Knowledge discovery through ontology matching: An approach based on an Artificial Neural Network model

M. Rubiolo, M.L. Caliusco, G. Stegmayer, M. Coronel, M. Gareli Fabrizi
2012 Information Sciences  
With the emergence of the Semantic Web several domain ontologies were developed, which varied not only in their structure but also in the natural language used to define them. The lack of an integrated view of all web nodes and the existence of heterogeneous domain ontologies drive new challenges in the discovery of knowledge resources which are relevant to a user's request. New approaches have recently appeared for developing web intelligence and helping users avoid irrelevant results on the
more » ... b. However, there remains some work to be done. This work makes a contribution by presenting an ANN-based ontology matching model for knowledge source discovery on the Semantic Web. Experimental results obtained on a real case study have shown that this model provides satisfactory responses. Email addresses: CONICET, CIDISI-UTN-FRSF (M. Rubiolo, M.L. Caliusco, G. Stegmayer, M. Coronel and M. Gareli Fabrizi), mrubiolo@santafe-conicet.gov.ar (M. Rubiolo, M.L. Caliusco, G. Stegmayer, M. Coronel and M. Gareli Fabrizi) (2011) Information Sciences, 194, pp. 107-119. sinc(ialso an adequate structure to receive mobile software agents that will travel through the net (for example, looking for knowledge required by an end-user) [30] . Although the capabilities and scope of the Semantic Web are impressive today, its continuous evolution presents many problems to be faced. For instance, whenever a node on the Web needs to initiate a dynamic collaborative relationship with another, it certainly finds it difficult to know which node to contact or where to look for the required knowledge. Therefore, it can be seen that the knowledge resource discovery in such an open distributed system becomes a major challenge. This is due to the lack of an integrated view of all the available knowledge resources. Besides, the existence of multiple domain-specific heterogeneous ontologies and their distributed development introduces another problem: on the Semantic Web, many independently developed ontologies describing the same or very similar fields of knowledge, co-exist. These ontologies are either non-identical or present minor differences, such as different naming conventions or different structures in the way they represent knowledge. Any application that involves multiple ontologies must establish semantic mappings among them to ensure interoperability. Examples of such applications arise in many domains, including e-commerce, e-learning, information extraction, bioinformatics, web services, tourism, among others [15] . For that reason, ontology-matching is necessary to solve the problem. Ontology-matching techniques, essentially, identify semantic affinity between concepts belonging to different ontologies. Among recent proposals, machine learning methods can process the matching problem through the presentation of many correct (positive) and incorrect (negative) examples. Such algorithms require sample data from which to learn. Matchers using machine learning usually operate in two phases: (i) the learning or training phase and (ii) the classification or matching phase. During the first phase, training data for the learning process of a matcher is created, for example by manually matching two ontologies. During the second phase, the trained matcher is used for matching new ontologies. Learning can be processed on-line -the system can continuously learn-or off-line, so its speed is not as relevant as its accuracy. Usually, this process is carried out by dividing the available data set. For instance, considering a set of positive and negative examples of alignments, the examples would be divided into a training set (typically 80% of data) and a validation or test set (typically 20% of data), which would be used for evaluating the performance of the learning algorithm [17] . This paper presents a novel approach for improving web knowledge resource discovery on the Semantic Web based on recently developed intelligent techniques. This approach combines agentbased technologies with a machine-learning classifier (in particular an Artificial Neural Network (ANN) model) used for defining the matching operation between ontologies, and makes a proposal for codifying ontology knowledge as inputs to the neural model. The method proposed in this paper takes into account both schema-level and instance-level information from ontologies and semantic annotations, and proposes a machine learning solution to the ontology-matching problem, which has been proved in a real world case study, obtaining high matching rates. The paper is organized as follows. Section 2 introduces some issues related to ontology-matching and ANN concepts. Section 3 presents related work. In section 4, a motivating scenario for knowledge discovery is introduced, and an intelligent web-agent is presented. Section 5 explains the ANN-based ontology-matching model used by this agent. Section 6 presents the model evaluation by considering a case study. Section 7 compares the ANN-based model with the H-Match ontology matching algorithm. Finally, conclusions and future work can be found in Section 8. 2 sinc(i
doi:10.1016/j.ins.2011.08.008 fatcat:bzlrqiouvvbanl5jcphcnylnta