Article Preview
TopIntroduction
The Spreading Activation technique (hereafter SA) introduced by (Collins & Loftus, 1975), in the field of psycho linguistics and semantic priming, proposes a model in which all relevant information is mapped on a graph as nodes with a certain “activation value.” Relations between two concepts are represented by a weighted edge. If a node is activated their activation value is spread to their neighbor nodes. This technique was adopted by the computer science community and applied to the resolution of different problems, see Section Related Work, and it is relevant to the medical fields sector in the scope of: (1) construction of hybrid semantic search engines; (2) ranking of information resources according to an input set of weighted resources; (3) recommendation of medical terms using well-known ontologies in a particular sector; and (4) decision-support easing the access to the information in large databases. Thus this technique provides a connectionist method to retrieve data like brain can do. Although SA is widely used, more specifically in recent years has been successfully applied to ontologies, a common and standard framework is missing and each third party interested in its application must to implement its own version (Todorova et al., 2009) of SA.
Taking into account the new information realm and the leading features of putting together the SA technique and the Semantic Web and Linked Data initiatives, new enriched services of searching, matchmaking, recommendation or contextualization can be implemented to fulfill the requirements of access information in different trending scopes (Lytras & de Pablos, 2011) like e-health, e-procurement, e-tourism or legal document databases (Berrueta, Labra, & Polo, 2006). More specifically in the e-health sector there is a growing need to automate the processes related to the tagging of electronic clinical records and to create tools for the clinical decision support. That is why SA is relevant to the field of clinical knowledge management technologies through its capacity to process large databases and exploit the know-how of previous records easing the recommendations of information resources.
The proposed work aims to provide a framework for SA to ease the configuration, customization and execution over graph-based structures and more specifically over RDF graphs and ontologies. It is relevant to medical systems access and interoperability due to the fact that this technique is based on a set of proven algorithms for retrieving and recommending information resources in large knowledge bases. Following the specific contributions of this work are listed: (1) study and revision of the classical constrained SA; (2) study and definition of new restrictions for SA applied to RDF graphs and ontologies; (3) implementation of a whole and extensible framework (called ONTOSPREAD) to customize and perform the SA based; (4) outlining of a methodology to configure and refine the execution of SA; and (5) an example of configuration and refinement applying SA over two well-known ontologies: GALEN and SNOMED-CT.
Organization
This paper is structured as follows: First we review the relevant work in medical systems and the common applications of SA. Then, we provide a description of the design and implementation of an open framework for SA technique, explaining the algorithm, restrictions, etc. Afterwards, we apply the ONTOSPREAD framework over the GALEN and SNOMED-CT ontologies to evaluate the SA technique for recommending concepts in medical systems. Finally, we evaluate the results of the previous executions and present some conclusions.
Since SA was introduced by (Collins & Loftus, 1975) in the field of psycho linguistics and semantic priming it has been applied to the resolution of problems trying to simulate the behavior of the brain using a connectionist method to provide an “intelligent” way to retrieve information and data.
The use of SA was motivated due to the research on graph exploration (Preece, 1981; Anderson, 1983). Nevertheless the success of this technique is especially relevant to the fields of Document (Turtle, 1991) and Information Retrieval (Cohen & Kjeldsen,1987). It has been also demonstrated its application to extract correlations between query terms and documents analyzing user logs (Cui, Wen, Nie, & Ma, 2003) and to retrieve resources amongst multiple systems (Schumacher, Sintek, & Sauermann, 2008) in which ontologies are used to link and annotate resources.