2021/09/15 by Sameer Bansal, Bansal, Sameer, Adrian Benton +1
Computer Science · #Advanced Graph Neural Networks #Computation and Language (cs.CL) #FOS: Computer and information sciences #I.2.7 #Machine Learning (cs.LG) #Natural Language Processing Techniques #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2109.07488
openalex publication_date 2021/09/15 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Nickel and Kiela (2017) present a new method for embedding tree nodes in the\nPoincare ball, and suggest that these hyperbolic embeddings are far more\neffective than Euclidean embeddings at embedding nodes in large, hierarchically\nstructured graphs like the WordNet nouns hypernymy tree. This is especially\ntrue in low dimensions (Nickel and Kiela, 2017, Table 1). In this work, we seek\nto reproduce their experiments on embedding and reconstructing the WordNet\nnouns hypernymy graph. Counter to what they report, we find that Euclidean\nembeddings are able to represent this tree at least as well as Poincare\nembeddings, when allowed at least 50 dimensions. We note that this does not\ndiminish the significance of their work given the impressive performance of\nhyperbolic embeddings in very low-dimensional settings. However, given the wide\ninfluence of their work, our aim here is to present an updated and more\naccurate comparison between the Euclidean and hyperbolic embeddings.\n