基于情感词典和Transformer模型的情感分析算法研究

Sentiment analysis method based on sentiment lexicon and Transformer

随着情感分析研究的不断深入,情感词典和深度学习技术被广泛地应用于情感分析任务中。针对情感词典不能考虑词的上下文语义信息,循环神经网络获取整个句子序列信息有限和网络在反向传播时梯度消失或梯度爆炸问题,提出一种基于情感词典和Transformer的文本情感分析方法。该方法不仅可以充分地利用情感词典的特征信息,还能将与情感词相关联的其他词融入到该情感词中,帮助情感词更好地编码。此外,该方法还能够更专注于情感词的不同位置,更好地理解输入句子的单词顺序和表示词与词之间的距离。最后在NLPCC2014情感分析数据集进行实验,取得了比普通卷积神经网络,基于注意力机制的卷积神经网络还要好的分类效果。

With the development of sentiment analysis research,sentiment lexicon and deep learning technologies are widely used in sentiment analysis tasks.Aiming at the problem that the sentiment lexicon cannot consider the context and semantic words,a circular neural network can obtain the limited information of the whole sentence sequence and the gradient disappears or explodes when the network propagates in the reverse direction,a text sentiment analysis method based on the sentiment lexicon and the Transformer is proposed.The method can fully utilize the feature information of the sentiment lexicon,and integrate other words associated with the sentiment word into the sentiment word to help the sentiment word to be better coded.In addition,the method can focus more on the different positions of the sentiment words,better understand the word order of the input sentences and the distance between the words and the words.Experimental results on NLPCC2014 sentiment analysis data set show that the classification effect is better than that of ordinary convolution neural network and convolution neural network based on self-attention.