Adaptive Experience-Based Composition of Continuously Changing Quality of Context

Contemporary systems increasingly rely on information provided by autonomous agents. The autonomous agents provide inherently inaccurate information due to, for example, rounding, calibration error or subjectivity. Moreover, the level of information inaccuracy may change without notice. Regardless the reason of the inaccuracy, a system relying on such information needs to adapt to the quality of the momentary information. In this paper, we propose a method for this adaption. The method bases on evidence theory and probability theory to compose a ground truth from disjoint information in a proposition. This ground truth is used to evaluate the disjoint information and determine this experience’s score. Each experience adds to the history of experiences in an agent, i.e., to the amount and character of evidence an agent has on another’s performance. Moreover, the method features a forgetting parameter facilitating adaption in case of, for example, maintenance to the providing agent. The method output is one parameter denoting the level of confidence the system certifies the composed information with. The presented method is validated by a case study on a dataset of in-house temperature data.