Thisiscompetition At Semeval-2019 Task 9: BERT Is Unstable For Out-of-domain Samples

Park Cheoneum, Kim Juae, Lee Hyeon-gu, Amplayo Reinald Kim, Kim Harksoo, Seo Jungyun, Lee Changki. Arxiv 2019

[Paper]    
BERT Model Architecture Pretraining Methods Security Training Techniques Transformer

This paper describes our system, Joint Encoders for Stable Suggestion Inference (JESSI), for the SemEval 2019 Task 9: Suggestion Mining from Online Reviews and Forums. JESSI is a combination of two sentence encoders: (a) one using multiple pre-trained word embeddings learned from log-bilinear regression (GloVe) and translation (CoVe) models, and (b) one on top of word encodings from a pre-trained deep bidirectional transformer (BERT). We include a domain adversarial training module when training for out-of-domain samples. Our experiments show that while BERT performs exceptionally well for in-domain samples, several runs of the model show that it is unstable for out-of-domain samples. The problem is mitigated tremendously by (1) combining BERT with a non-BERT encoder, and (2) using an RNN-based classifier on top of BERT. Our final models obtained second place with 77.78% F-Score on Subtask A (i.e. in-domain) and achieved an F-Score of 79.59% on Subtask B (i.e. out-of-domain), even without using any additional external data.

Similar Work