Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (14): 160-166.DOI: 10.3778/j.issn.1002-8331.2101-0043

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Specific Target Viewpoint Extraction Using Context Focused Mechanism

CHEN Yupeng, CHEN Jiawei, HUANG Rong, HAN Fang   

  1. 1.College of Information Science and Technology, Donghua University, Shanghai 201620, China
    2.Engineering Research Center of Digitalized Textile & Fashion Technology, Donghua University, Shanghai 201620, China
  • Online:2022-07-15 Published:2022-07-15

采用上下文专注机制的特定目标观点抽取

陈聿鹏,陈佳伟,黄荣,韩芳   

  1. 1.东华大学 信息科学与技术学院,上海 201620 
    2.东华大学 数字化纺织服装技术教育部工程研究中心,上海 201620

Abstract: Aiming at the problem that the existing target and viewpoint extraction models fail to fully consider the connection between the two, a specific target viewpoint extraction model based on contextual focus mechanism is proposed. First, it splices the extracted target feature vector with the context word vector at each location to form the final sentence representation, strengthens the interaction between the target and the sentence, and achieves target fusion; secondly, it uses the context focus mechanism to focus more attention around the target word, weakens the semantic features of distant words. The proposed model uses BiLSTM(bi-directional long short-term memory) network to encode sentences and extract features. Compared with existing models, the accuracy, recall and F1 value of the proposed model have been improved to a certain extent, which proves the effectiveness of the proposed algorithm. At the same time, the pre trained BERT model is applied in the current task to further improve the effect of the model.

Key words: target fusion, context focus mechanism, bi-directional long short-term memory(BiLSTM), BERT model

摘要: 针对现有的目标和观点抽取模型未能充分考虑两者的联系的问题,提出一种基于上下文专注机制的特定目标观点抽取模型。将抽取出的目标特征向量与每个位置的上下文词向量拼接构成最终的句子表示,加强目标与句子之间的交互,实现目标融合;采用上下文专注机制把注意力更多地放在目标词的周围,削弱远距离词的语义特征。提出的模型采用双向长短时记忆(bi-directional long short-term memory,BiLSTM)网络将句子编码,并提取特征。与现有模型相比,所提模型的精确率、召回率和F1值都有一定程度的提升,证明了所提算法的有效性。同时,预训练的BERT模型也被应用到当前任务中,使模型效果获得了进一步的提升。

关键词: 目标融合, 上下文专注机制, 双向长短时记忆(BiLSTM)网络, BERT模型