| 摘 要: 针对方面级情感分析中未能充分利用显式句法依赖的问题，提出基于语法依赖的分层注意力网络。对每个单词与方面词之间的语法路径进行建模，表征每个词对方面词的句法表示，将生成的句法表示反馈到关注层来推断权重。通过分层注意力对单词和句子赋予不同的注意力权重，多方面帮助模型增加对重要部分的注意力。实验结果表明，该方法在SemEval-2014中优于现有的算法。 |
| 关键词: 情感分析；句法依赖；注意力机制 |
|
中图分类号: TP391.1
文献标识码: A
|
|
| Aspect-Level Sentiment Analysis Combining Explicit Syntactic Dependencies and Hierarchical Attention |
|
FAN Mingwei, ZHANG Yunhua
|
(School of Information Science and Technology, Zhejiang Sci-Tech University, Hangzhou 310018, China)
719068852@qq.com; 605498519@qq.com
|
| Abstract: Aiming at the problem of underutilization of explicit syntactic dependencies in aspect-level sentiment analysis, this paper proposes a hierarchical attention network based on grammar dependencies. The syntax path between each word and aspect word is modeled to represent the syntactic representation of each word to the aspect word, and the generated syntactic representation is fed back to the attention layer to infer the weight. Through hierarchical attention, different attention weights are assigned to words and sentences, which helps the model to pay more attention to important parts in many ways. Experimental results show that this method is superior to existing algorithms in SemEval-2014. |
| Keywords: sentiment analysis; syntactic dependency; attention mechanism |
