# US 11,113,468 B1 · Systems and methods for word segmentation based on a competing neural character language model

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

- Number: US 11,113,468 B1
- Document ID: US11113468B1
- Status: Granted
- Kind: B1
- Application number: 17/028,023
- Filed: 2020-09-22
- Granted: 2021-09-07
- Earliest claimed: 2020-05-08
- Assignees: Coupang Corp.
- Applicants: Coupang Corp.
- Inventors: Shusi Yu; Jing Li
- Web page: https://patentgrep.com/patent/US11113468B1

## CPC

- G06F 40/279: Recognition of textual entities
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/045: Combinations of networks
- G06N 3/048: Activation functions
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N 3/0895: Weakly supervised learning, e.g. semi-supervised or self-supervised learning
- G06Q 10/0877: {by inventory control or reporting using inventory tracking or counting}
- G06Q 30/0603: {Catalogue creation or management}
- G06Q 10/087: Inventory or stock management, e.g. order filling, procurement or balancing against orders
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR

## Abstract

Systems and methods are provided for detecting inaccuracy in a product title, comprising identifying, by running a string algorithm on a title associated with a product, at least one product type associated with the product, predicting, using a machine learning algorithm, at least one product type associated with the product based on the title, detecting an inaccuracy in the title, based on at least one of the identification or the prediction, and outputting, to a remote device, a message indicating that the title comprises the inaccuracy. Running the string algorithm may comprise receiving a set of strings, generating a tree based on the received set of strings, receiving the title, and traversing the generated tree using the title to find a match. Using the machine learning algorithm may comprise identifying words in the title, learning a vector representation for each character n-gram of each word, and summing each character n-gram.

## Claims

### Claim 1

A computer-implemented system for word segmentation, the system comprising:
a memory storing instructions; and
at least one processor configured to execute the instructions to:
receive a plurality of characters for segmentation;
convert, using an embedding model, a character of the plurality of characters into an embedding vector;
feed the embedding vector into a forward language model to retrieve a first output vector;
feed the embedding vector into a reverse language model to retrieve a second output vector;
compare the embedding vector to the first output vector and the second output vector; and
segment the plurality of characters based on the comparison,
wherein comparing the embedding vector to the first output vector and the second output vector comprises determining a reverse exponent of a Euclidean distance between the embedding vector and each of the first output vector and the second output vector.

### Claim 2 (depends on 1)

The system of claim 1, wherein the embedding model comprises a projection layer and a softmax classifier layer.

### Claim 3 (depends on 1)

The system of claim 1, wherein at least one of the forward language model or the reverse language model comprises a Recurrent Neural Network (RNN).

### Claim 4 (depends on 3)

The system of claim 3, wherein at least one of the forward language model or the reverse language model comprises at least one of a Gated Recurrent Unit (GRU) or a Long Short-Term Memory (LSTM).

### Claim 5 (depends on 1)

The system of claim 1, wherein the at least one processor is further configured to execute the instructions to feed a first context embedding vector sequence into the forward language model, the first context embedding vector sequence comprising a sequence of vectors associated with a predetermined number of characters preceding the character.

### Claim 6 (depends on 1)

The system of claim 1, wherein the at least one processor is further configured to execute the instructions to feed a second context embedding vector sequence into the reverse language model, the second context embedding vector sequence comprising a sequence of vectors associated with a predetermined number of characters following the character.

### Claim 7 (depends on 1)

The system of claim 1, wherein the at least one processor is configured to execute the instructions to pair the character with a predetermined number of characters preceding the character when the reverse exponent of the Euclidean distance between the embedding vector and the first output vector is greater than the reverse exponent of the Euclidean distance between the embedding vector and the second output vector.

### Claim 8 (depends on 1)

The system of claim 1, wherein the at least one processor is configured to execute the instructions to pair the character with a predetermined number of characters following the character when the reverse exponent of the Euclidean distance between the embedding vector and the second output vector is greater than the reverse exponent of the Euclidean distance between the embedding vector and the first output vector.

### Claim 9 (depends on 1)

The system of claim 1, wherein the at least one processor is configured to execute the instructions to repeat the steps of converting, feeding, comparing, and segmenting for each of the plurality of characters for segmentation until each character belongs to a word.

### Claim 10 (depends on 1)

The system of claim 1, wherein the at least one processor is configured to execute the instructions to predict a number of segments in the plurality of characters based on a number of characters in the plurality of characters and an average word length.

### Claim 11

A computer-implemented method for word segmentation, the method comprising:
receiving a plurality of characters for segmentation;
converting, using an embedding model, a character of the plurality of characters into an embedding vector;
feeding the embedding vector into a forward language model to retrieve a first output vector;
feeding the embedding vector into a reverse language model to retrieve a second output vector;
comparing the embedding vector to the first output vector and the second output vector; and
segmenting the plurality of characters based on the comparison,
wherein comparing the embedding vector to the first output vector and the second output vector comprises determining a reverse exponent of a Euclidean distance between the embedding vector and each of the first output vector and the second output vector.

### Claim 12 (depends on 11)

The method of claim 11, wherein the embedding model comprises a projection layer and a softmax classifier layer.

### Claim 13 (depends on 11)

The method of claim 11, wherein at least one of the forward language model or the reverse language model comprises a Recurrent Neural Network (RNN).

### Claim 14 (depends on 13)

The method of claim 13, wherein at least one of the forward language model or the reverse language model comprises at least one of a Gated Recurrent Unit (GRU) or a Long Short-Term Memory (LSTM).

### Claim 15 (depends on 11)

The method of claim 11, further comprising feeding a first context embedding vector sequence into the forward language model, the first context embedding vector sequence comprising a sequence of vectors associated with a predetermined number of characters preceding the character.

### Claim 16 (depends on 11)

The method of claim 11, further comprising feeding a second context embedding vector sequence into the reverse language model, the second context embedding vector sequence comprising a sequence of vectors associated with a predetermined number of characters following the character.

### Claim 17 (depends on 11)

The method of claim 11, further comprising pairing the character with a predetermined number of characters preceding the character when the reverse exponent of the Euclidean distance between the embedding vector and the first output vector is greater than the reverse exponent of the Euclidean distance between the embedding vector and the second output vector.

### Claim 18 (depends on 11)

The method of claim 11, further comprising pairing the character with a predetermined number of characters following the character when the reverse exponent of the Euclidean distance between the embedding vector and the second output vector is greater than the reverse exponent of the Euclidean distance between the embedding vector and the first output vector.

### Claim 19 (depends on 11)

The method of claim 11, further comprising repeating the steps of converting, feeding, comparing, and segmenting for each of the plurality of characters for segmentation until each character belongs to a word.

### Claim 20

A computer-implemented system for word segmentation, the system comprising:
a memory storing instructions; and
at least one processor configured to execute the instructions to:
receive a plurality of characters for segmentation;
convert, using an embedding model, a character of the plurality of characters into an embedding vector, wherein the embedding model comprises a projection layer and a softmax classifier layer;
feed the embedding vector into a forward language model to retrieve a first output vector;
feed the embedding vector into a reverse language model to retrieve a second output vector;
compare the embedding vector to the first output vector and the second output vector;
segment the plurality of characters based on the comparison; and
repeat the steps of converting, feeding, comparing, and segmenting for each of the plurality of characters for segmentation until each character belongs to a word,
wherein comparing the embedding vector to the first output vector and the second output vector comprises determining a reverse exponent of a Euclidean distance between the embedding vector and each of the first output vector and the second output vector.