AN APPROACH TO SEMANTIC-BASED MODEL DISCOVERY AND SELECTION

Model discovery and selection is an important step in component-based simulation model development. This paper proposes an efficient model discovery approach and quantifies the degrees of semantic similarity for selection of partially matched models. Models are represented as production strings as specified by an EBNF composition grammar. Together with a novel DHT overlay network, we achieve fast discovery of syntactically similar models with discovery cost independent of the model size. Next, we rank partially matched models for selection using semantic-based model attributes and behavior. Experiments conducted on a repository with 4,000 models show that on average DHT-based model lookup using production strings takes less than one millisecond compared with two minutes using naive string comparisons. Lastly, efficient model selection is a tradeoff between query representation and the computation cost of model ranking.