基于BiLSTM-CNN-ATT的方面级情感分析

Aspect-level Sentiment Analysis Based on BiLSTM-CNN-ATT

方面级情感分析旨在挖掘句子中关于特定方面的情感极性。根据特定方面是否存在于句子中,方面级情感分析细分为ACSA和ATSA两个子任务。过去的研究大多采用基于注意力机制的循环神经网络模型和卷积神经网络模型,但前者不能有效捕获局部情感特征,后者对全局语义信息挖掘不够充分。针对这些问题,提出BiLSTM-CNN-ATT模型。该模型通过双向长短期记忆网络捕获上下文和特定方面语义信息,之后利用注意力机制和自注意力机制对特定方面和上下文进行优化,最后融合优化后的上下文和特定方面,利用卷积神经网络提取局部情感特征并进行情感预测。为验证该模型的有效性,在SemEval2014任务4的Restaurant和Laptop数据集上进行测试。实验结果表明,在ACSA和ATSA的三分类和二分类任务中,该模型准确率均有提高。

Aspect-level sentiment analysis aims at exploring the emotional polarity of specific aspect in sentence.According to whether specific aspect exists in sentence,aspect-level sentiment analysis includes two sub-tasks:aspect-category sentiment analysis(ACSA)and aspect-term sentiment analysis(ATSA).In the past studies,most of them used the attention-based recurrent neural network model and the convolutional neural network model,the former cannot effectively capture local emotional features,while the latter cannot fully mine global semantic information.To solve these problems,the attention-based bidirectional long short term memory and convolutional neural network model(BiLSTM-CNN-ATT)is proposed.In the model,using the bidirectional long short memory network captures the semantic information of context and specific aspect,then using attention mechanism and self-attention mechanism optimizes specific aspect and context,finally using convolutional neural network extracts local emotional features from optimized context and specific aspect,predicting emotional polarity.To verify the validity of the model,test the model on the laptop and restaurant datasets in task 4 of Semeval2014.The experimental results show that the accuracy of the proposed model is improved in both three-way classification and binary classification on ACSA and ATSA.