基于深度学习的商品评价情感分析与研究

Commodity Evaluation Analysis and Research Based on Deep Learning

移动互联网助推的电子商务时代使得商品评价空前繁荣,论文提出一种基于深度学习的改进型模型来分析评价数据的情感。首先通过分词与综合停用词表等预处理数据集,然后使用Skip-gram模型训练出数据集中每个词的词向量,并使用自扩充情感词典对评价语句情感极性进行量化,量化的情感正负值与词向量形成融合矩阵输入,并通过分流规则设计进行差异网络输入,选择CNN或RNN完成抽象特征提取,即Shunt-C&RNN产品评价分类模型(改进型深度学习方法)。与传统机器学习SVM相比,改进型深度学习方法准确率大幅提升6.6%,较单一深度学习方法提高了近1.5%。

This paper proposes an improved deep learning model for commodity evaluation sentiment analysis. Firstly,this paper uses stop words and tokenizer to pretreatment the data,then Skip-gram model is used to generate word vectors. Secondly,an autogenerated sentiment lexicon is used to quantify the sentiment polarity of words in commodity reviews and integrate this information into the model input matrix. Lastly,this paper counts the differences between the network input through the distribution rules of designed and chose RNN or CNN for feature extraction. Above all is the Shunt-CRNN commodity reviews sentiment classification model（improved deep learning approach）. Compared with the traditional machine learning SVM and the single deep learning method the proposed method has improved the precision by 6.6% and 1.5% respectively.