Journal of Beijing University of Posts and Telecommunications

  • EI核心期刊

Journal of Beijing University of Posts and Telecommunications ›› 2023, Vol. 46 ›› Issue (4): 123-128.

Previous Articles    

Aspect Category Classification Integrated in Syntactic Dependency and BERT-Att-BiLSTM

BAO Qianhui, WEN Juan, SHI Shuzhen, DONG Mengping, LIU Xue   

  • Received:2022-06-22 Revised:2022-08-18 Online:2023-08-28 Published:2023-08-24

Abstract: Current aspect category classification in fine-grained sentimental analysis suffers from low accuracy problems. An aspect category classification method is proposed integrates in syntactic dependency and bidirectional encoder representations from Transformers-attention mechanism-bidirectional short-term memory network ( BERT-Att-BiLSTM). Firstly, in the target information extraction layer, aspect-opinion pairs from comments is extracted by syntactic dependency. In the word embedding layer, BERT module is used to combined the dynamic features of context to achieve word vector representation. Then, in feature extraction layer, the BiLSTM module is integrated in attention mechanism to reduce the dimension of feature space. Finally, the aspect category is obtained at the classification layer through activation function. Experimental results show that the precision, recall rate, and F1 value of the proposed method reached 85.25% , 72.38% and 77.06% , surpassed the other chosen models and proved its effectiveness.

Key words: aspect category extraction , syntactic dependency , aspect category classification ,  bidirectional encoder representations from Transformers
,
attention mechanism

CLC Number: