Contextual Hierarchy Driven Ontology Learning

Contextual Hierarchy Driven Ontology Learning

Lobna Karoui (Ecole Supérieure d’Electricité, France)
Copyright: © 2009 |Pages: 22
DOI: 10.4018/978-1-60566-028-8.ch001
OnDemand PDF Download:
$37.50

Abstract

Research in ontology learning had always separated between ontology building and evaluation tasks. Moreover, it had used for example a sentence, a syntactic structure or a set of words to establish the context of a word. However, this research avoids accounting for the structure of the document and the relation between the contexts. In our work, we combine these elements to generate an appropriate context definition for each word. Based on the context, we propose an unsupervised hierarchical clustering algorithm that, in the same time, extracts and evaluates the ontological concepts. Our results show that our concept discovery approach improves the conceptual quality and the relevance of the extracted ontological concepts, provides a support for the domain experts and facilitates the evaluation task for them.
Chapter Preview
Top

Introduction

Some current work in data annotation, data integration, information retrieval, building multi-agents application, semantic web services depends on ontologies. The development and the deployment of these applications are related to the richness of the conceptualization inside the ontology. Moreover, giving a semantic to the web can be realized in an incremental manner by using the ontology vocabulary. In this paper, we focus on the use of context in the ontology learning process. Until now, most works have investigated various issues of ontology building such as methodology of ontology extraction and ontology evaluation but separately. In our research, we are interested in defining an approach that permits, at the same time, to extract and evaluate the ontological concepts by using the notion of “context”. Generally, the idea of defining the context of a word is limited to a sentence, a syntactic structure, a set of words or a set of sentences without tacking into account the document’s structure and the relation between the contexts. In this work, we combine these elements to establish an appropriate context definition for each word. So, our contributions are to propose firstly a new context definition that takes into account the document characteristics and secondly to define a new approach that not only extracts the ontological concepts but also evaluates them. This approach is based on our proposed context definition that takes into account the position of a word inside the HTML structure and the relation between the deduced contexts in the HTML document. For this purpose, we propose an unsupervised hierarchical clustering algorithm namely Contextual Concept Discovery (CCD) based on an incremental use of the partitioning Kmeans algorithm, guided by a structural context and producing a support for an easy evaluation task. Our context definition is based on the html structure and the location of each word in the documents. Each context is deduced from the various analyses included in the pre-processing step (Karoui et al., 2006). This explicitly contextual representation, titled contextual hierarchy, guides the clustering algorithm to delimit the context of each word by improving the word weighting, the words pair’s similarity and the semantically closer cooccurent selection for each word. By performing an incremental process and by recursively dividing each cluster, the CCD algorithm refines the context of each word cluster. It improves the conceptual quality of the extracted concepts. The CCD algorithm offers the choice between either an automatic execution or a user interactive one. In order to help the domain expert during the evaluation task, the last part of the CCD algorithm exploits a web collection and extracts the existing contexts. This information is used to compute the credibility degree associated to each word cluster in order to inform about it and facilitate the experts’ semantic interpretation. So, the CCD algorithm extracts the domain concepts and proposes a quantitative and a qualitative evaluation for the experts. Our evaluation proposition permits the ontology reuse and evolution since the informing elements that support the experts’ interpretation are driven by the web changes and are stored with the experts’ comments for a later use. We experiment the contextual clustering algorithm on French html document corpus related to the tourism domain. The results show that our algorithm improves the relevance of the extracted concepts in comparison with a simple Kmeans. Also, our observations and discussions with experts confirm that our evaluation process helps and assists the user.

The remainder of the paper is organized as follows: section 2 presents the state of the art related, sections 3 answers the question: What is the impact of the context on the intelligent interpretation, section 4 presents defines the context and the Contextual Concept Discovery” (CCD) algorithm, section 5 experiments our algorithm proposition and section 6 concludes and gives our future directions.

Complete Chapter List

Search this Book:
Reset