A Rank-Based Similarity Metric for Word Embeddings

Enrico Santus (1, 2) , Hongmin Wang (2, 3) , Emmanuele Chersoni (4, 5) , Yue Zhang (2)

1 MIT - Massachusetts Institute of Technology
2 SIT - Singapore Institute of Technology [Singapore]
3 UC Santa Barbara - University of California [Santa Barbara]
4 LPL - Laboratoire Parole et Langage
5 AMU - Aix Marseille Université

Word Embeddings (WE) have recently imposed themselves as a standard for representing word meaning in NLP. Semantic similarity between word pairs has become the most common evaluation benchmark for these representations, with vector cosine being typically used as the only similarity metric. In this paper, we report experiments with a rank-based metric for WE, which performs comparably to vector cosine in similarity estimation and out-performs it in the recently-introduced and challenging task of outlier detection, thus suggesting that rank-based measures can improve clustering quality.

Domaines: Informatique [cs]; Informatique et langage [cs.CL]

Enrico Santus, Hongmin Wang, Emmanuele Chersoni, Yue Zhang. A Rank-Based Similarity Metric for Word Embeddings. 56th annual meeting of the Association for Computational Linguistics (ACL), Jul 2018, Melbourne, Australia. ⟨hal-01838253⟩