Multi-Cultural Interlinking of Web Taxonomies with ACROSS

Abstract

The Web hosts a huge variety of multi-cultural taxonomies. They encompass product catalogs of e-commerce, general-purpose knowledge bases and numerous domain-specific category systems. The enormous heterogeneity of those sources is a challenging aspect when multiple taxonomies have to be interlinked. In this paper we introduce ACROSS system to support the alignment of independently created Web taxonomies. For mapping categories across different taxonomies, ACROSS harnesses instance-level features as well as distant supervision from an intermediate source like multiple Wikipedia editions. ACROSS includes a reasoning step, which is based on combinatorial optimization. In order to reduce the run time of the reasoning procedure without sacrificing the quality, we study two models of user involvement. Our experiments with heterogeneous taxonomies for different domains demonstrate the viability of our approach and improvement over state-of-the-art baselines.