# US 2025/0148211 A1 · Semantically Classifying Sets Of Data Elements

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0148211 A1
- Document ID: US20250148211A1
- Status: Published application
- Kind: A1
- Application number: 18/501,282
- Filed: 2023-11-03
- Published: 2025-05-08
- Earliest claimed: 2023-11-03
- Assignees: Oracle International Corporation
- Applicants: Oracle International Corporation
- Inventors: Dima Alberg; Guy Shaked
- Web page: https://patentgrep.com/patent/US20250148211A1

## CPC

- G06F 40/30: Semantic analysis
- G06F 16/353: into predefined classes
- G06F 21/60: Protecting data
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

One or more embodiments include operations associated with semantic classification of data columns. The operations may include receiving a set of data elements corresponding to a data column to be semantically classified, applying a machine learning model to the set of data elements to predict a set of candidate semantic types for the set of data elements, selecting a particular semantic type from the set of candidate semantic types based at least in part on a semantic fit score corresponding to the particular semantic type predicted by the machine learning model, and presenting the particular semantic type as a recommended semantic classification for the data column.

## Claims

### Claim 1

One or more non-transitory computer-readable media storing instructions, which when executed by one or more hardware processors, cause performance of operations comprising:
receiving a set of data elements corresponding to a data column to be semantically classified;
applying a machine learning model to the set of data elements to predict a set of candidate semantic types for the set of data elements, wherein the set of candidate semantic types comprises, for each data element of the set of data elements, at least one candidate semantic type corresponding to the data element;
selecting a particular semantic type from the set of candidate semantic types based at least in part on a semantic fit score corresponding to the particular semantic type predicted by the machine learning model;
presenting the particular semantic type as a recommended semantic classification for the data column.

### Claim 2 (depends on 1)

The media of claim 1, wherein the operations further comprise:
generating the set of data elements at least in part by transforming structured data into unstructured data.

### Claim 3 (depends on 1)

The media of claim 1, wherein the semantic fit score is based at least in part on an occurrence frequency of the particular semantic type in the set of candidate semantic types predicted by the machine learning model.

### Claim 4 (depends on 3)

The media of claim 3, wherein selecting the particular semantic type from the set of candidate semantic types comprises:
determining that the occurrence frequency of the particular semantic type meets a threshold occurrence frequency; and
selecting the particular semantic type from the set of candidate semantic types based at least in part on the occurrence frequency meeting the threshold occurrence frequency.

### Claim 5 (depends on 1)

The media of claim 1,
wherein the semantic fit score is based at least in part on a confidence level for the particular semantic type,
wherein the confidence level represents a degree of confidence that the particular semantic type predicted for a particular data element accurately represents a semantic meaning of the particular data element.

### Claim 6 (depends on 5)

The media of claim 5, wherein selecting the particular semantic type from the set of candidate semantic types comprises:
determining that the confidence level for the particular semantic type meets a threshold confidence level; and
selecting the particular semantic type from the set of candidate semantic types based at least in part on the confidence level meeting the threshold confidence level.

### Claim 7 (depends on 1)

The media of claim 1, wherein selecting the particular semantic type from the set of candidate semantic types comprises:
determining, for each candidate semantic type of the set of candidate semantic types, a candidate semantic fit score corresponding to the particular candidate semantic type; and
determining that a first candidate semantic fit score, corresponding to a first candidate semantic type from among the set of candidate semantic types, is a highest candidate semantic fit score with respect to the set of candidate semantic types; and
selecting the first candidate semantic type as the particular semantic type based at least in part on the first candidate semantic fit score being the highest candidate semantic fit score with respect to the set of candidate semantic types, wherein the semantic fit score corresponding to the particular semantic type is the first candidate semantic fit score.

