针对电商平台提出一个基于评论的商品特征抽取及情感分析框架,并将该框架在京东生鲜类商品的评论中进行应用.实验结果表明该框架确实能够成功抽取出商品的典型特征及该特征对应的情感极性,且在小样本数据集上测试了特征词和观点词抽取算法以及情感极性计算方法的性能,其中显式＜特征词,观点词＞词对抽取的准确率达到了53.6％,召回率达到了81.5％,极性判断的准确率达到了98.3％.主要贡献包括:提出一种依据观点词与特征词关联度的隐含特征词映射方法;基于word2vec词向量模型计算特征词相似度,并利用改进的半监督层次聚类算法对特征词进行典型特征聚类,建立特征词关联表.%This paper dedicates our work to propose a framework of product feature extraction and sentiment analysis based on comments of e-commerce platform.By applying this framework to the comments of fresh goods in JD.COM,the results of the experiments show that our framework can successfully extract the typical features and their corresponding sentiment polarities of every fresh goods.Also,we evaluated the performance of the feature and opinion extraction algorithms on a small data set;the accuracy and recall of extraction the explicit ＜ feature,opinion ＞ pairs reach 53.6％ and 81.5％ respectively.Furthermore,the accuracy of sentiment analysis reaches 98.3％.Overall,we make the following contributions:we proposed an implicit product feature mapping method,which is based on the correlations of opinions and product features;we proposed an improved semi-supervised hierarchical clustering algorithm to cluster product features and then establish an associative table of similar product features,where we use a new toolkit word2vec,to compute the similarity between any two feature words.
展开▼
