Institute of Information Science, Academia Sinica

In this work, we propose a framework for automatically extracting semantic relationships from heterogeneous, unstructured user-generated Web data, which are contributed from numerous users in different sources such as search engines, Open Directory Project (ODP), and Twitter. These heterogeneous data are integrated based on their most common characteristics- which could enrich the semantic meanings of the integrated data. Based on these semantic meanings, concepts are detected and relationships between associated concepts are constructed. The constructed relationships can be classified into: predefined relationships which are common in existing ontologies (such as ``is a,'' ``has subclass,'' and ``is part of,'') and user generated relationships. The experimental results demonstrate that our framework is able to extract semantic relationships with satisfactory accuracy.