### Claim 8 (depends on 7)

The media of claim 7, wherein, for the first candidate semantic type, determining the candidate semantic fit score corresponding to the particular candidate semantic type comprises:
computing, for each particular data element of the set of data elements, a confidence level for the particular candidate semantic type having been predicted for the particular data element of the set of data elements;
computing a denominator comprising a first sum of each particular confidence level corresponding to each particular data element of the set of data elements;
computing a numerator comprising a second sum of each particular confidence level corresponding to each occurrence of the first candidate semantic type having been predicted; and
computing a quotient comprising the numerator divided by the denominator, wherein the quotient is the first candidate semantic fit score corresponding to the first candidate semantic type.

### Claim 9 (depends on 1)

The media of claim 1,
wherein selecting the particular semantic type from the set of candidate semantic types comprises:
determining, for each particular candidate semantic type of the set of candidate semantic types, a particular semantic fit score corresponding to the particular candidate semantic type;
comparing each particular semantic fit score to a threshold semantic fit score; and
selecting a plurality of particular semantic types from the set of candidate semantic types, wherein each particular semantic type of the plurality of particular semantic types is selected based at least in part on the particular semantic fit score corresponding to the particular semantic type meeting the threshold semantic fit score; and
wherein presenting the particular semantic type as the recommended semantic classification comprises:
presenting each particular semantic type of the plurality of particular semantic types as a candidate recommended semantic classifications for the data column.

### Claim 10 (depends on 1)

The media of claim 1, wherein the operations further comprise:
determining that the recommended semantic classification is associated with a personal information classification; and
identifying the data column as containing one or more items of personal information.

### Claim 11 (depends on 10)

The media of claim 10, wherein the operations further comprise:
determining that the data column is associated with a data breach; and
generating an alert, based at least in part on the data column having been identified as containing the one or more items of personal information, indicating that the data breach includes the one or more items of personal information.

### Claim 12 (depends on 1)

The media of claim 1, wherein the operations further comprise:
determining that a first candidate semantic type, of the set of candidate semantic types, is associated with a personal information classification;
determining a first semantic fit score corresponding to the first candidate semantic type;
comparing the first semantic fit score to a threshold semantic fit score; and
identifying the data column as containing one or more items of personal information based at least in part on the first semantic fit score meeting the threshold semantic fit score.

### Claim 13 (depends on 12)

The media of claim 12, wherein the first candidate semantic type differs from the particular semantic type presented as the recommended semantic classification for the data column.

### Claim 14 (depends on 1)

The media of claim 1, wherein the operations further comprise:
determining that a first candidate semantic type, of the set of candidate semantic types, is associated with a non-personal information classification;
determining a first semantic fit score corresponding to the first candidate semantic type;
comparing the first semantic fit score to a threshold semantic fit score;
identifying the data column as containing non-personal information based at least in part on the first semantic fit score meeting the threshold semantic fit score.

### Claim 15 (depends on 14)

The media of claim 14, wherein the operations further comprise:
determining that the data column is associated with a data breach; and
generating an alert, based at least in part on the data column having been identified as containing non-personal information, indicating that the data breach includes non-personal information.

### Claim 16 (depends on 1)

The media of claim 1, wherein the operations further comprise:
receiving feedback corresponding to one or more outputs of the machine learning model; and
updating the machine learning model based at least in part on the feedback;
wherein the one or more outputs of the machine learning model comprises at least one of: the set of candidate semantic types predicted by the machine learning model, or the particular semantic type presented as the recommended semantic classification for the data column.

### Claim 17 (depends on 1)

The media of claim 1, wherein the operations further comprise:
training the machine learning model, based on one or more initial supervised learning datasets, to predict candidate semantic types for data elements; and
retraining the machine learning model based on one or more subsequent supervised learning datasets, wherein at least one of the one or more subsequent supervised learning datasets is based at least in part on feedback corresponding to at least one of: the set of candidate semantic types predicted by the machine learning model, or the particular semantic type presented as the recommended semantic classification for the data column.

### Claim 18 (depends on 1)

The media of claim 1, wherein the operations further comprise:
determining a semantic similarity between the recommended semantic classification for the data column and at least one schema element;
determining that the semantic similarity meets a threshold semantic similarity; and
presenting the set of data elements corresponding to the data column as being semantically related to the at least one schema element based at least in part on the semantic similarity meeting the threshold semantic similarity.

