Domain post-training of BERT with MLM, NSP, an [EOT] token, variable fine-tuning, and data augmentation sets a new state of the art on Ubuntu Corpus V1 and the DSTC7 Advising Corpus.
Datasets and Training Setup We evaluate our model on two multi-turn dyadic data sets, Ubuntu IRC (Internet Relay Chat) Corpus V1[1] and Advis- ing Corpus1 [17]
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
An Effective Domain Adaptive Post-Training Method for BERT in Response Selection
Domain post-training of BERT with MLM, NSP, an [EOT] token, variable fine-tuning, and data augmentation sets a new state of the art on Ubuntu Corpus V1 and the DSTC7 Advising Corpus.