# US 2021/0334309 A1 · Classification device, classification method, generation method, classification program, and generation program

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

- Number: US 2021/0334309 A1
- Document ID: US20210334309A1
- Status: Published application
- Kind: A1
- Application number: 16/486,660
- Filed: 2018-05-21
- Published: 2021-10-28
- Earliest claimed: 2018-05-21
- Assignees: Rakuten, Inc.
- Applicants: Rakuten, Inc.
- Inventors: Makoto TSUTSUMI; Koji MURAKAMI; Takashi UMEDA
- Web page: https://patentgrep.com/patent/US20210334309A1

## CPC

- G06F 16/906: Clustering; Classification
- G06F 40/30: Semantic analysis
- G06F 16/90332: {Natural language query formulation or dialogue systems}
- G06F 16/93: Document management systems
- G06F 40/295: Named entity recognition
- G06N 20/00: Machine learning
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Even when there are many candidates for a linked entity, accuracy of estimating an entity related to a term that appears on given writing is improved. A classification device stores a first model generated by using a corpus including pieces of first writing related to predetermined entities, and a second model generated by using a corpus including the pieces of first writing. Features of terms included in the first writing are acquired from the first model. Features of the pieces of first writing are acquired from the second model. The classification device causes the first model to output first information indicating entities indicated by terms, which has similarity in a feature to a second term included in second writing, among first terms indicating the predetermined entities. The classification device causes the second model to output second information indicating entities related to pieces of writing, which has similarity in a feature to the second writing, among the pieces of first writing. The classification device determines linking of the second term based on the first information and the second information.

## Claims

### Claim at position 1

1-10. (canceled)

### Claim 11

A classification device comprising:
at least one memory configured to store computer program code;
a storage that stores a first model which is generated by machine learning using a first corpus including one or more pieces of first writing related to one or more respective predetermined entities and from which features of terms included in the one or more pieces of first writing can be acquired, and a second model which is generated by machine learning using a second corpus including the one or more pieces of first writing and from which features of the one or more pieces of first writing can be acquired;
at least one processor configured to access said computer program code and operate as instructed by the computer program code, the computer program code including:
first control code configured to cause at least one of the at least one processor to cause the stored first model to output, in response to an input of a second term that is included in given second writing, first information indicating one or more first entities indicated by one or more respective terms, which have similarity in a feature to the second term, among one or more first terms indicating the one or more predetermined respective entities;
second control code configured to cause at least one of the at least one processor to cause the stored second model to output, in response to an input of the second writing, second information indicating one or more second entities related to one or more respective pieces of writing, which have similarity in a feature to the second writing, among the one or more pieces of first writing; and
determination code configured to cause at least one of the at least one processor to determine linking of the second term based on the output first information and the output second information.

### Claim 12 (depends on 11)

The classification device according to claim 11,
wherein the first corpus includes the one or more first terms, and
the second corpus has been generated by removing the one or more first terms from the first corpus and does not include the one or more first terms.

### Claim 13 (depends on 11)

The classification device according to claim 11,
wherein the first corpus is the second corpus,
the first model is the second model, and both of the features of the terms included in the one or more pieces of first writing and the features of the one or more pieces of first writing can be acquired from the first model, and
the second control code is configured to cause at least one of the at least one processor to cause the first model to output the second information.

### Claim 14 (depends on 11)

The classification device according to claim 11,
wherein the second control code is configured to cause at least one of the at least one processor to cause the second model to output the second information indicating the one or more second entities related to the one or more pieces of writing, which has relatively high similarity in a feature to the second writing, among the one or more pieces of writing related to the one or more first entities indicated by the first information output from the first model, and
the determination code is configured to cause at least one of the at least one processor to determine linking of the second term based on the second information.

### Claim 15 (depends on 11)

The classification device according to claim 11,
wherein the features of the terms included in the one or more pieces of first writing is acquired in a unit of a word from the first model, and the first model calculates, when the second term includes a plurality of words, a feature of the second term by a predetermined arithmetic operation with respect to a plurality of features of the plurality of respective words.

### Claim 16 (depends on 11)

The classification device according to claim 11,
wherein when outputting the second information, the second model acquires a feature of the second writing by machine learning using the second writing.

### Claim 17 (depends on 11)

The classification device according to claim 11,
wherein the linking of the second term includes determining an entity related to the second term among the one or more predetermined entities.

### Claim 18 (depends on 11)

The classification device according to claim 11,
wherein the linking of the second term includes determining whether there is an entity related to the second term among the one or more predetermined entities.

### Claim 19 (depends on 11)

The classification device according to claim 11,
wherein the determining code is configured to cause at least one of the at least one processor to determine an entity related to the second term among one or more third entities included in both of the one or more first entities and the one or more second entities.

### Claim 20

A classification method performed by a computer, the method comprising:
storing, in a storage, a first model which is generated by machine learning using a first corpus including one or more pieces of first writing related to one or more respective predetermined entities and from which features of terms included in the one or more pieces of first writing can be acquired;
storing, in the storage, a second model which is generated by machine learning using a second corpus including the one or more pieces of first writing and from which features of the one or more pieces of first writing can be acquired;
causing the stored first model to output, in response to an input of a second term that is included in given second writing, first information indicating one or more first entities indicated by one or more respective terms, which have similarity in a feature to the second term, among one or more first terms indicating the one or more predetermined respective entities;
causing the stored second model to output, in response to an input of the second writing, second information indicating one or more second entities related to one or more respective pieces of writing, which have similarity in a feature to the second writing, among the one or more pieces of first writing; and
determining linking of the second term based on the output first information and the output second information.

### Claim 21

A generation method executed by a computer, the method comprising:
generating, by machine learning using a first corpus including one or more pieces of first writing related to one or more respective predetermined entities, a first model from which features of terms included in the one or more pieces of first writing can be acquired;
generating a second model, from which features of the one or more pieces of first writing can be acquired, by machine learning using a second corpus including the one or more pieces of first writing;
generating a classifier including the first model, the second model and a determination unit that determines linking of a second term that is included in given second writing; and
storing the generated classifier in a storage,
wherein the first model outputs, in response to an input of the second term, first information indicating one or more first entities indicated by one or more respective terms, which have similarity in a feature to the second term, among one or more first terms indicating the one or more predetermined respective entities,
the second model outputs, in response to an input of the second writing, second information indicating one or more second entities related to one or more respective pieces of writing, which have similarity in a feature to the second writing, among the one or more pieces of first writing,
the determination unit determines linking of the second term based on the output first information and the output second information, and
the classifier is read from the storage, and the first model and the second model included in the read classifier are used to output the first information and the second information.
