基于卷积神经网络与情感倾向点互信息算法的农产品情感词典构建

Agricultural products sentiment dictionary construction based on convolutional neural network and semantic orientation pointwise mutual information algorithm

针对现有的情感词典无法精准地对在线农产品情感词进行捕捉分析的问题,通过对农产品在线评论进行分析,构建专用情感词典,创新地提出C-TF算法。首先,将卷积神经网络(CNN)与词频(TF)结合计算得出情感种子词,并利用文本词性标注进行词语过滤得到候选词;接着,利用情感倾向点互信息(SO-PMI)算法计算每个候选词与种子情感词的相似度从而对词语极性标注;最终,形成农产品情感词典。为验证所提词典的准确性,用构建的情感词典对不同平台的农产品评论进行情感分类。实验结果显示,与其他情感词典相比,在精确率、召回率和F1值3个评价指标上均提升了5.00以上个百分点。所构建的农产品情感词典对农产品情感分类效果更好,便于商家更准确地掌握消费者的情感倾向。

Aiming at the problem that the existing sentiment dictionary cannot accurately capture and analyze the sentiment words of online agricultural products,by analyzing the online comments of agricultural products,a special sentiment dictionary was constructed,and the Convolutional neural network-Term Frequency(C-TF)algorithm was innovatively proposed.First,Convolutional Neural Network(CNN)and Term Frequency(TF)were combined to calculate emotional seed words,and used text part-of-speech tagging to filter words to obtain candidate words.Then,the similarity between each candidate word and the seed sentiment word was calculated by the Semantic Orientation Pointwise Mutual Information(SO-PMI)algorithm to mark the polarity of the word.Finally,the sentiment dictionary of agricultural products was formed.In order to verify the accuracy of the proposed dictionary,the constructed sentiment dictionary was used to classify the sentiment of agricultural product reviews on different platforms.The experiment results show that compared with other sentiment dictionaries,the three evaluation indicators(Precision,Recall,and F1 score)are improved by more than 5.00 percentage points.The construction of agricultural products sentiment dictionary is more effective for agricultural product sentiment classification,and it is convenient for merchants to more accurately grasp the sentiment tendencies of consumers.