# US 12,182,091 B2 · Semantic vectorization for feature engineering

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

- Number: US 12,182,091 B2
- Document ID: US12182091B2
- Status: Granted
- Kind: B2
- Application number: 18/168,982
- Filed: 2023-02-14
- Published: 2024-07-04
- Granted: 2024-12-31
- Earliest claimed: 2022-12-30
- Assignees: Sap SE
- Applicants: Sap SE
- Inventors: Mohamed Bouadi; Arta Alavi; Salima Benbernou; Mourad Ouziri
- Web page: https://patentgrep.com/patent/US12182091B2

## CPC

- G06F 16/221: {Column-oriented storage; Management thereof}
- G06N 3/006: based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
- G06N 3/045: Combinations of networks
- G06N 3/08: Learning methods
- G06N 3/092: Reinforcement learning
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Systems and methods include determination of a plurality of features, determination, for each of the plurality of features, of a feature vector based on a taxonomy of logical entities, combination of the determined feature vectors into a composite feature vector, determination of an operator based on the composite feature vector, and determination of a new feature based on the operator.

## Claims

### Claim 1

A system comprising:
a memory storing processor-executable program code; and
at least one processing unit to execute the processor-executable program code to cause the system to:
determine a plurality of database columns of one or more database tables;
for each of the plurality of database columns, determine a feature vector based on a taxonomy of logical entities;
combine the determined feature vectors of each of the plurality of database columns into a composite feature vector;
determine an operator based on the composite feature vector; and
determine a feature based on the operator and the plurality of database columns,
wherein each entry of each of the feature vectors corresponds to a respective one of the logical entities, and
wherein combination of the determined feature vectors comprises determination, for each one of the logical entities, of a composite value based on the entry of each of the feature vectors which corresponds to the logical entity.

### Claim 2 (depends on 1)

A system according to claim 1, the at least one processing unit to execute the processor-executable program code to cause the system to:
determine a second feature vector based on the determined feature;
combine the second feature vector with the composite feature vector to generate a second composite feature vector;
determine a second operator based on the second composite feature vector; and
determine a third feature based on the second operator.

### Claim 3 (depends on 2)

A system according to claim 2, wherein the determined feature vectors, the composite feature vector, the second feature vector and the second composite feature vector are a same size.

### Claim 4 (depends on 3)

A system according to claim 3, the at least one processing unit to execute the processor-executable program code to cause the system to:
determine the operator by inputting the composite feature vector to a deep reinforcement learning network; and
determine the second operator by inputting the second composite feature vector to the deep reinforcement learning network.

### Claim 5 (depends on 1)

A system according to claim 1, wherein the determined feature vectors and the composite feature vector are a same size.

### Claim 6

A method comprising:
determining a plurality of features of one or more database tables;
for each of the plurality of features, determining a feature vector based on a taxonomy of logical entities;
combining the determined feature vectors of each of the plurality of features into a composite feature vector;
determining an operator based on the composite feature vector; and
determining a new feature based on the operator and the plurality of features,
wherein each entry of each of the feature vectors corresponds to a respective one of the logical entities, and
wherein combining the determined feature vectors comprises determining, for each one of the logical entities, a composite value based on the entry of each of the feature vectors which corresponds to the logical entity.

### Claim 7 (depends on 6)

A method according to claim 6, further comprising:
determining a second feature vector based on the new feature;
combining the second feature vector with the composite feature vector to generate a second composite feature vector;
determining a second operator based on the second composite feature vector; and
determining a third feature based on the second operator.

### Claim 8 (depends on 7)

A method according to claim 7, wherein the determined feature vectors, the composite feature vector, the second feature vector and the second composite feature vector are a same size.

### Claim 9 (depends on 8)

A method according to claim 8, further comprising:
determining the operator by inputting the composite feature vector to a deep reinforcement learning network; and
determining the second operator by inputting the second composite feature vector to the deep reinforcement learning network.

### Claim 10 (depends on 6)

A method according to claim 6, wherein the determined feature vectors and the composite feature vector are a same size.

### Claim 11

A non-transitory medium storing executable program code executable by at least one processing unit of a computing system to cause the computing system to:
determine a plurality of features, each feature associated with a database column;
for each of the plurality of features, determine a feature vector based on a taxonomy of logical entities;
combine the determined feature vectors into a composite feature vector;
determine an operator based on the composite feature vector; and
determine a feature based on the operator,
wherein each entry of each of the feature vectors corresponds to a respective one of the logical entities, and
wherein combination of the determined feature vectors comprises determination, for each one of the logical entities, of a composite value based on the entry of each of the feature vectors which corresponds to the logical entity.

### Claim 12 (depends on 11)

A medium according to claim 11, the program code executable by at least one processing unit of a computing system to cause the computing system to:
determine a second feature vector based on the determined feature;
combine the second feature vector with the composite feature vector to generate a second composite feature vector;
determine a second operator based on the second composite feature vector; and
determine a third feature based on the second operator.

### Claim 13 (depends on 12)

A medium according to claim 12, wherein the determined feature vectors, the composite feature vector, the second feature vector and the second composite feature vector are a same size.

### Claim 14 (depends on 13)

A medium according to claim 13, the program code executable by at least one processing unit of a computing system to cause the computing system to:
determine the operator by inputting the composite feature vector to a deep reinforcement learning network; and
determine the second operator by inputting the second composite feature vector to the deep reinforcement learning network.

### Claim 15 (depends on 11)

A medium according to claim 11, wherein the determined feature vectors and the composite feature vector are a same size.