ADVAnced Analytics (AI/ML)

Online social networks (OSNs) allow users to freely share online posts, which may contain ideas, opinions, digital photos and videos, or breaking news on current events and activities. Information is diffused over OSNs when social media users examine input streams of posts and selectively share/forward them over their networks. This results in a cascade of information, as multimedia content spreads from user to user throughout the social network. These interactions produce large volumes and varieties of data that can quickly spread through numerous OSNs to millions of users. However, the open nature of OSNs enables users to freely self-publish online with little to no editing, fact checking, or accountability. This lack of accountability and verification gives OSN users the opportunity to deliberately spread deceptive and inaccurate information throughout social networks.

To provide early warning of emerging threats, IvySys develops social network analysis (SNA) tools that accurately model and predict the way information and behaviors propagate over online social networks. Orchestrated campaigns to spread misinformation across OSNs can produce different patterns of propagation than naturally occurring information diffusion behavior, and IvySys SNA tools leverage a modeling and simulation approach that exploits this fact. The goal is to accurately detect – early in the diffusion process – alarming social media signals representing orchestrated spreading of misinformation that will propagate widely in the future. IvySys SNA tools use model-based predictive analytic techniques based on digital signal processing and machine learning to understand and detect unfolding disinformation campaigns. IvySys SNA solutions leverage the following core competencies:

• Predictive analytics 

• Behavioral modeling and simulation

• Information diffusion modeling and simulation

• Social media signal processing

• Graph analytics

• Machine learning

• Synthetic social network data generation

• Social media data collection (while protecting personally identifiable information)

• Data visualization

Solution examples include:

Social Media Topic Propagation Prediction Tool – A Java-based social network analysis tool that enables the collection, generation, and analysis of the diffusion of social media topics. The social media data collection component, built on the IBM Streams platform, employs a customizable Twitter data ingestor that connects to Twitter's Streaming API and filters tweets according to keywords and other parameters. The resulting tweets are then automatically organized according to the original topic and subsequent retweets, and analyzed to predict their popularity. The social network generation and analysis component is a Java application with three main modes of operation: diffusion simulation, sampling analysis, and multi-message analysis. The diffusion simulation mode allows the user to simulate social media information diffusion scenarios subject to a number of model parameters. These parameters include:

• Number of social network nodes

• Diffusion model type 

• Connectivity statistical distribution for a node's contacts (e.g., Poisson, power law)

• Time-varying social influence of network nodes on friends

The sampling analysis mode allows the user to perform sampling analysis on data previously generated by a diffusion simulation. When using the sampling analysis mode, the user can modify the sampling rate and observe how that rate affects the computed value of message topic popularity. This mode enables users to determine the impact that social media message sampling has on prediction performance. The multi-message analysis mode is an extension of the Sampling Analysis mode, allowing the user to perform sampling analysis on multiple message topics at once and compare their computed message popularity values. Users can also vary how early or late in the information diffusion process to predict message popularity in order to determine the impact on prediction performance.