融合句法规则和CNN的旅游评论情感分析

Sentiment analysis of tourist reviews combined with syntactic rules and CNN

旅游评论文本冗长且句式复杂,现有的旅游类情感分析算法较少考虑文本特点和句法变化规则,导致分类准确率降低。针对该问题,提出融合句法规则和卷积神经网络(convolutional neural network,CNN)的旅游评论情感分析算法SCNN(syntactic rules for convolutional neural network)。建立4个词典,根据词典对文本进行直接分类词、总结句和转折句的初步提取,通过CNN进行分类。实验结果表明,与传统的CNN算法相比,SCNN算法在准确率、召回率和F1值上分别平均提升了3.9%、4.2%和4.6%,AUC值提升了13%。

The tourist reviews are long and the sentence pattern is complex.The existing algorithms of sentiment analysis seldom consider the text characteristics and the syntactic change rules,which leads to the decrease of classification accuracy.A tourist reviews sentiment analysis algorithm SCNN(syntactic rules for convolutional neural network)was proposed,which combined syntactic rules and convolutional neural network(CNN).Four dictionaries were established,the direct classified words,summary sentences and transitional sentences were extracted from the text according to the dictionary,and they were classified through CNN.Experimental results show that compared with the traditional CNN algorithm,the average accuracy,recall rate and F1 value of SCNN are improved by 3.9%,4.2%and 4.6%,respectively,and the AUC value is improved by 13%.