# US 2020/0349226 A1 · Dictionary Expansion Using Neural Language Models

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

- Number: US 2020/0349226 A1
- Document ID: US20200349226A1
- Status: Published application
- Kind: A1
- Application number: 16/403,232
- Filed: 2019-05-03
- Published: 2020-11-05
- Earliest claimed: 2019-05-03
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Petar Ristoski; Daniel Gruhl; Alfredo Alba; Anna Lisa Gentile; Ismini Lourentzou; Chad Eric DeLuca; Linda Ha Kato; Steven R. Welch; Chris Kau
- Web page: https://patentgrep.com/patent/US20200349226A1

## CPC

- G06F 40/242: Dictionaries
- G06F 40/30: Semantic analysis
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/08: Learning methods
- G06N 3/044: Recurrent networks, e.g. Hopfield 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

Embodiments relate to a system, program product, and method for iterative expansion and application of a domain-specific dictionary. One or more dictionary instances are applied against a text corpus. The dictionary is iteratively expanded and selectively populated with one or more additional dictionary instances, including semantically similar instances to the applied dictionary instances and extension instances contextually related to the applied dictionary instances. The iteratively expanded dictionary is applied to an unexplored corpus to identify matching corpus data to populated instances of the dictionary.

## Claims

### Claim 1

A computer system comprising:
a processing unit operatively coupled to memory;
an artificial intelligence (AI) platform in communication with the processing unit, the AI platform to manage dictionary expansion, including:
a dictionary manager to apply one or more dictionary instances against a text corpus;
a director operatively coupled to the dictionary manager, the director to iteratively expand a dictionary, including populate the dictionary with one or more additional dictionary instances, the expansion comprising:
identification of one or more semantically similar instances to the one or more applied dictionary instances, and selectively addition of the identified one or more semantically similar instances to the dictionary; and
extension of the applied dictionary instances, including identification of one or more extension instances contextually related to the one or more applied dictionary instances, and selective addition of the identified one or more extension instances to the dictionary; and
a text manager operatively coupled to the AI platform, the text manager to apply the iteratively expanded dictionary to an unexplored corpus to identify matching corpus data to populated instances of the dictionary.

### Claim 2 (depends on 1)

The computer system of claim 1, further comprising the text corpus represented as a neural language model, and wherein the identification of one or more semantically similar instances further comprises the director to represent the one or more dictionary instances as a vector and calculate similarity between the one or more dictionary instances in vector form and all instances in the text corpus.

### Claim 3 (depends on 2)

The computer system of claim 2, wherein the similarity calculation includes the director to use a cosine similarity measure.

### Claim 4 (depends on 1)

The computer system of claim 1, wherein the selective addition of the identified one or more contextually related extension instances to the dictionary includes the director to expand the dictionary beyond terms present in the text corpus.

### Claim 5 (depends on 4)

The computer system of claim 4, wherein expansion of the dictionary further comprises the director to generate new multi-token instances and replace tokens in an existing dictionary instance with similar tokens from the text corpus.

### Claim 6 (depends on 4)

The computer system of claim 4, further comprising the director to concatenate additional tokens to one or more existing dictionary instances with contextually related tokens from the text corpus.

### Claim 7

A computer program product for dictionary expansion management, the computer program product comprising a computer readable storage medium having program code embodied therewith, the program code executable by a processor to:
apply one or more dictionary instances against a text corpus;
iteratively expand a dictionary, including populate the dictionary with one or more additional dictionary instances, comprising:
identify one or more semantically similar instances to the one or more applied dictionary instances, and selectively add the identified one or more semantically similar instances to the dictionary; and
extend the applied dictionary instances, including identify one or more extension instances contextually related to the one or more applied dictionary instances, and selectively add the identified one or more extension instances to the dictionary; and
apply the iteratively expanded dictionary to an unexplored corpus to identify matching corpus data to populated instances of the dictionary.

### Claim 8 (depends on 7)

The computer program product of claim 7, further comprising the text corpus represented as a neural language model, and the program code to identify one or more semantically similar instances further comprises program code to represent the one or more dictionary instances as a vector and calculate similarity between the one or more dictionary instances in vector form and all instances in the text corpus.

### Claim 9 (depends on 8)

The computer program product of claim 8, wherein the similarity calculation includes the program code using a cosine similarity measure.

### Claim 10 (depends on 7)

The computer program product of claim 7, wherein the program code to selectively add the identified one or more contextually related extension instances to the dictionary includes expansion of the dictionary beyond terms present in the text corpus.

### Claim 11 (depends on 10)

The computer program product of claim 10, wherein expansion of the dictionary further comprises program code to generate new multi-token instances and replace tokens in an existing dictionary instance with similar tokens from the text corpus.

### Claim 12 (depends on 10)

The computer program product of claim 10, further comprising program code to concatenate additional tokens to one or more existing dictionary instances with contextually related tokens from the text corpus.

### Claim 13

A method comprising:
applying one or more dictionary instances against a text corpus;
iteratively expanding a dictionary, including populating the dictionary with one or more additional dictionary instances, comprising:
identifying one or more semantically similar instances to the one or more applied dictionary instances, and selectively adding the identified one or more semantically similar instances to the dictionary; and
extending the applied dictionary instances, including identifying one or more extension instances contextually related to the one or more applied dictionary instances, and selectively adding the identified one or more extension instances to the dictionary; and
applying the iteratively expanded dictionary to an unexplored corpus to identify matching corpus data to populated instances of the dictionary.

### Claim 14 (depends on 13)

The method of claim 13, further comprising the text corpus represented as a neural language model, and wherein identifying one or more semantically similar instances further comprises representing the one or more dictionary instances as a vector and calculating similarity between the one or more dictionary instances in vector form and all instances in the text corpus.

### Claim 15 (depends on 14)

The method of claim 14, wherein the similarity calculation includes using a cosine similarity measure.

### Claim 16 (depends on 13)

The method of claim 13, wherein selectively adding the identified one or more contextually related extension instances to the dictionary includes expanding the dictionary beyond terms present in the text corpus.

### Claim 17 (depends on 16)

The method of claim 16, wherein expanding the dictionary further comprises generating new multi-token instances and replacing tokens in an existing dictionary instance with similar tokens from the text corpus.

### Claim 18 (depends on 16)

The method of claim 16, further comprising concatenating additional tokens to one or more existing dictionary instances with contextually related tokens from the text corpus.
