Charles River Analytics partners with Raytheon BBN on bias detection program

Charles River Analytics Inc., based in Cambridge, MA, announced on December 19 that it has partnered with Raytheon BBN to develop bias detection and credibility reporting tools for military analysts, who must quickly understand the value of data. We are applying our systemic functional grammar toolkit and linguistic feature-based machine learning approach to bias classification for DARPA’s Causal Exploration program. 

Currently, credibility assessment is an entirely subjective and manual effort for military analysts. It is also limited by the rate that a human can read. Recently, adversaries have increased the scale and effectiveness of propaganda and information operation campaigns. Evolving adversary actions require military analysts to adopt new approaches to identify systemic, wide-spread bias.

As the Causal Exploration program develops a modeling platform to help military analysts understand and address underlying causal factors that drive complex conflict situations, bias and reliability detection is a critical piece to understand what information to include and how to understand it in context.

“The US military needs better methods to deal with adversary scale and complexity; these adversarial techniques interweave into open and free press and information spaces,” said Robert Hyland, principal scientist at Charles River Analytics. “Alongside Raytheon BBN, we’re establishing the foundation for a generalized propaganda and information operation detection system.”

Our bias detection tools use high-fidelity bias classifiers that work at the sentence- or speaker-level. These tools further our growing portfolio of efforts in natural language processing. Under the ALARMM and THREAT efforts, we used advanced natural language processing to help detect, analyze, and act on complex information signals in news and technical documentation. These efforts are powered by our machine learning and systemic functional grammar toolkit capabilities. 

Source: Charles River Analytics