Assessment of uncertainty in soft data: A case study

This paper presents a theoretical framework and preliminary results for the assessment of uncertainty in soft data. Soft data involves statements that contain incomplete, ambiguous and biased information. Many of those statements are subjective. For this work, an incremental model is developed to explore dimensions of soft data uncertainty, such as ambiguities of natural language, inaccuracies of reported information or even misleading intentions. If some dimensions are specific to humans' attitudes and use of language, some others highlight limitations in their perceptions or lack of knowledge. The model identifies conceptual and linguistic characteristic of each dimension and provides a richer description of soft data uncertainties in terms of vagueness, ambiguity, self-confidence and weight of evidence. Furthermore, a joint approach based on natural language techniques and semantic analysis is used to estimate those criteria. This approach was adopted to carry out a case study aiming at assessing the uncertainty of HUMINT messages. The results reveal an overall promising picture of the presence and the characterization of uncertainty in soft data, suitable for high level information fusion.