Context-Guided Self-supervised Relation Embeddings

A semantic relation between two given words a and b can be represented using two complementary sources of information: (a) the semantic representations of a and b (expressed as word embeddings) and, (b) the contextual information obtained from the co-occurrence contexts of the two words (expressed in the form of lexico-syntactic patterns). Pattern-based approach suffers from sparsity while methods rely only on word embeddings for the related pairs lack of relational information. Prior works on relation embeddings have pre-dominantly focused on either one type of those two resources exclusively, except for a notable few exceptions. In this paper, we proposed a self-supervised context-guided Relation Embedding method (CGRE) using the two sources of information. We evaluate the learnt method to create relation representations for word-pairs that do not co-occur. Experimental results on SemEval-2012 task2 dataset show that the proposed operator outperforms other methods in representing relations for unobserved word-pairs