Computing Terms Semantic Relatedness by Knowledge in Wikipedia

Abstract

Many researchers have recognized Wikipedia as a resource of huge dynamic knowledge base in recent years. This paper provides a new approach for obtaining measures of terms semantic relatedness, which maps terms to relevant Wikipedia articles as the background information for analyzing. The proposed algorithm WLA focuses on the hyperlink structure and summary paragraph extracted from the topic pages to compute two terms similarity. Comparing with other similar techniques, the approach is less computationally intensive, because only the first paragraph is analyzed, not the entire text. Our method achieves good performance on the widely used test set WS-353.