Phase II Amount
$1,380,811
ION, a proposed software platform for Influence Operations aNalysis, will help an end user uncover adversary or competitor strategy and improve capability to influence target audiences. It will help USG and commercial end users alike answer: What will an adversary or competitor do next and how can I preempt or counter them? Our proposed human-in-the-loop system for anticipating an adversary or competitorÂ’s behaviors will directly enable a new way of carrying out analysis of entitiesÂ’ events (actions) and communications in context to (a) identify tactics and appropriate counter-measures, and (b) inform a better understanding of strategy. This type of technology, built on methodologically sound frameworks for analysis and proven techniques from the commercial marketing and advertising world, will also provide improved rigor to the influence space, build on best practices, and create an environment for the development of high-quality standards in the field. The principal strength of our proposed technical approach for ION is the incorporation of multiple proven methods that reduce the burden on the analyst. The unique combination of these methods sets the system apart from other tools. In particular, pre-trained language models are revolutionizing the Natural Language Processing (NLP) landscape. ION will leverage these advancements to reduce the amount of input required from analysts and adapt to new domains. Similarly, the proposed techniques for integrated modeling and forecasting leverage proven time series methods with demonstrated efficacy in a wide range of data regimes. These methodological approaches provide a strong foundation for automating large components of the analytical process used in the case studies below. The novel application of state-of-the-art NLP methods is a core component of our technical solution. Large language models have advanced the state of the art in a wide range of benchmark tasks over recent years. Our approach uses these technologies to build a powerful foundation that is further aided by the incorporation of analyst feedback. The combination of these methods is relatively unexplored, particular in the joint analysis of communications and events; however, a large body of literature indicates they hold tremendous promise. A final unique attribute of our technical approach is the way we treat sequences of communications and events as a combined pool of complex signals to detect and forecast behavior. This allows for the application of a diverse set of computational methods to address the integrated modeling and forecasting challenge. Many of these methods have been proven over decades of research, thus minimizing risk. Other proposed techniques are derived from more recent breakthroughs and hold tremendous promise for a revolutionary advancement in jointly modeling communications and behaviors.
