Using Concept Lattices to Reconcile Semantic Heterogeneity in Data

点击翻译按钮获取中文摘要

In modern information systems, data capture often cannot be closely controlled by traditional organizational mechanisms intended to enforce common meaning. Additionally, data are no longer used only for established and predetermined purposes. Indeed, one of the key promises of Big Data and Open Data is the opportunity to repurpose data for uses that were not anticipated when systems to collect and store the data were designed. This increasingly diverse information landscape creates significant challenges in integrating or combining information from various sources in ways that retain the semantics (meaning) of the underlying data. Integration difficulties arise from the independence of sources, resulting in a high level of semantic heterogeneity between sources. Achieving high-quality semantic integration is essential to take full advantage of the opportunities afforded by Big Data to support decision-making. The proposed research introduces the notion of “concept lattice” as a mechanism to provide semantics for data that can be used to integrate heterogeneous independent sources. In a concept lattice, nodes are concepts, interpreted as predicates applied to phenomena (instances) (e.g., Male(John)). A directed arc indicates a “precedence” relationship between two nodes; that is, possessing one property implies possessing another (e.g., Voter(John) -> Adult(John)). In this framework, the semantics of a node is entirely determined by the pattern of (and relationships among) arcs entering and leaving it. A concept lattice is a “lightweight” conceptual model of a domain, in contrast with traditional approaches that rely on “heavyweight” models (global or mediated schemas). Nodes may be considered either classes or properties, depending only on the pattern of incoming and outgoing arcs. This semantic relativism facilitates integration of concept lattices, where a concept may be a class in one lattice but a property in another. Moreover, two semantically equivalent nodes from heterogeneous sources can have different manifestations, indicating different semantic interpretations of the values of a property. The proposed research program will build on my earlier work by focusing on four objectives, working with graduate students and academic colleagues. First, we will formally define the core elements of a concept lattice structure, and develop reasoning mechanisms to navigate lattices. Second, we will design and implement a prototype to store and process concept lattices. Third, we will use concept lattices as a foundation to integrate information across heterogeneous sources by defining “merge nodes” – nodes from independent lattices that carry the same meaning and can be used to combine lattices. Fourth, we will evaluate the effectiveness of concept lattices in improving information retrieval from heterogeneous data sources.