روش نوين انطباق هستي‌شناسي با استفاده از پيكره‌هاي متني

A New Method for Ontology Matching by Using Textual Corpus

امروزه، استفاده از هستي‌شناسي براي مقاصد گوناگوني در حال گسترش است. اما، در مواقعي به خاطر تفاوت‌هايي كه در ساخت هستي‌شناسي در مراكز مختلف وجود دارد، امكان تبادل دانش بين دو هستي‌شناسي ميسر نيست. براي حل اين مشكل، روش‌هاي مختلفي براي انطباق هستي‌شناسي ارايه شده است كه برخي از آنها مبتني بر فنون يادگيري ماشين است. انطباق هستي‌شناسي يا به عبارت ديگر تشخيص مفاهيم متناظر در هستي‌شناسي‌هاي مختلف داراي كاربرد‌هاي متنوعي است. در اين مقاله، يك روش جديد ارايه شده است كه با بهره گيري از يادگيري ماشين و نيز پيكره هاي متني به عنوان منبع دانش از شباهت‌هاي معنايي بين هستي‌شناسي‌‌ها جهت انطباق استفاده مي‌كند.

The aim of ontology matching is to find similarities or matches between concepts of different ontologies. There are many new applications which need a sort of ontology matching. Some examples comprises of semantic web applications, multi agent systems, applications mash up and so on. One may be interested in either finding lexical similarity or semantic similarity, but in the both cases, the result of such a matching process can be useful for relating distinct ontologies. Leveraging ontology matching system enables us to reuse existing ontologies in new applications and save costs by eliminating the need for developing new ontologies. Among current algorithms proposed for matching onologies applying machine learning techniques is a promising one. However, there are some problems regarding the results of these methods which are mainly due to poor features used in learning process. In this paper we propose a new method in which a text corpus is used as the source of knowledge in conjunction with a machine learning method to find matching between two ontologies .The main objective in this new method is to find similarity of two concepts based on similarity of their instances. We show how contextual knowledge hidden in domain specific documents can help us to boost the machine learning methods by providing enough features. Also we show how taking benefit from this knowledge transcends the current approaches merely detect lexical similarity by either recognizing semantic similarity of concepts