基于局部注意力的方面级别情感分析算法

A local attention network for aspect-level sentiment analysis

考虑到对方面进行情感分析时,距方面词越近的上下文内容对方面的情感倾向影响可能越大,因此将结合距离因素,提出了基于局部自注意力机制的算法用于方面级别的情感分类。将输入分为全局文本序列、方面序列以及局部文本序列,通过自注意力分别对三个序列进行特征提取,对全局文本和方面序列进行交互注意力计算,更新全局文本表示,对局部文本和方面序列进行交互注意力计算,更新局部文本表示。最终将全局文本和局部文本表示连接起来输入分类器。实验结果显示本文的结果优于近两年提出的方面级别情感分析算法。

Considering that the context content closer to the aspect words may have a greater influence on the sentiment of the aspect when analyzing the emotion of the aspect,an algorithm based on the local attention mechanism is proposed to classify the emotion of the aspect level based on the distance factor.Input can be divided into global text,aspect,and the local text.Extract feature of the three sequences by self-attention,and then apply interactive attention on global text and aspect to update global text representation,apply interactive attention on local context and aspect to update the local text representation.Finally,the global text and the local text representation are joined into the input classifier.Experimental results show that our results are superior to the aspect-level sentiment analysis algorithm proposed in the past two years.