An Approach to Automatically Constructing Domain Ontology

An Approach to Automatically Constructing Domain Ontology 1 Tingting He1 2 3 Xiaopeng Zhang 1 3 Xinghuo Ye1 3 1 Department of Computer Science, Huazhong Normal University 430079 Wuhan, China 2 Software College of Tsinghua University 102201 Beijing, China 3National Language Resources Monitor & Research Center (network media) 430079 Wuhan, China tthe@mail.ccnu.edu.cn zhangxiaopeng@mails.ccnu.edu.cn yxhez@tom.com Abstract. In this paper we present an approach to mining domain-dependent ontologies using term extraction and relationship discovery technology. There are two main innovations in our approach. One is extracting terms using log-likelihood ratio, which is based on the contrastive probability of term occurrence in domain corpus and background corpus. The other is fusing together information from multiple knowledge sources as evidences for discovering particular semantic relationships among terms. In the experiment, we also improve the traditional k-mediods algorithm for multi-level clustering. We have applied our approach to produce an ontology for the domain of computer science and obtained promising results. 1 Introduction Ontologies have become an important means for structuring knowledge and building knowledge-intensive systems. Ontologies have shown their usefulness in application areas such as intelligent information integration, information retrieval and natural language processing, to name but a few. For this purpose, efforts have been made to facilitate the ontology engineering process, in particular the acquisition of ontologies from domain texts [1]. Constructing an ontology is an extremely laborious effort. Even with some reuse of “core” knowledge from an Upper Model, the task of creating an ontology for a particular domain has a high cost, incurred for each new domain. Tools that could automate, or semi-automate, the construction of ontologies for different domains could dramatically reduce the knowledge creation cost. One approach to developing such tools is to rely on information implicit in collections of texts in a particular domain. If it were possible to automatically extract terms and their semantic relations from the text corpus, domain ontology could be built conveniently. This would be more cost-effective than having a human develop the ontology from scratch [2][3]. Domain specific ontologies are useful in systems involved with artificial reasoning, and information retrieval. Ontologies give such systems a vocabulary of terms, and concepts relating one term to another. In this paper, we present a method that automatically mining an ontology from any large corpus in a specific domain, to support data integration and information retrieval tasks. The induced ontology consists of domain concepts only related by parent-child links, not including more specialized relations. Our approach is comprised of two main phases: term extraction and relationship discovery. In the former phase, meaningful terms are extracted using LLR formula. In the latter one, parent-child relations among terms are induced through multilevel clustering. The remainder of the paper is as follows. In Section 2, we will discuss the related works in this field. In Section 3, we will give an overview of the overall system architecture and explain the means and progress of ontology construction. Subsequently, an example will show some promising results we have 1 The paper is supported …

Content truncated to 500 words