### Claim 19 (depends on 1)

The media of claim 1, wherein the operations further comprise:
determining that the semantic fit score is below a threshold semantic fit score; and
presenting the set of data elements corresponding to the data column as being a candidate for data cleansing based at least in part on the semantic fit score being below the threshold semantic fit score.

### Claim 20

A method, comprising:
receiving a set of data elements corresponding to a data column to be semantically classified;
applying a machine learning model to the set of data elements to predict a set of candidate semantic types for the set of data elements, wherein the set of candidate semantic types comprises, for each data element of the set of data elements, at least one candidate semantic type corresponding to the data element;
selecting a particular semantic type from the set of candidate semantic types based at least in part on a semantic fit score corresponding to the particular semantic type predicted by the machine learning model;
presenting the particular semantic type as a recommended semantic classification for the data column;
wherein the method is performed by at least one device including a hardware processor.

### Claim 21 (depends on 20)

The method of claim 20, further comprising:
generating the set of data elements at least in part by transforming structured data into unstructured data.

### Claim 22 (depends on 20)

The method of claim 20, wherein the semantic fit score is based at least in part on an occurrence frequency of the particular semantic type in the set of candidate semantic types predicted by the machine learning model.

### Claim 23 (depends on 22)

The method of claim 22, wherein selecting the particular semantic type from the set of candidate semantic types comprises:
determining that the occurrence frequency of the particular semantic type meets a threshold occurrence frequency; and
selecting the particular semantic type from the set of candidate semantic types based at least in part on the occurrence frequency meeting the threshold occurrence frequency.

### Claim 24 (depends on 20)

The method of claim 20,
wherein the semantic fit score is based at least in part on a confidence level for the particular semantic type,
wherein the confidence level represents a degree of confidence that the particular semantic type predicted for a particular data element accurately represents a semantic meaning of the particular data element.

### Claim 25 (depends on 24)

The method of claim 24, wherein selecting the particular semantic type from the set of candidate semantic types comprises:
determining that the confidence level for the particular semantic type meets a threshold confidence level; and
selecting the particular semantic type from the set of candidate semantic types based at least in part on the confidence level meeting the threshold confidence level.

### Claim 26 (depends on 20)

The method of claim 20, wherein selecting the particular semantic type from the set of candidate semantic types comprises:
determining, for each candidate semantic type of the set of candidate semantic types, a candidate semantic fit score corresponding to the particular candidate semantic type; and
determining that a first candidate semantic fit score, corresponding to a first candidate semantic type from among the set of candidate semantic types, is a highest candidate semantic fit score with respect to the set of candidate semantic types; and
selecting the first candidate semantic type as the particular semantic type based at least in part on the first candidate semantic fit score being the highest candidate semantic fit score with respect to the set of candidate semantic types, wherein the semantic fit score corresponding to the particular semantic type is the first candidate semantic fit score.

### Claim 27 (depends on 26)

The method of claim 26, wherein determining the candidate semantic fit score corresponding to the particular candidate semantic type comprises:
for the first candidate semantic type:
computing, for each particular data element of the set of data elements, a confidence level for the particular candidate semantic type having been predicted for the particular data element of the set of data elements;
computing a denominator comprising a first sum of each particular confidence level corresponding to each particular data element of the set of data elements;
computing a numerator comprising a second sum of each particular confidence level corresponding to each occurrence of the first candidate semantic type having been predicted; and
computing a quotient comprising the numerator divided by the denominator, wherein the quotient is the first candidate semantic fit score corresponding to the first candidate semantic type.

### Claim 28 (depends on 20)

The method of claim 20,
wherein selecting the particular semantic type from the set of candidate semantic types comprises:
determining, for each particular candidate semantic type of the set of candidate semantic types, a particular semantic fit score corresponding to the particular candidate semantic type;
comparing each particular semantic fit score to a threshold semantic fit score; and
selecting a plurality of particular semantic types from the set of candidate semantic types, wherein each particular semantic type of the plurality of particular semantic types is selected based at least in part on the particular semantic fit score corresponding to the particular semantic type meeting the threshold semantic fit score; and
wherein presenting the particular semantic type as the recommended semantic classification comprises:
presenting each particular semantic type of the plurality of particular semantic types as a candidate recommended semantic classifications for the data column.

