Adapting BERT for Continual Learning of a Sequence of Aspect Sentiment Classification Tasks

This paper studies continual learning (CL) of a sequence of aspect sentiment\nclassification (ASC) tasks. Although some CL techniques have been proposed for\ndocument sentiment classification, we are not aware of any CL work on ASC. A CL\nsystem that incrementally learns a sequence of ASC tasks should address the\nfollowing two issues: (1) transfer knowledge learned from previous tasks to the\nnew task to help it learn a better model, and (2) maintain the performance of\nthe models for previous tasks so that they are not forgotten. This paper\nproposes a novel capsule network based model called B-CL to address these\nissues. B-CL markedly improves the ASC performance on both the new task and the\nold tasks via forward and backward knowledge transfer. The effectiveness of\nB-CL is demonstrated through extensive experiments.\n