IP Library Granted Patent US 12,026,465
Granted Patent B2
US 12,026,465 · App. 17/553,800 · Granted Jul 2, 2024

Method and system for classifying word as obscene word

Inventor: Mikhail Borisovich Libman (Moscow, RU)
Assignee: Direct Cursus Technology L.L.C
G06F40/279G06F40/232G06F40/40G06V30/19147
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,026,465
App. No.
17/553,800
Granted
Jul 2, 2024
Kind
B2
Abstract

There is disclosed a method and system for classifying a word as an obscene word, the method comprising, at a training phrase: acquiring a first word, the first word corresponding to a given obscene word; generating a first set of misspelled words, the first set of misspelled words comprising a plurality of misspelled variations of the first word; generating a training pairs, the training pairs comprising: a set of positive training pairs comprising the first word paired with each misspelled variations of the first word; training a machine learning algorithm, the training comprising: determining, for each training pairs, a set of features representative of a property of the training pairs; generating an inferred function based on the set of features, the inferred function being configured to assign, in use, an indecency score, the decency score being indicative of a likelihood of the word being obscene.

Claims (59)

1. A computer implemented method for classifying a word as an obscene word, the method being executable by one or more servers, the method comprising, at a training phase:

acquiring a first word, the first word corresponding to a given obscene word;

generating a first set of misspelled words, the first set of misspelled words comprising a plurality of misspelled variations of the first word;

generating training pairs, the training pairs comprising:

a set of positive training pairs comprising the first word paired with each misspelled variations of the first word;

training a machine learning algorithm (MLA), the training comprising:

determining, for each of the training pairs, a set of features representative of a property of a given one of the training pairs; and

generating an inferred function based on the set of features, the inferred function being configured to assign, in use, an indecency score, the indecency score being indicative of a likelihood of the word being obscene.

2. The method of claim 1 , wherein the method further comprises:

acquiring a second word, the second word not corresponding to any obscene word;

generating a second set of misspelled words, the second set of misspelled words comprising a plurality of misspelled variations of the second word; and

wherein the training pairs further comprise a set of negative training pairs comprising the second word paired with each misspelled variations of the second word.

3. The method of claim 2 , wherein the first set of misspelled words and the second set of misspelled words are generated by executing a misspelling algorithm.

4. The method of claim 3 , wherein the misspelling algorithm is configured to generate the plurality of misspelled variations of the first word by at least one of:

substituting a given character within the first word with a typographical symbol;

switching one or more characters within the first word;

removing, one or more characters within the first word;

substituting the given character of a first alphabet within the first word with a corresponding character of a second alphabet, the first alphabet and the second alphabet being different alphabets.

5. The method of claim 4 , wherein the misspelling algorithm is created using a set of heuristics.

6. The method of claim 1 ,

further comprising, at an in-use phase:

acquiring, from an electronic device, a text comprising a plurality of words to be published on a web resource;

inputting the text into the MLA, the MLA being configured to assign the decency indecency score to each of the words included within the plurality of words; and

in response to the indecency score of at least one word of the text being above a predetermined threshold, applying a restrictive action.

7. The method of claim 6 , wherein the restrictive action is one of:

removing the at least one word prior to publishing the text on the web resource;

preventing the text from being published on the web resource.

8. The method of claim 6 , wherein the web resource is associated with a social media application.

9. The method of claim 6 , wherein the web resource is a live-chat application on a gaming platform.

10. The method of claim 1 , wherein the MLA is a neural network.

11. A system for classifying a word as an obscene word, the system comprising a server, the server comprising at least one processor and at least one non-transitory computer-readable memory storing instructions, which, when executed by the at least one processor, at a training phase, cause the system to:

acquire, a first word, the first word corresponding to a given obscene word;

generate, a first set of misspelled words, the first set of misspelled words comprising a plurality of misspelled variations of the first word;

generate training pairs, the training pairs comprising:

a set of positive training pairs comprising the first word paired with each misspelled variations of the first word;

train a machine learning algorithm (MLA) executable by the server, the training comprising:

determine, for each of the training pairs, a set of features representative of a property of a given one of the training pairs; and

generate an inferred function based on the set of features, the inferred function being configured to assign, in use, an indecency score, the indecency score being indicative of a likelihood of the word being obscene.

12. The system of claim 11 , wherein the at least one processor further causes the system to:

acquire a second word, the second word not corresponding to any obscene word;

generate a second set of misspelled words, the second set of misspelled words comprising a plurality of misspelled variations of the second word; and

wherein the training pairs further comprise a set of negative training pairs comprising the second word paired with each misspelled variations of the second word.

13. The system of claim 12 , wherein the first set of misspelled words and the second set of misspelled words are generated by executing a misspelling algorithm.

14. The system of claim 13 , wherein the misspelling algorithm is configured to generate the plurality of misspelled variations of the first word by at least one of:

substituting a given character within the first word with a typographical symbol;

switching one or more characters within the first word;

removing, one or more characters within the first word;

substituting the given character of a first alphabet within the first word with a corresponding character of a second alphabet, the first alphabet and the second alphabet being different alphabets.

15. The system of claim 14 , wherein the misspelling algorithm is trained using a set of heuristics.

16. The system of claim 11 , wherein the system further comprises a web resource hosted by the server; and wherein the at least one processor, at an in-use phase, further causes the system to:

acquire from an electronic device, a text comprising a plurality of words to be published on the web resource;

input, the text into the MLA, the MLA being configured to assign the indecency score to each of the words included within the plurality of words; and

in response to the indecency score of at least one word of the text being above a predetermined threshold, applying a restrictive action.

17. The system of claim 16 , wherein the restrictive action is one of:

removing the at least one word prior to publishing the text on the web resource;

preventing the text from being published on the web resource.

18. The system of claim 16 , wherein the web resource is associated with a social media application.

19. The system of claim 16 , wherein the web resource is a live-chat application on a gaming platform.

20. The system of claim 11 , wherein the MLA is a neural network.

Assignments (5)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Aug 9, 2024
From: DIRECT CURSUS TECHNOLOGY L.L.C
To: Y.E. HUB ARMENIA LLC
Reel/Frame 068534/0687 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Sep 15, 2023
From: YANDEX EUROPE AG
To: DIRECT CURSUS TECHNOLOGY L.L.C
Reel/Frame 065692/0720 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Dec 17, 2021
From: LIBMAN, MIKHAIL BORISOVICH
To: YANDEX.TECHNOLOGIES LLC
Reel/Frame 058418/0621 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Dec 17, 2021
From: YANDEX.TECHNOLOGIES LLC
To: YANDEX LLC
Reel/Frame 058418/0681 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Dec 17, 2021
From: YANDEX LLC
To: YANDEX EUROPE AG
Reel/Frame 058418/0741 →
Priority Claims (1)
RU RU2020142418 · Dec 22, 2020 · national
Continuity (1)
Related Publication 20220198143A1 · Jun 23, 2022