Semantic Annotation Using Horizontal and Vertical Contexts

Abstract

This paper addresses the issue of semantic annotation using horizontal and vertical contexts. Semantic annotation is a task of annotating web pages with ontological information. As information on a web page is usually two-dimensionally laid out, previous semantic annotation methods that view a web page as an ‘object’ sequence have limitations. In this paper, to better incorporate the two-dimensional contexts, semantic annotation is formalized as a problem of block detection and text annotation. Block detection is aimed at detecting the text block by making use of context in one dimension and text annotation is aimed at detecting the ‘targeted instance’ in the identified blocks using the other dimensional context. A two-stage method for semantic annotation using machine learning has been proposed. Experimental results indicate that the proposed method can significantly outperform the baseline method as well as the sequence-based method for semantic annotation.