Article Preview
Top1. Introduction
Ontology techniques still are important research area in information systems. Extensive application of ontology signalizes principle role of ontology on the system. Several researchers utilize ontology in biomedicine (e.g. Ruiz-Martínez et al., 2011), knowledge management and knowledge sharing (e.g. (Yoo & No, 2014)) and e-learning systems and generation of document (e.g. Gillani & Ko, 2015; Peroni, Shotton, & Vitali, 2013). Indeed, several researches applies ontology in recommender systems (e.g. Likavec, Osborne, & Cena, 2015; Cantador, Castells, & Bellogín, 2011).
Semantic web community relies on heterogeneous ontologies to dominate the crucial problem of semantic heterogeneity (Khiat & Benaissa, 2015). Hence, the ontology is mostly applied in semantic relatedness, semantic similarity and information retrieval. Hence, many researchers apply ontology in the applications of semantic relatedness and similarity (e.g. Sánchez & Batet, 2013; Kara et al., 2012).
Moreover, many researchers attempt to apply ontology and compute similarity between concept pairs (e.g. Likavec, Osborne, & Cena, 2015; Sánchez, Batet, Isern, & Valls, 2012; Sánchez & Batet, 2013). Singer et al. computes Semantic Relatedness from Human Navigational Paths (Singer, Niebler, Strohmaier, & Hotho, 2013).
Generally, ontology is formed on graph and concepts are nodes of the graph (Gauch, Chaffee, & Pretschner, 2003). These nodes are connected by taxonomic (i.e. is a link) or non-taxonomic links. The authors nominate this type of ontology as graph-based ontology. All the methods compute similarity based on graph-based ontology. Therefore, to compute similarity between concept pairs based on graph-based ontology, graph traversal is needed. Hence, time complexity of graph traversal has direct effect on the results performance.
In this paper, the authors introduce a new representation of ontology based on N-dimensional Euclidean space. By this means, the authors apply N-dimensional cube as an N-dimensional matrix and propose Matrix-based ontology. In fact, Matrix-based ontology is an ontology which is established based on N-dimensional matrix. The N-dimensional matrix provides a condition which the authors can specify position of any concept of ontology. Hence, each concept in graph-based ontology, which is shown as a node, is an entity in Matrix-based ontology. In addition, depth of ontology is matched with number of dimensions in matrix (i.e. N). Matrix-based ontology is capable to decrease time complexity of computation which has direct effect on computation similarity.
The authors also introduce a new measure to compute similarity which is performed on N-dimensional Matrix-based ontology. Our proposed measure attempts to present high accuracy with low complexity in similarity computation. The proposed measure is evaluated with the three aforementioned methods (i.e. Edge-Counting method, Feature-based method and Information-Content method). Furthermore, the authors collect the reported results of the related studies and compare the results with the results of our model by using two well-known benchmarks. Moreover, the accuracy of our model with different dimensions is presented and compared with the related studies and benchmarks.
The remaining sections of this paper are organized as follows: In related studies section, the authors explain the related studies concerning ontology and ontology-based semantic similarity methods. In the next section, the problem is explained. Our proposed model which is called N-dimensional Matrix-based ontology is introduced in fourth section. In fifth section, the authors propose a new measure to compute semantic similarity between concept pairs which is applied in N-dimensional Matrix-based ontology. The results and validation of the results are illustrated in the next section which a 5-dimensional Matrix-based ontology is exemplified as the results; moreover, the results are validated with two well-known benchmarks. In discussion section, the discussion regarding the advantages and disadvantages of our model is provided. The conclusion of this paper will be presented in the last section.