Social Media Influence Detection Using Post-URL Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting coordinated influence operations on social media are limited by their inability to distinguish between organic and coordinated activity, particularly as they often involve a mix of manual and automated actions, lack scalability, and fail to provide real-time detection.
Innovation Solution
A machine learning-based system that uses content-based features from post-URL pairs to identify coordinated influence efforts by training classifiers iteratively, incorporating manual verification, and retraining based on labeled data to improve detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If unsupervised machine learning approaches are used to identify coordinated influence efforts, then detection capability is improved, but scalability and real-time detection are worsened due to manual inspection requirements
Solution Approach 1:
The patent uses supervised machine learning models that are trained on labeled datasets of coordinated influence operations and organic activity. These models learn to automatically classify new posts without requiring manual inspection, thereby scaling the detection capability while maintaining accuracy. The trained models can process large volumes of social media data in real-time, resolving the contradiction between detection precision and scalability.
2Measurement precision
If supervised machine learning models are trained on labeled data, then detection accuracy is improved, but the ability to detect new and evolving tactics is worsened due to static training data
Solution Approach 1:
The patent implements a feedback mechanism where detected coordinated influence operations and their characteristics are fed back into the training process. The system continuously retrains supervised models with newly labeled data, allowing the models to adapt to evolving tactics and strategies used by influence operators. This continuous learning loop maintains high detection accuracy while improving adaptability to new threats.
Solution Approach 2:
The system employs dynamic model retraining where the supervised learning models are not static but are periodically updated with new training data. This allows the detection system to evolve alongside the tactics of influence operations, maintaining both accuracy and adaptability by incorporating lessons from newly detected campaigns.
3Reliability
If manual verification is incorporated into the detection process, then detection reliability is improved, but processing speed and real-time capability are worsened
Solution Approach 1:
The patent performs preliminary automated classification using supervised machine learning models before manual verification. The models pre-screen large volumes of data and identify high-confidence detections that can be processed automatically. Only cases that fall into uncertain categories or meet specific thresholds are forwarded for manual verification, thereby maintaining reliability while preserving processing speed for the majority of cases.
Data Source
AI summary
According to various embodiments, a machine learning based method, system, and non-transitory computer-readable medium for identifying content on social media related to one or more coordinated influence efforts are disclosed. The method includes generating one or more datasets of post-uniform resource locator (URL) pairs produced from one or more known coordinated influence efforts on one or more social media platforms. The method further includes generating one or more datasets of post-URL pairs produced from one or more random users on one or more social media platforms. The method additionally includes extracting a plurality of content-based features from the post-URL pairs from known coordinated influence efforts and random users. The method also includes iteratively training a classifier over a predetermined period of time to distinguish between a post-URL pair produced from a coordinated influence effort and a post-URL pair produced from a random user using the extracted plurality of content-based features.