Machine Learning Influence Operation Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for analyzing information to detect influence operations are inadequate, as they often rely on sentiment analysis, which may fail to identify influence operations with positive sentiment or those that are not detected by sentiment analysis.
Innovation Solution
A method and system that utilize a supervised machine-learning framework to detect and analyze influence operations by receiving content items from internet sources, processing them through primary and secondary machine-learning models, and identifying associations with predefined influence operations and diverse narratives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sentiment analysis is used to detect influence operations, then negative sentiment influence operations can be identified, but positive sentiment influence operations remain undetected
Solution Approach 1:
The patent divides the detection system into multiple specialized models: a primary machine-learning model for identifying influence operations regardless of sentiment, and secondary machine-learning models for analyzing diverse narratives. This segmentation allows each model to specialize in specific aspects, improving overall detection reliability while covering both positive and negative sentiments without relying solely on sentiment analysis
Solution Approach 2:
The primary machine-learning model is designed to detect influence operations across all sentiment types (positive, negative, and neutral), making it a universal detector that doesn't rely on sentiment polarity. This multi-functional approach ensures that both positive and negative influence operations are detected with high accuracy, resolving the limitation of sentiment-based detection
2Reliability
If a single machine-learning model is used for detection, then the system remains simple, but detection accuracy and narrative analysis capability are insufficient
Solution Approach 1:
The patent implements a hierarchical multi-model architecture where a primary model performs initial influence operation detection, and secondary models perform specialized narrative analysis on detected content. This segmentation of functions across multiple models achieves high detection accuracy (85-91%) and comprehensive narrative analysis while keeping each individual model relatively simple and manageable
Solution Approach 2:
The system introduces an intermediary processing stage where the primary model's detections are passed to secondary models for further narrative analysis. This intermediary layer enables complex multi-stage analysis without requiring a single overly complex model, maintaining modularity and ease of maintenance while achieving high overall accuracy
Data Source
AI summary
Methods and systems for detecting influence operations are provided. In some examples, methods include receiving a plurality of content items, and providing each content item of the plurality of content items to a primary machine-learning model which is trained to determine whether one or more content items are associated with one or more predefined influence operations. The method further includes receiving, from the primary machine-learning model, an indication that at least one content item of the plurality of content items is associated with the one or more predefined influence operations, and providing the at least one content item to at least one secondary machine-learning model which is trained to determine whether one or more content items are associated with one or more predefined diverse narratives for the one or more predefined influence operations. In some examples, each predefined diverse narrative corresponds to a diagnostic frame and/or a prognostic frame.