Content-Agnostic Fake News Detection via Website Structural Analysis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing automated fake news detection systems are limited by language barriers and require training on specific languages, making them ineffective for non-English news websites, and manual fact-checking is not scalable during events like elections or pandemics.

Innovation Solution

A content-agnostic method that uses machine learning to detect fake information items by analyzing structural, traffic, and network features of websites, such as IP-related activity, DNS-related details, and HTTP-related details, without relying on content, reputation, or sentiment, and can be deployed at either the ISP or user side for scalability and privacy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If automated fake news detection systems use linguistic analysis and content-based markers, then detection accuracy is improved for specific languages, but the systems become ineffective for non-English news websites due to language barriers

Engineering Contradiction:
Improvedetection accuracyVSAvoidlanguage coverage
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent segments the detection approach into two independent components: (1) content-agnostic structural analysis of website code, network traffic, and hosting patterns, and (2) optional linguistic analysis. This segmentation allows the system to operate effectively across all languages using structural features alone, while maintaining high detection accuracy by adding linguistic analysis when language-specific training data is available.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent creates a universal detection framework that functions across all languages and regions by relying on language-independent structural characteristics of fake news websites. The system analyzes common structural patterns, hosting behaviors, and network traffic features that are consistent regardless of language, making the detector universally applicable while maintaining effectiveness.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Reliability

If manual fact-checking is performed to ensure high detection accuracy, then reliability is improved, but scalability deteriorates during high-volume events like elections or pandemics

Engineering Contradiction:
Improvedetection reliabilityVSAvoiddetection scalability
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent implements self-service detection by training machine learning models on structural characteristics of fake news websites. Once trained, the system automatically detects and flags fake news without requiring manual fact-checking for each article. The models learn to identify patterns in website structure, hosting behavior, and network traffic, enabling autonomous, scalable detection that maintains reliability while handling high volumes during elections or pandemics.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent performs preliminary action by pre-training detection models on labeled datasets of fake and real news websites before deployment. This preliminary training phase captures the structural characteristics of fake news sites, enabling the system to rapidly detect new fake news without requiring real-time manual analysis. The pre-trained models can be quickly deployed and scaled to handle surge volumes during critical events.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If content analysis is performed to detect fake information, then detection precision is improved, but processing time increases and response time deteriorates

Engineering Contradiction:
Improvedetection precisionVSAvoidresponse time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent extracts and analyzes only the structural and metadata features of news websites, separating these from the actual content. By taking out the content analysis component and focusing solely on structural characteristics (website code, hosting patterns, network traffic), the system achieves detection precision while dramatically reducing processing time. This extraction approach allows rapid analysis without the computational burden of full content processing.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies partial action by analyzing only the necessary structural features required for detection rather than performing complete content analysis. The system extracts and processes only the relevant structural metadata and network characteristics needed to identify fake news patterns, achieving sufficient detection precision with minimal processing time and avoiding excessive analysis of full article content.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4195072A1Method, system and computer programs for detecting online fake information items
Publication Date: 2023.06.14 TELEFONICA INNOVACION DIGITAL SL
  • EP4195072A1 patent drawingFigure 1
  • EP4195072A1 patent drawingFigure 2
  • EP4195072A1 patent drawing

AI summary

A method, system and computer programs for detecting online fake information items are proposed. The method comprises collecting a plurality of information items from a plurality of websites using a cloud server; extracting, by the cloud server, structural, traffic and network features of each website of the plurality of websites, without analyzing any content of the information items, obtaining a feature dataset for each website as a result; executing, by the cloud server, a selection process on each feature dataset based on a performance and size criteria, obtaining a best-feature dataset for each website as a result; generating, by the cloud server, a filterlist including detected fake information items by means of executing a machine-learning classification process on each best-feature dataset.