基于特征级注意力的方面级情感分类模型研究

Research on aspect level sentiment classification model based on feature level attention

近年来大数据、自然语言处理等技术得到了飞速发展。情感分析作为自然语言处理细分领域的前沿技术之一,得到了极大的重视。然而,低参数量、高精度依然是制约情感分析的关键因素之一。为实现模型参数少、模型分类精度高的情感分析需求,通过改进特征级注意力机制的输入向量,以及前馈神经网络与注意力编码的前后位置关系,得到可复位特征级注意力机制,并基于该机制提出了基于可复位特征级注意力方面级情感分类模型(RFWA)和基于可复位特征级自注意力方面级情感分类模型(RFWSA),实现了高精度的方面级情感分析效果。在公开数据集上的实验结果表明,相比现有的主流情感分析方法,所提出的模型有明显的优势,尤其是在取得相当分类效果的情况下,模型的参数量仅为最新AOA网络的1/4。

In recent years,big data,natural language processing and other technologies have been developed rapidly.As one of the cutting-edge technologies in the field of natural language processing,emotion analysis has received great attention.However,High precision and high performance are still the key factors restricting emotional analysis.In order to achieve high-precision emotion analysis,based on the feature-level neural network,this paper improves the reset feature level attention mechanism,and proposes an aspect level emotion classification model based on the reset feature level attention(RFWA)and an aspect level emotion classification model based on the reset feature level self-attention(RFWSA).Finally,combined with Bi-LSTM-CRF,high quality aspect level emotion analysis is realized by aspect level phrase extraction in the network.The experimental results show that compared with the existing mainstream emotion analysis model,the model proposed in this paper has obvious advantages.Especially when the classification effect is quite good,the parameters of the model are only 1/4 of the AOA Network.