An ontological analysis of uncertainty in soft data

This paper introduces several aspects to take into account in order to describe uncertainties in soft data. Various types of uncertainty are identified to express ambiguities of natural language, inaccuracies of information reporting, lack of knowledge or even misleading intentions. The analysis explores uncertainties inherent to soft data, along with contradictions or weightings revealed when information items are considered in combination. Quality measures corresponding to each aspect are developed by using the Uncertainty Representation and Reasoning Evaluation Framework (URREF) as a basis. While some criteria are intended to capture the way humans assess uncertainty through their use of language, some others highlight inconsistencies with respect to domain knowledge. A joint approach, based on natural language processing and shallow semantic analysis is proposed to evaluate those criteria. This analysis provides a richer description of soft data in terms of precision, ambiguity, vagueness, weight of evidence and consistence, suitable for high-level information fusion.