Description

queryid

bigint

Hybrid search ID

subquery_seq

smallint

Serial number of subquery conducted within a hybrid search

context_id

uuid

Returned text chunk ID

chunk

text

Returned text chunk

rank

integer

Rank of text chunks within the same subquery type

score

real

Score of text chunk.

Hybrid search overall : Score calculated by RRF

Semantic text search : Score based on distance between vectors in text meaning search

Full-text search : Score calculated by score_expression in hybrid search processing

In the case of text meaning search, the score is calculated using the distance d between the vector representations of the text chunk and the query text, according to the following formula.

pgx_vectorizer.pgx_context_metrics table

This is a list of text chunks returned by either hybrid search or subquery. When the pgx_list_contexts function is executed, records for the specified query ID are inserted. You can enter any evaluation value in the context_metrics column.

Column

Type

Description

queryid

bigint

Hybrid search ID

context_id

uuid

Returned text chunk ID

context_metrics

jsonb

You can input evaluation values for text chunks.

pgx_vectorizer.pgx_trace_contexts view

Provide information about the text chunks returned by either hybrid search or subqueries.

For hybrid searches conducted with trace information enabled, there are records for each text chunk of search results for each subquery.

Column

Type

Description

queryid

bigint

Hybrid search ID

retrieval_time

timestamp with timezone

Time when hybrid search was conducted

retrieval_user

oid

Database user who performed the hybrid search

application_name

text

Application name that issued the hybrid search. The value of application_name specified

query

jsonb

Search criteria for hybrid search

subquery_type

text

Types of subqueries

'-' : Overall hybrid search (main query)

'semantic' : Semantic text search

'fulltext' : Full-text search

subquery_text

text

Content of the executed subquery (SQL statement)

context_id

uuid

Returned text chunk ID

chunk

text

Returned text chunk

rank

integer

Rank of text chunks within the same subquery type

score

real

Score of text chunk.

Hybrid search overall : Score calculated by RRF

Semantic text search : Score based on distance between vectors in text meaning search

Full-text search : Score calculated by score_expression in hybrid search processing

context_metrics

jsonb

You can input evaluation values for text chunks.