System and method for database schema matching

1 . A computer-implemented method for matching a source schema to a target schema, said method being executed by a computing device and comprising:

accessing a source table that has a first table schema, wherein the source table includes a source attribute;

generating a source document that uses first text to describe the source attribute, wherein the first text semantically describes the source attribute, and the first text is structured in accordance with a passage-based structured document format;

accessing a plurality of target tables, at least some of which have a second table schema;

generating a plurality of target documents that use second text to describe target attributes included in the plurality of target tables, wherein the second text semantically describes the target attributes, and the second text is also structured in accordance with the passage-based structured document format;

searching the plurality of target documents to identify a set of target documents whose target attributes are determined to correspond to the source attribute;

based on the identified set of target documents, retrieving a set of target tables that correspond to the identified set of target documents, such that the set of target tables include the target attributes that are determined to correspond to the source attribute;

selecting a set of the target attributes from the retrieved set of target tables; and

generating a ranked list that is formed from the set of target attributes, wherein target attributes that are included in the ranked list are ones that are determined to that match the first-source attribute.

2 . The computer-implemented method of claim 1 , wherein the source attribute comprises contents of a column in the source table.

3 . The computer-implemented method of claim 1 , wherein a large language model (LLM) selects the set of the target attributes.

4 . The computer-implemented method of claim 1 , wherein a large language model determines that the target attributes correspond to the source attribute.

5 . The computer-implemented method of claim 1 , wherein searching the plurality of target documents is performed based on semantic similarity between the source attribute and the target attributes.

6 . The computer-implemented method of claim 5 , further comprising:

encoding the second text using an embedding model to generate embeddings of the second text.

7 . The computer-implemented method of claim 6 , wherein the embeddings serve as a basis for measuring semantic similarity.

8 . The computer-implemented method of claim 3 , wherein selecting the set of the target attributes includes generating a prompt for use by the LLM.

9 . A computer system comprising:

one or more processors; and

one or more hardware storage devices that store instructions that are executable by the one or more processors to cause the computer system to:

access a source table that has a first table schema, wherein the source table includes a source attribute;

generate a source document that uses first text to describe the source attribute, wherein the first text semantically describes the source attribute, and the first text is structured in accordance with a passage-based structured document format;

access a plurality of target tables, at least some of which have a second table schema;

generate a plurality of target documents that use second text to describe target attributes included in the plurality of target tables, wherein the second text semantically describes the target attributes, and the second text is also structured in accordance with the passage-based structured document format;

search the plurality of target documents to identify a set of target documents whose target attributes are determined to correspond to the source attribute;

based on the identified set of target documents, retrieve a set of target tables that correspond to the identified set of target documents, such that the set of target tables include the target attributes that are determined to correspond to the source attribute;

select a set of the target attributes from the retrieved set of target tables; and

generate a ranked list that is formed from the set of target attributes, wherein target attributes that are included in the ranked list are ones that are determined to match the source attribute.

10 . The computer system of claim 9 , wherein the source attribute is encoded.

11 . The computer system of claim 9 , wherein the target documents are encoded.

12 . The computer system of claim 9 , wherein the source attribute is a column descriptor of a column in the source table.

13 . The computer system of claim 9 , wherein the source document includes an overview of a purpose of the source table.

14 . The computer system of claim 9 , wherein the source document includes a description of primary key attributes of the source table.

15 . The computer system of claim 9 , wherein the source document includes a description of foreign key attributes of the source table.

16 . The computer system of claim 9 , wherein the source document includes descriptions of other source attributes of the source table.

17 . The computer system of claim 9 , wherein a target document that is included among the plurality of target documents includes: an overview of a purpose of the target document, a description of primary key attributes of the target table, a description of foreign key attributes of the target table, and other target attributes of the target table.

18 . One or more hardware storage devices that store instructions that are executable by one or more processors to cause the one or more processors to:

access a source table that has a first table schema, wherein the source table includes a source attribute;

generate a source document that uses first text to describe the source attribute, wherein the first text semantically describes the source attribute, and the first text is structured in accordance with a passage-based structured document format;

access a plurality of target tables, at least some of which have a second table schema;

generate a plurality of target documents that use second text to describe target attributes included in the plurality of target tables, wherein the second text semantically describes the target attributes, and the second text is also structured in accordance with the passage-based structured document format;

search the plurality of target documents to identify a set of target documents whose target attributes are determined to correspond to the source attribute;

based on the identified set of target documents, retrieve a set of target tables that correspond to the identified set of target documents, such that the set of target tables include the target attributes that are determined to correspond to the source attribute;

select a set of the target attributes from the retrieved set of target tables; and

generate a ranked list that is formed from the set of target attributes, wherein target attributes that are included in the ranked list are ones that are determined to match the source attribute.

19 . The one or more hardware storage devices of claim 18 , wherein the source attribute is encoded.

20 . The one or more hardware storage devices of claim 18 , wherein the target documents are encoded.