Abstract:

A sentiment classification that is specific to one aspect is a fine-grained task in the field of sentiment analysis. Deep neural networks can better extract context features and aspect features, and use the attention mechanism to assign corresponding weight values according to the different importance of context features and aspect features. The model focuses on extracting context and aspect features and better integrating context and aspect vectors, and proposes a deep neural network with mixed extraction and multi-layer attention. Based on Bi-LSTM and CNN, there are significant results in extracting features, and a merge model of two networks is introduced. Finally, it is verified on the classic Laptop, Resteraunt and Twitter datasets, showing a better classification effect than the benchmark model.

Key words: aspect-level, sentiment classification, multi-layer attention

摘要：

特定于某一方面的情感分类是情感分析领域中的一项细粒度任务。深层的神经网络可以更好地提取上下文特征与方面特征，同时利用Attention机制可以根据上下文特征和方面特征不同的重要性赋予相应的权重值。模型着重从提取上下文与方面特征和更好地融合上下文与方面向量入手，提出了一种混合提取与多层注意的深度神经网络。基于Bi-LSTM和CNN在提取特征方面都有显著的成效，引入两种网络的合并模型。最后，在经典的Laptop，Resteraunt和Twitter数据集上进行了验证，展示了比基准模型更好地分类效果。

关键词: 方面级, 情感分类, 多层注意