Learning Meta Word Embeddings by Unsupervised Weighted Concatenation of Source Embeddings

04/26/2022
by   Danushka Bollegala, et al.
0

Given multiple source word embeddings learnt using diverse algorithms and lexical resources, meta word embedding learning methods attempt to learn more accurate and wide-coverage word embeddings. Prior work on meta-embedding has repeatedly discovered that simple vector concatenation of the source embeddings to be a competitive baseline. However, it remains unclear as to why and when simple vector concatenation can produce accurate meta-embeddings. We show that weighted concatenation can be seen as a spectrum matching operation between each source embedding and the meta-embedding, minimising the pairwise inner-product loss. Following this theoretical analysis, we propose two unsupervised methods to learn the optimal concatenation weights for creating meta-embeddings from a given set of source embeddings. Experimental results on multiple benchmark datasets show that the proposed weighted concatenated meta-embedding methods outperform previously proposed meta-embedding learning methods.

READ FULL TEXT

Please sign up or login with your details

Forgot password? Click here to reset