Extracting meta-measures from data for fuzzy aggregation of crowd sourced information

Fuzzy measures (FMs) have been used to model the (typically subjective) “worth” of subsets of information sources relative to a decision making problem. The fuzzy integral (FI) is a way to fuse the information encoded in a FM with the (typically objective) confidences in the strength of a hypothesis arising from the information sources. In prior work, Yager discussed a set of aggregation functions for general FMs. However, that work is primarily focused on theoretical exploration versus application. Herein, we investigate the direct extraction of different FMs from data, one for specificity and another for agreement, in the context of crowd sourcing. In crowd sourcing, one often has a lack of a ground truth or information regarding the reliability of sources. That is, all sources must be assumed equal (in terms of knowledge, experience level, etc.). Our goal is the intelligent fusion of this data taking into account as much information as possible from the data itself. Once a set of FMs are extracted from the data, we aggregate the FMs (resulting in what we herein refer to as a meta-measure) and use it in fuzzy integration. The novel aspect of this work is the extraction of multiple FMs directly from the original pool of data and the use of the resultant meta-measure and a FI to fuse the data from which the FMs were extracted. Herein, our data is interval-valued, thus we focus on fusion with respect to the generalized interval FI.