# US 12,223,265 B2 · Explainable propaganda detection

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

- Number: US 12,223,265 B2
- Document ID: US12223265B2
- Status: Granted
- Kind: B2
- Application number: 17/855,248
- Filed: 2022-06-30
- Published: 2023-01-05
- Granted: 2025-02-11
- Earliest claimed: 2021-06-30
- Assignees: Qatar Foundation for Education, Science and Community Development; Massachusetts Institute of Technology
- Applicants: Qatar Foundation for Education, Science and Community Development; Massachusetts Institute of Technology
- Inventors: Preslav I. Nakov; Giovanni Da San Martino; Seunghak Yu
- Web page: https://patentgrep.com/patent/US12223265B2

## CPC

- G06F 40/205: Parsing
- G06F 40/30: Semantic analysis
- G06F 3/04842: Selection of displayed objects or displayed text elements (G06F3/0482 takes precedence)
- G06F 40/216: using statistical methods
- G06F 40/289: Phrasal analysis, e.g. finite state techniques or chunking
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Explainable propaganda detection is provided by parsing each sentence of a plurality of sentences in an article for structural details; identifying, via a machine learning model, dimensional features across the plurality of sentences based on the structural details; identifying, via the machine learning model, a propagandizing tactic demonstrated in each sentence of the plurality of sentences based a covariance score to the propagandizing tactic and the plurality of dimensional features identified for a given sentence; and displaying, in a user interface that includes the article, propaganda labels corresponding to the propagandizing tactic identified for each sentence of the plurality of sentences.

## Claims

### Claim 1

A method, comprising:
parsing each sentence of a plurality of sentences in an article for structural details, wherein the plurality of sentences represent a threshold number of sentences selected from a beginning of the article;
identifying, via a machine learning model, dimensional features across the plurality of sentences based on the structural details;
identifying, via the machine learning model, a propagandizing tactic demonstrated in each sentence of the plurality of sentences based on a covariance score to the propagandizing tactic and the dimensional features identified for a given sentence; and
displaying, in a user interface that includes the article, propaganda labels corresponding to the propagandizing tactic identified for each sentence of the plurality of sentences.

### Claim 2 (depends on 1)

The method of claim 1, wherein the propagandizing tactic is selected by the machine learning model from a list comprising:
name calling;
repetition;
slogans;
appeal to fear;
doubt;
exaggeration;
flag-waving;
loaded language;
association with an out-group;
bandwagon;
casual oversimplification;
obfuscation;
appeal to authority;
binary fallacy;
thought-terminating clichés;
red herrings;
strawman arguments;
whataboutisms; and
none.

### Claim 3 (depends on 2)

The method of claim 2, further comprising, in response to receiving a veracity selection from a user to include the veracity of information included in the sentence when identifying the propagandizing tactic demonstrated in each sentence of the plurality of sentences:
identifying whether statements included in each of the sentences of the plurality of sentences are true or false;
categorizing any sentences of the plurality of sentences that includes a false statement demonstrating a lying propagandizing tactic; and
selecting the propagandizing tactic for any sentences of the plurality of sentences that exclude false statements from the list.

### Claim 4 (depends on 2)

The method of claim 2, further comprising, in response to receiving a veracity selection from a user to ignore the veracity of information included in the sentence when identifying the propagandizing tactic demonstrated in each sentence of the plurality of sentences:
categorizing false statements identically to true statements when selecting the propagandizing tactic from the list.

### Claim 5 (depends on 1)

The method of claim 1, wherein the dimensional features include:
a relative position of a given sentence in the article relative to other sentences;
a semantic stance of the given sentence relative to a title of the article;
a sentiment of the given sentence; and
an article-level prediction of the article as a whole being propaganda.

### Claim 6 (depends on 5)

The method of claim 5, wherein the semantic stance is selected from a group including:
related agreeing;
related disagreeing;
related discussing; and
unrelated.

### Claim 7 (depends on 5)

The method of claim 5, wherein the sentiment is selected from a group including:
positive;
negative;
neutral; and
compound.

