Frustratingly Easy Meta-Embedding -- Computing Meta-Embeddings by Averaging Source Word Embeddings

Coates, Joshua and Bollegala, Danushka ORCID: 0000-0003-4476-7003 (2018) Frustratingly Easy Meta-Embedding -- Computing Meta-Embeddings by Averaging Source Word Embeddings. In: NAACL-HLT, 2018-6-1 - 2018-6-6, New Orleans, USA.

Text average-meta-embedding.pdf - Author Accepted Manuscript Download (208kB)

Abstract

Creating accurate meta-embeddings from pre-trained source embeddings has received attention lately. Methods based on global and locally-linear transformation and concatenation have shown to produce accurate meta-embeddings. In this paper, we show that the arithmetic mean of two distinct word embedding sets yields a performant meta-embedding that is comparable or better than more complex meta-embedding learning methods. The result seems counter-intuitive given that vector spaces in different source embeddings are not comparable and cannot be simply averaged. We give insight into why averaging can still produce accurate meta-embedding despite the incomparability of the source vector spaces.

Item Type:	Conference or Workshop Item (Unspecified)
Additional Information:	Accepted to NAACL-HLT 2018
Uncontrolled Keywords:	cs.CL, cs.CL
Depositing User:	Symplectic Admin
Date Deposited:	12 Mar 2018 09:17
Last Modified:	19 Jan 2023 06:38
Related URLs:	Author
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3018858