Aligning Multi-Cultural Knowledge Taxonomies by Combinatorial Optimization
Résumé
Large collections of digital knowledge have become valuable assets for search and recommendation applications. The taxonomic type systems of such knowledge bases are often highly heterogeneous, as they reflect different cultures, languages, and intentions of usage. We present a novel method to the problem of multi-cultural knowledge alignment, which maps each node of a source taxonomy onto a ranked list of most suitable nodes in the target taxonomy. We model this task as combinatorial optimization problems, using integer linear programming and quadratic programming. The quality of the computed alignments is evaluated, using large heterogeneous taxonomies about book categories.