The PRIOR+: Results for OAEI Campaign 2007

Ontology mapping is to find semantic correspondences between similar elements of different ontologies. It is critical to achieve semantic interoperability in the WWW. This paper summarizes the results of the PRIOR+ participating at OAEI campaign 2007. The PRIOR+ is a generic and automatic ontology mapping tool, based on propagation theory, information retrieval technique and artificial intelligence model. The approach utilizes both linguistic and structural information of ontologies, and measures the profile similarity of different elements of ontologies in a vector space model (VSM). Furthermore, the PRIOR+ adaptively aggregate different similarities according to the harmony of similarity matrix. Finally the PRIOR+ deals with ontology constraints using interactive activation and competitive neural network. The preliminary results of benchmark task are presented, followed by a discussion. Some future works are given at the end. 1 Presentation of the system 1.1 State, purpose, general statement The World Wide Web (WWW) now is widely used as a universal medium for information exchange. Semantic interoperability among different information systems in the WWW is limited due to information heterogeneity, and the non semantic nature of HTML and URLs. Ontologies have been suggested as a way to solve the problem of information heterogeneity by providing formal and explicit definitions of data. They may also allow for reasoning over related concepts. Given that no universal ontology exists for the WWW, work has focused on finding semantic correspondences between similar elements of different ontologies, i.e., ontology mapping. Automatic ontology mapping is important to various practical applications such as the emerging Semantic Web [3], information transformation and data integration [2], query processing across disparate sources [7], and many others [4]. Ontology mapping can be done either by hand or using automated tools. Manual mapping becomes impractical as the size and complexity of ontologies increases. Fully or semi-automated mapping approaches have been examined by several research studies, e.g., analyzing linguistic information of elements in ontologies [15], treating ontologies as structural graphs [12], applying heuristic rules to look for specific mapping patterns [8] and machine learning techniques [1]. More comprehensive surveys of ontology mapping approaches can be found in [9][14]. This paper proposes a new generic and scalable ontology mapping approach, the PRIOR+ approach. The architecture of the PRIOR+ is shown in Fig. 1. The PRIOR+ takes advantage of propagation theory, information retrieval technique and artificial intelligence model to solve ontology mapping problem. It utilizes both linguistic and structural information of ontologies, and measures the profile similarity of different elements of ontologies in a vector space model (VSM). Finally, the PRIOR+ adaptively aggregates different similarities according to the harmony of the matrix and deals with ontology constraints using interactive activation network. Fig. 1. The architecture of the PRIOR+ approach 1.2 Specific techniques used The PRIOR+ is extended from the PRIOR [10][11]. In addition to the profile similarity and the edit distance of elements’ name used in the PRIOR, the PRIOR+ considers structure similarity as well and adaptively aggregate different similarities based on their harmony. Furthermore, the PRIOR+ has a brand new NN-based Constraint Satisfaction Solver. 1.2.1 Similarity Generation The similarity generation model aims to generate the similarity of both linguistic and structural information of ontologies. The details of calculating profile similarity and the edit distance of elements’ name have been presented in the PRIOR [10][11]. To calculate the structure similarity of two elements, various structural features are extracted, e.g. the number of its sub-elements, the number of its direct property, the depth of the element to the root etc. Afterwards, the difference between these structural features are calculated and normalized to represent its structure similarity. The outputs of the similarity generation model are three similarity matrixes. Each matrix denotes a kind of similarity of two ontologies. 1.2.2 Harmony Estimation The heterogeneities of information result in differences between ontologies, either from a linguistic view or structural view. Therefore, given two ontologies, it is critical to estimate the difference between ontologies, and then to adjust mapping strategies according to the difference. Here we define a term called harmony to represent the similarity between ontologies. Three types harmony of ontologies, i.e. name harmony, profile harmony and structure harmony, are calculated based on the similarity matrixes output from similarity generation model. Ideally, if two ontologies are very similar in either linguistic or structural view, two true should-be-mapped elements should own a similarity equal to 1 or larger than the similarity of all other cells standing in the same row and column of those two elements in the corresponding similarity matrix. Therefore, the harmony of ontologies can be defined using Equation 1, where hk denotes different types of harmony (i.e., name harmony, profile harmony and structure harmony), 1 O E and 2 O E denote the number of elements in ontologies, O1 and O2, k M CMAX denotes the number of cells that own the highest similarity in its corresponding row/column in similarity matrix Mk.