### Claim 8

A system, comprising:
a processor; and
a memory, that includes instructions that when executed by the processor perform operations, including:
parsing each sentence of a plurality of sentences in an article for structural details, wherein the plurality of sentences represent a threshold number of sentences selected from a beginning of the article;
identifying, via a machine learning model, dimensional features across the plurality of sentences based on the structural details;
identifying, via the machine learning model, a propagandizing tactic demonstrated in each sentence of the plurality of sentences based on a covariance score to the propagandizing tactic and the dimensional features identified for a given sentence; and
displaying, in a user interface that includes the article, propaganda labels corresponding to the propagandizing tactic identified for each sentence of the plurality of sentences.

### Claim 9 (depends on 8)

The system of claim 8, wherein the propagandizing tactic is selected by the machine learning model from a list comprising:
name calling;
repetition;
slogans;
appeal to fear;
doubt;
exaggeration;
flag-waving;
loaded language;
association with an out-group;
bandwagon;
casual oversimplification;
obfuscation;
appeal to authority;
binary fallacy;
thought-terminating clichés;
red herrings;
strawman arguments;
whataboutisms; and
none.

### Claim 10 (depends on 9)

The system of claim 9, further comprising, in response to receiving a veracity selection from a user to ignore the veracity of information included in the sentence when identifying the propagandizing tactic demonstrated in each sentence of the plurality of sentences:
categorizing false statements identically to true statements when selecting the propagandizing tactic from the list.

### Claim 11 (depends on 8)

The system of claim 8, wherein the dimensional features include:
a relative position of a given sentence in the article relative to other sentences;
a semantic stance of the given sentence relative to a title of the article;
a sentiment of the given sentence; and
an article-level prediction of the article as a whole being propaganda.

### Claim 12 (depends on 11)

The system of claim 11, wherein the semantic stance is selected from a group including:
related agreeing;
related disagreeing;
related discussing; and
unrelated.

### Claim 13 (depends on 11)

The system of claim 11, wherein the sentiment is selected from a group including:
positive;
negative;
neutral; and
compound.

### Claim 14

A memory apparatus that includes instructions that when executed by a processor perform operations, comprising:
parsing each sentence of a plurality of sentences in an article for structural details, wherein the plurality of sentences represent a threshold number of sentences selected from a beginning of the article;
identifying, via a machine learning model, dimensional features across the plurality of sentences based on the structural details;
identifying, via the machine learning model, a propagandizing tactic demonstrated in each sentence of the plurality of sentences based on a covariance score to the propagandizing tactic and the dimensional features identified for a given sentence; and
displaying, in a user interface that includes the article, propaganda labels corresponding to the propagandizing tactic identified for each sentence of the plurality of sentences.

### Claim 15 (depends on 14)

The memory apparatus of claim 14, wherein the propagandizing tactic is selected by the machine learning model from a list comprising:
name calling;
repetition;
slogans;
appeal to fear;
doubt;
exaggeration;
flag-waving;
loaded language;
association with an out-group;
bandwagon;
casual oversimplification;
obfuscation;
appeal to authority;
binary fallacy;
thought-terminating clichés;
red herrings;
strawman arguments;
whataboutisms; and
none.

### Claim 16 (depends on 15)

The memory apparatus of claim 15, the operations further comprising, in response to receiving a veracity selection from a user to ignore the veracity of information included in the sentence when identifying the propagandizing tactic demonstrated in each sentence of the plurality of sentences:
categorizing false statements identically to true statements when selecting the propagandizing tactic from the list.

### Claim 17 (depends on 14)

The memory apparatus of claim 14, wherein the dimensional features include:
a relative position of a given sentence in the article relative to other sentences;
a semantic stance of the given sentence relative to a title of the article;
a sentiment of the given sentence; and
an article-level prediction of the article as a whole being propaganda.

### Claim 18 (depends on 17)

The memory apparatus of claim 17, wherein the semantic stance is selected from a group including:
related agreeing;
related disagreeing;
related discussing; and
unrelated.

### Claim 19 (depends on 17)

The memory apparatus of claim 17, wherein the sentiment is selected from a group including:
positive;
negative;
neutral; and
compound.