Assessing trust over uncertain rules and streaming data

Decision makers (humans or software agents alike) are increasingly faced with the challenge of examining large volumes of information originating from heterogeneous sources requiring them to ascertain trust in various pieces of information. While several authors have explored various trust computation models on static data and certain rules, past work has typically assumed: (i) a statistically significant number of ratings are available prior to trust assessment, and (ii) assessed trust values tend to vary slowly over time. In contrast, military settings warrant: (i) trust assessment over partial, uncertain and streaming (live and real-time) information from heterogeneous sources, (ii) coping up with the dynamic and evolving nature of the ground truth, and (iii) and more importantly, rules used for making inferences may by themselves be uncertain. Within the context of executing the OODA loop for decision making our research objective is to develop a family of trust operators for dynamic information flows for assessing trust over data-in-motion rather than a large corpus of static data. In this paper, we show how to exploit the computational toolset of subjective logic to build a framework for trust assessment in this case. Furthermore, we describe an implementation of the framework (using Information Fabric [6] and Controlled English Fact Store[5]) and present an experimental evaluation that quantifies the efficacy with respect to accuracy and overhead of the proposed framework.