A Framework for Knowledge Derivation Incorporating Trust and Quality of Data

Abstract

Today, across all major industries gaining insight from data is seen as an essential part of business. However, while data gathering is becoming inexpensive and relatively easy, analysis and ultimately deriving knowledge from it is increasingly difficult. In many cases, there is the problem of too much data such that important insights are hard to find. The problem is often not lack of data but whether knowledge derived from it is trustworthy. This means distinguishing "good" from "bad" insights based on factors such as context and reputation. Still, modeling trust and quality of data is complex because of the various conditions and relationships in heterogeneous environments. The new TrustKnowOne framework and architecture developed in this dissertation addresses these issues by describing an approach to fully incorporate trust and quality of data with all its aspects into the knowledge derivation process. This is based on Berlin, an abstract graph model we developed that can be used to model various approaches to trustworthiness and relationship assessment as well as decision making processes. In particular, processing, assessment, and evaluation approaches are implemented as graph expressions that are evaluated on graph components modeling the data. We have implemented and applied our framework to three complex scenarios using real data from public data repositories. As part of their evaluation we highlighted how our approach exhibits both the formalization and flexibility necessary to model each of the realistic scenarios. The implementation and evaluation of these scenarios confirms the advantages of the TrustKnowOne framework over current approaches.