Lessons learned in the creation of a data set for hard/soft information fusion

Abstract

Recent experiences in urban operations highlight the need for automated methods to help execute the data collection and association functions in the intelligence process. Technology development in this area has been challenged by the heterogeneity of data to be processed and by the diversity of potential applications. Here, we report lessons learned during the creation of a hard/soft data set. The data set was built using 'hard' physical sensor reports and 'soft' documents captured during training exercises simulating urban conditions. We discuss data collection issues, and we examine the information elements that can be exploited from the data. A key product in the data set is the identification of threads of insurgent activities that were observed by both hard and soft sensing modalities. We discuss the process by which the threads were identified, including an annotation procedure that structures the text information to enable association with the hard information elements.