A multi-task learning model for Chinese-oriented aspect polarity classification and aspect term extraction

Aspect-based sentiment analysis (ABSA) task is a fine-grained of natural language processing and consists two subtasks: aspect term extraction (ATE) polarity classification (APC). Most the related works merely focus on subtask Chinese inferring fail to emphasize research Chinese-oriented ABSA multi-task learning. Based local context (LCF) mechanism, this paper firstly proposes learning model for aspect-based analysis, namely LCF-ATEPC. Compared with other models, equips capability extracting synchronously. The experimental results four review datasets outperform state-of-the-art performance ATE APC subtask. And by integrating domain-adapted BERT model, LCF-ATEPC achieves in most commonly used SemEval-2014 task4 Restaurant Laptop datasets. Moreover, effective analyze both English reviews collaboratively multilingual mixed dataset prove its effectiveness.