Development of a conceptual graph-based information retrieval model for medical question databases

Evidence-based answers for family practice are concise answers to clinical questions from family physicians. Electronic availability of these answers could be used to improve patient care and lower costs if an automated computer system could be developed to access the information. To find the precise answer efficiently, the first step is to match the really asked questions with the prior well-built questions in a clinical question database. This match can direct the search to relevant and precise answers. In this thesis, we propose a new conceptual graph based information retrieval model. The clinical questions are represented as conceptual graphs, thus the information retrieval task can be expressed as a graph matching problem. To solve the problem that typical graph matching algorithms are of exponential complexity, we introduce a hybrid tree/ternary-tree approach of indexing and storage of conceptual graphs. An efficient algorithm to detect the largest common subgraph which uses the prior knowledge of the conceptual graph database is used to reduce the time taken during on-line retrieval. The semantic similarity measurement between two questions represented as conceptual graphs is also presented. Finally, we evaluate the retrieval performance of this conceptual graph based model against a clinical question database, and compare the performance with that of the classic vector space model.