Computer-implemented method for improving classification of labels and categories of a database

Patent metadata

Field	Value
Publication number	US-11983202-B2
Application number	US-202217734153-A
Country	US
Kind code	B2
Filing date	May 2, 2022
Priority date	Jun 29, 2021
Publication date	May 14, 2024
Grant date	May 14, 2024

Abstract

There is disclosed a method for using a computer to enable correction of misclassified labels in a database. The computer initially applies a dataset (including labels pointing respectively to categories) to a first classifier, which includes a first loss function. Pursuant to the initial application, the computer determines that one or more labels have been misclassified. Responsive to such determination, the computer changes the first loss function to a second loss function to form a second classifier including the second loss function. The computer then applies the dataset to the second classifier for enabling correction of the one or more misclassified labels.

First claim

What is claimed is:

1. A method implemented with one or more processors for improving classification of labels and categories of a database stored in memory that includes a set of labels and a set of categories where (1) each label in the set of labels points to at least one of the categories in the set of categories, and (2) each label in the set of labels is associated with a hierarchical category path, comprising:
the one or more processors applying both a subset of the set of labels and a subset of the set of categories of the database stored in the memory to a first classifier for classifying the subset of labels with respect to the subset of categories, the first classifier including a first loss function;
the one or more processors determining, based on said applying both the subset of labels and the subset of categories to the first classifier, whether at least one label in the subset of labels of the database stored in the memory has been misclassified;
in response to said determining that at least one label in the subset of labels of the database stored in the memory has been misclassified based on said applying both the subset of labels and the subset of categories to the first classifier:
the one or more processors changing the first loss function of the first classifier to a second loss function to form a second classifier including the second loss function, and
the one or more processors applying both the subset of labels and the subset of categories to the second classifier for classifying the subset of labels with respect to the subset of categories of the database stored in the memory for improving the second classifier's classification of labels and categories;

wherein the first loss function comprises a global categorical cross-entropy loss function, and wherein said changing the first loss function to the second loss function comprises changing the global categorical cross-entropy loss function to a weighted-by-sample categorical cross entropy loss function; and
wherein the weighted-by-sample categorical cross entropy loss function comprises:

ℒ
G
′

=

1
N

⁢

∑

i
=
1

N

a

y
^

i

,

y
i

⁢

ℒ

G
,
i

where:
G,i is the global catigorical cross-entory loss for sample i and

a

y
^

i

,

y
i

=

{

a
shorter

,

if
⁢

t
^

.

prefix_path

⁢
_of
⁢

(

t
i
′

)

a
longer

,

if
⁢

t
i
′

.

prefix_path

⁢
_of
⁢

(

t
^

i

)

1
,

otherwise

.

{circumflex over (t)} i is the path corresponding to the ŷ i prediction;
t i ′ is the observed path corresponding to y i ;
a shorter denotes the cost of predicting a shorter path than an observed one; and
a longer denotes the cost of predicting a longer path than the observed one.

2. The method of claim 1 , wherein the cost assigned to a shorter is greater than the cost assigned to a longer .

3. A method implemented with one or more processors for improving classification of labels and categories of a database, comprising:
the one or more processors storing in memory the labels and categories of the database that includes a set of labels and a set of categories where (1) each label in the set of labels points to at least one of the categories in the set of categories, (2) the labels in the set of labels are disposed in a label hierarchy with both the labels and the categories being arranged throughout a plurality of levels, and (3) each label in the set of labels is associated with a p

Assignees

• Naver Corp

Inventors

• Lagos Nikolaos
• Calapodescu Ioan
• Lee Jinhee
• Ait-Mokhtar Salah

Classifications

• G06F16/285 Primary

  Clustering or classification · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/082

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title