Knowledge representation is a subdomain within the field of artificial intelligence that is concerned with the structured representation of beliefs, intentions, and value judgments. The goal is to allow for computer interpretation and reasoning of this representation. Symbols are used in knowledge-based systems as computational placeholders for real world concepts, like disease, clinical events, and relationships. The extent to which these symbols reflect the real world are bounded by the domain of interest, which is the targeted portion of the real world to be modeled.
A knowledge base stores the symbolic notations associated with statements about a particular domain. In the clinical setting, this might allow for meshing of patient information with external information sources to aid the clinician with information that can inform decision making. An advanced knowledge base might be expected to process patient information actively to identify pertinent knowledge and provide actionable support, as well as facilitate semantic interoperability between data at different institutions. Knowledge representation forms are generally represented by logic and/or rules and semantic networks.
Forms of Knowledge Representation
Rules capture domain-specific knowledge through the use of if-then statements. The underlying principle of rules is derivation, which simply means that knowledge is derived from a given constellation of facts or concepts. Medical knowledge can be understood as declarative knowledge and procedural knowledge. Declarative knowledge is comprised of propositions and sentences. Propositions are statements that take on the value of “true” or “false.” To form sentences, propositions can be interrelated using Boolean operators. Procedural knowledge enumerates actionable information or conclusions that can be drawn from declarative knowledge. In this way, declarative knowledge comprise the “if” portion and procedural knowledge represents the “then” portion of if-then statements.
Semantic networks aim to capture knowledge as conceptual graphs, represent by vertices (aka nodes) and edges (aka arcs). Nodes represent classes of concepts, like clinician, or instances of these concepts, like resident physician. Edges between the nodes interrelate concepts. Through edges like inheritance, concepts can be organized into categories and larger hierarchies, making semantic networks useful for representation of taxonomic structures. Semantic networks are the underlying structure for knowledge graphs and Linked Data recommendations.
The Linked Data movement set out a list of guidelines for the sharing and integration of data sets. The idea is that any kind of data can be associated with ontologies using the Resource Description Framework, which connects data and also adds semantic information to the data.
Given a knowledge base, a computer could process that knowledge and derive new statements. Generally, there are two basic operations that a system can perofrm on its knowledge base. One adds a new statement to the base (tell), whereas the other queries what is known (ask). By processing over the statements it has been told from within a knowledge base, a system could generate implicit knowledge. For example, a knowledge base might explicitly state that “G6PD can cause red blood cell hemolysis” and that “hemolysis is a cause of anemia.” While not explicitly stated by the knowledge base, the answer to the question “Is G6PD a cause of anemia?” is yes, as the knowledge base entails this statement based on the previous two.
To allow for automated reasoning, ontologies are conceptual models that make the knowledge base of a domain available to computers. An ontology provides a representation of 1) entities within that domain, 2) the hierarchy of entities, and 3) the relations between entities.
- Knowledge Management
- Applied Ontology
- Transforming the EHR into a knowledge platform to ensure improved health and healthcare
- Barros M, Couto FM. Knowledge Representation and Management: a Linked Data Perspective. Yearb Med Inform. 2016;(1):178-183. doi:10.15265/IY-2016-022.
- Grimm S. Knowledge Representation and Ontologies. In: Gaber MM, ed. Scientific Data Mining and Knowledge Discovery: Principles and Foundations. Berlin, Heidelberg: Springer Berlin Heidelberg; 2010. doi:10.1007/978-3-642-02788-8.
- Kong G, Xu D, Yang J. Clinical Decision Support Systems: A Review on Knowledge Representation and Inference Under Uncertainties. Int J Comput Intell Syst. 2008;1(2):159-167. doi:10.1080/18756891.2008.9727613.
- Schulz S, Balkanyi L, Cornet R, Bodenreider O. From concept representations to ontologies: a paradigm shift in health informatics? Healthc Inform Res. 2013;19(4):235-242. doi:10.4258/hir.2013.19.4.235.
Submitted by Mark Mai