Concept Matching in Informal Node-Link Knowledge Representations

Abstract

Information stored by managed organizations in free text documents, databases, and engineered knowledge repositories can often be processed as networks of conceptual nodes and relational links (concept graphs). However, these models tend to be informal as related to new or multi-source tasks. This work contributes to the understanding of techniques for matching knowledge elements: in informal node-link knowledge representations, drawn from existing data resources, to support user-guided analysis. Its guiding focus is the creation of tools that compare, retrieve, and merge existing information resources.Three essays explore important algorithmic and heuristic elements needed to leverage concept graphs in real-world applications. Section 2 documents an algorithm which identifies likely matches between student and instructor concept maps aiming to support semi-automatic matching and scoring for both classroom and unsupervised environments. The knowledge-anchoring, similarity flooding algorithm significantly improves on term-based matching by leveraging map structure and also has potential as a methodology for combining other informal, human-created knowledge representations. Section 3 describes a decompositional tagging approach to organizing (aggregating) automatically extracted biomedical pathway relations. We propose a five-level aggregation strategy for extracted relations and measure the effectiveness of the BioAggregate tagger in preparing extracted information for analysis and visualization. Section 4 evaluates an importance flooding algorithm designed to assist law enforcement investigators in identifying useful investigational leads. While association networks have a long history as an investigational tool, more systematic processes are needed to guide development of high volume cross-jurisdictional data sharing initiatives. We test path-based selection heuristics and importance flooding to improve on traditional association-closeness methodologies.Together, these essays demonstrate how structural and semantic information can be processed in parallel to effectively leverage ambiguous network representations of data. Also, they show that real applications can be addressed by processing available data using an informal concept graph paradigm. This approach and these techniques are potentially useful for workflow systems, business intelligence analysis, and other knowledge management applications where information can be represented in an informal conceptual network and that information needs to be analyzed and converted into actionable, communicable human knowledge.