### Claim 29 (depends on 20)

The method of claim 20, further comprising:
determining that the recommended semantic classification is associated with a personal information classification; and
identifying the data column as containing one or more items of personal information.

### Claim 30 (depends on 29)

The method of claim 29, further comprising:
determining that the data column is associated with a data breach; and
generating an alert, based at least in part on the data column having been identified as containing the one or more items of personal information, indicating that the data breach includes the one or more items of personal information.

### Claim 31 (depends on 20)

The method of claim 20, further comprising:
determining that a first candidate semantic type, of the set of candidate semantic types, is associated with a personal information classification;
determining a first semantic fit score corresponding to the first candidate semantic type;
comparing the first semantic fit score to a threshold semantic fit score; and
identifying the data column as containing one or more items of personal information based at least in part on the first semantic fit score meeting the threshold semantic fit score.

### Claim 32 (depends on 31)

The method of claim 31, wherein the first candidate semantic type differs from the particular semantic type presented as the recommended semantic classification for the data column.

### Claim 33 (depends on 20)

The method of claim 20, further comprising:
determining that a first candidate semantic type, of the set of candidate semantic types, is associated with a non-personal information classification;
determining a first semantic fit score corresponding to the first candidate semantic type;
comparing the first semantic fit score to a threshold semantic fit score;
identifying the data column as containing non-personal information based at least in part on the first semantic fit score meeting the threshold semantic fit score.

### Claim 34 (depends on 33)

The method of claim 33, further comprising:
determining that the data column is associated with a data breach; and
generating an alert, based at least in part on the data column having been identified as containing non-personal information, indicating that the data breach includes non-personal information.

### Claim 35 (depends on 20)

The method of claim 20, further comprising:
receiving feedback corresponding to one or more outputs of the machine learning model; and
updating the machine learning model based at least in part on the feedback;
wherein the one or more outputs of the machine learning model comprises at least one of: the set of candidate semantic types predicted by the machine learning model, or the particular semantic type presented as the recommended semantic classification for the data column.

### Claim 36 (depends on 20)

The method of claim 20, further comprising:
training the machine learning model, based on one or more initial supervised learning datasets, to predict candidate semantic types for data elements; and
retraining the machine learning model based on one or more subsequent supervised learning datasets, wherein at least one of the one or more subsequent supervised learning datasets is based at least in part on feedback corresponding to at least one of: the set of candidate semantic types predicted by the machine learning model, or the particular semantic type presented as the recommended semantic classification for the data column.

### Claim 37 (depends on 20)

The method of claim 20, further comprising:
determining a semantic similarity between the recommended semantic classification for the data column and at least one schema element;
determining that the semantic similarity meets a threshold semantic similarity; and
presenting the set of data elements corresponding to the data column as being semantically related to the at least one schema element based at least in part on the semantic similarity meeting the threshold semantic similarity.

### Claim 38 (depends on 20)

The method of claim 20, further comprising:
determining that the semantic fit score is below a threshold semantic fit score; and
presenting the set of data elements corresponding to the data column as being a candidate for data cleansing based at least in part on the semantic fit score being below the threshold semantic fit score.

### Claim 39

A system comprising:
at least one hardware processor;
the system being configured to execute operations, using the at least one hardware processor, the operations comprising:
receiving a set of data elements corresponding to a data column to be semantically classified;
applying a machine learning model to the set of data elements to predict a set of candidate semantic types for the set of data elements, wherein the set of candidate semantic types comprises, for each data element of the set of data elements, at least one candidate semantic type corresponding to the data element;
selecting a particular semantic type from the set of candidate semantic types based at least in part on a semantic fit score corresponding to the particular semantic type predicted by the machine learning model;
presenting the particular semantic type as a recommended semantic classification for the data column.
