中圖分類號： TP391
文獻標識碼： A
DOI： 10.19358/j.issn.2096-5133.2020.06.005
引用格式： 宋婷，陳戰偉. 基于方面情感的層次化雙注意力網絡[J].信息技術與網絡安全，2020，39(6)：24-30.

Hierarchical double attention network for aspect sentiment analysis

Abstract： Aspect-level sentiment analysis based on deep learning, and a neural network model combined with attention mechanisms achieve good classification results. But common methods only consider single-level attention mechanisms and cannot obtain inter-sentence dependencies. In this paper，a hierarchical dual attention neural network model for aspect-level sentiment analysis was designed. It introduces the attention mechanism of aspect targets for specific aspects, and the text context self-attention mechanism to obtain aspect characteristic information and global dependency information of sentences; designs Hierarchical GRU network, in which the word layer embeds specific aspect information to obtain the internal characteristic information of the sentence for the aspect target，and the sentence layer network uses the dual attention mechanism and the input of the word layer to obtain the feature-dependent information between sentences and to achieve a deep level aspect emotion classification. A comparison experiment was performed on the two SemEval 2014 datasets and the Twitter dataset to verify the effectiveness of the method. The classification accuracy rate for aspect-level emotions was effectively improved.

Key words : hierarchical；dual attention；aspect emotion；attention mechanism；GRU