Cross-Lingual Taxonomy Alignment with Bilingual Knowledge Graph Embeddings

Abstract

Recently, different knowledge graphs have become the essential components of many intelligent applications, but no research has explored the use of knowledge graphs to cross-lingual taxonomy alignment (CLTA), which is the task of mapping each category in the source taxonomy of one language onto a ranked list of most relevant categories in the target taxonomy of another language. In this paper, we study how to perform CLTA with a multilingual knowledge graph. Firstly, we identify the candidate matched categories in the target taxonomy for each category in the source taxonomy. Secondly, we find the relevant knowledge denoted as triples for each category in the given taxonomies. Then, we propose two different bilingual knowledge graph embedding models called BTransE and BTransR to encode triples of different languages into the same vector space. Finally, we perform CLTA based on the vector representations of the relevant RDF triples for each category. Preliminary experimental results show that our approach is comparable and complementary to the state-of-the-art method.