State of the Art on Ontology Alignment

State of the Art on Ontology Alignment

Maria Vargas-Vera (Universidad Adolfo Ibanez, Chile) and Miklos Nagy (The Open University, UK)
DOI: 10.4018/978-1-4666-8751-6.ch099
OnDemand PDF Download:
List Price: $37.50


Ontology mapping as a semantic data integration approach has evolved from traditional data integration solutions. The core problems and open issues related to early data integration approaches are also applicable to ontology mapping on the Semantic Web community. Therefore, in this review the authors present the related literature, starting from the traditional data integration approaches, in order to highlight the evolution of data integration from the early approaches. Once the roots of semantic data integration have been presented, the authors proceed to introduce the state-of-the-art of the ontology mappings systems including the early approaches and the systems that can be compared through the Ontology Alignment Initiative (OAEI).
Chapter Preview

Importance Of The Problem

The problem of ontology mapping is important because its solution have an immediate application in search engines, question answering systems, etc. In fact the Information Retrieval community could in principle profoundly benefit from this proposed work. Based on the systems currently available at competitive level, we have visualized that the key challenges are as follows:

  • a.

    Robustness across domains: Most systems use multiple techniques such as heuristics, machine learning or Natural Language Processing (NLP) in order to transform the information in the ontologies into their internal representation. For example, ASMOV (Jean-Mary et al., 2009) uses domain specific background knowledge whereas RiMOM (Zhang et al., 2009) applies pre-defined rules to assess similarities. Anchor-Flood (Seddiqui & Aono, 2009) and TaxoMap (Hamdi etal., 2009) have been designed to exploit large textual descriptions of the ontology concepts, which is an assumption that cannot be satisfied across domains. These techniques have the problem that could impact domain independence negatively because they require a-priori knowledge from a designer. It is important to emphasize that ontology designers will always have the freedom to model their domain according to their need. In the same way, database designers can come up with different models for the same problem. To overcome this problem existing systems utilize various types of domain knowledge (heuristic rule or training data set).

  • b.

    Uncertain reasoning: Some ontology alignment systems provide limited reasoning support, in order to derive new information from the input ontologies. Unfortunately, not enough emphasis is placed on the reasoning part in spite of the fact that it has the potential to provide an added value to the systems. Furthermore, the uncertain reasoning possibility is completely missing from many of the existing systems.

  • c.

    Managing conflicts: Conflict detection is only provided as a post-processing functionality. However, conflicts that normally appear during the mapping process are not treated properly by the current solutions. Managing conflicting information does have the potential to improve the end results slightly or moderately depending on the domains.

  • d.

    Mapping optimization: Only two systems (RiMOM, TaxoMap) consider optimization of the mapping process, while the rest of the systems do not even consider it as a problem at this stage. This can be explained by the fact that most of the systems have not faced the problem of processing large-scale real world ontologies. While it is true that optimization issues can be addressed later on, it is important that the mapping solutions are conceived with scalability options.

  • e.

    Mapping visualization: Each system presents the mapping result to the users, although little emphasis has been placed on how these results are presented. Most systems show the results as a list of mapping pairs, and only some employ two-dimensional graph-based visualization. Additionally, there is no way to examine how the system produced these results as only the end results are kept.

Complete Chapter List

Search this Book: