# US 2024/0004880 A1 · Data augmentation system, data augmentation method, and information storage medium

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

- Number: US 2024/0004880 A1
- Document ID: US20240004880A1
- Status: Published application
- Kind: A1
- Application number: 18/337,061
- Filed: 2023-06-19
- Published: 2024-01-04
- Earliest claimed: 2022-06-29
- Assignees: Rakuten Group, Inc.
- Applicants: Rakuten Group, Inc.
- Inventors: Chen ZHAO; Yuki NAKAYAMA; Koji MURAKAMI
- Web page: https://patentgrep.com/patent/US20240004880A1

## CPC

- G06F 16/2455: Query execution
- G06F 16/3322: {using system suggestions (G06F16/3325 takes precedence)}
- G06F 16/24542: {Plan optimisation}
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/047: Probabilistic or stochastic networks
- G06N 3/0475: Generative networks
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Provided is a data augmentation system including at least one processor configured to: acquire a first search query including a first named entity, which was actually input in a search executed in a past; execute training of a first model that outputs a second search query including a virtual second named entity, which is different from the first named entity, based on the first search query; and execute data augmentation based on the second search query output by the trained first model.

## Claims

### Claim 1

A data augmentation system, comprising at least one processor configured to:
acquire a first search query including a first named entity, which was actually input in a search executed in a past;
execute training of a first model that outputs a second search query including a virtual second named entity, which is different from the first named entity, based on the first search query; and
execute data augmentation based on the second search query output by the trained first model.

### Claim 2 (depends on 1)

The data augmentation system according to claim 1,
wherein the first model comprises a model having output determined in one of a stochastic manner or a random manner, and
wherein the at least one processor is configured to execute the data augmentation based on the second search query output in one of the stochastic manner or the random manner by the trained first model.

### Claim 3 (depends on 2)

The data augmentation system according to claim 2,
wherein the first model comprises a variational auto-encoder (VAE) model having output determined in one of the stochastic manner or the random manner with a standard normal distribution followed by a probability density function to which a latent variable conforms, and
wherein the at least one processor is configured to execute the data augmentation based on the second search query output in one of the stochastic manner or the random manner by the trained VAE model.

### Claim 4 (depends on 3)

The data augmentation system according to claim 3,
wherein the VAE model includes:
an encoder configured to learn mapping from the first search query input during the training to an array of a cell state of a long short-term memory (LSTM) assumed to be a normal distribution; and
a decoder configured to learn mapping from an average value of the cell state in the array to a third search query output during the training, and
wherein the first model is configured to cause the decoder to generate the second search query by setting the average value in the array sampled from the normal distribution as an initial cell state.

### Claim 5 (depends on 2)

The data augmentation system according to claim 2, wherein the at least one processor is configured to execute the training so that, when the first search query is input to the first model, a third search query that is the same as the first search query is output from the first model.

### Claim 6 (depends on 1)

The data augmentation system according to claim 1, wherein the at least one processor is configured to:
compare each of a plurality of fourth search queries actually input in the search to a dictionary database relating to the first named entity; and
acquire, as the first search query, each of the plurality of fourth search queries that matches the dictionary database.

### Claim 7 (depends on 1)

The data augmentation system according to claim 1,
wherein the first search query includes a plurality of characters,
wherein each of the plurality of characters is associated with a tag relating to the first named entity, and
wherein the at least one processor is configured to execute the training based on the tag associated with each of the plurality of characters.

### Claim 8 (depends on 1)

The data augmentation system according to claim 1,
wherein the first search query is associated with a first attribute relating to the first named entity, and
wherein the at least one processor is configured to execute the training based on the first search query and the first attribute.

### Claim 9 (depends on 1)

The data augmentation system according to claim 1, wherein the at least one processor is configured to:
determine whether a predetermined condition is satisfied based on the second search query; and
execute the data augmentation when it is determined that the predetermined condition is satisfied.

### Claim 10 (depends on 9)

The data augmentation system according to claim 9, wherein the at least one processor is configured to determine whether the predetermined condition is satisfied by determining whether the virtual second named entity included in the second search query has a predetermined number of characters or more.

### Claim 11 (depends on 9)

The data augmentation system according to claim 9, wherein the at least one processor is configured to determine whether the predetermined condition corresponding to types of characters in the virtual second named entity included in the second search query is satisfied.

### Claim 12 (depends on 1)

The data augmentation system according to claim 1, wherein the at least one processor is configured to execute the data augmentation by augmenting training data for a second model that recognizes a fifth named entity included in a fifth search query based on the second search query.

### Claim 13 (depends on 12)

The data augmentation system according to claim 12, wherein the at least one processor is configured to:
estimate a second attribute relating to the virtual second named entity based on the second model that has learned the training data; and
augment the training data based on the second search query and the second attribute.

### Claim 14 (depends on 13)

The data augmentation system according to claim 13,
wherein the training data includes the first search query and a first attribute relating to the first named entity, and
wherein the at least one processor is configured to estimate the second attribute based on the second model that has learned the training data including the first search query and the first attribute.

### Claim 15 (depends on 12)

The data augmentation system according to claim 12, wherein the at least one processor is configured to execute additional training of the trained second model based on new training data output through the data augmentation.

### Claim 16 (depends on 1)

The data augmentation system according to claim 1, wherein the at least one processor is configured to execute the data augmentation by augmenting training data for a third model, which is different from the first model, based on the second search query.

### Claim 17 (depends on 1)

The data augmentation system according to claim 1, wherein the at least one processor is configured to acquire the first search query that was actually input in the search executed during a specific period in the past.

### Claim 18 (depends on 1)

The data augmentation system according to claim 1,
wherein the search comprises a search for a product in an electronic commerce service,
wherein the first search query comprises a search query actually input in the electronic commerce service, and
wherein the second search query includes the virtual second named entity relating to a virtual product.

### Claim 19

A data augmentation method, comprising:
acquiring a first search query including a first named entity, which was actually input in a search executed in a past;
executing training of a first model that outputs a second search query including a virtual second named entity, which is different from the first named entity, based on the first search query; and
executing data augmentation based on the second search query output by the trained first model.

### Claim 20

A non-transitory information storage medium having stored thereon a program for causing a computer to:
acquire a first search query including a first named entity, which was actually input in a search executed in a past;
execute training of a first model that outputs a second search query including a virtual second named entity, which is different from the first named entity, based on the first search query; and
execute data augmentation based on the second search query output by the trained first model.
