基于深度学习的电商评论信息多刻面情感分类研究  

Deep Learning Based Multi-aspect Sentiment Classification of E-commerce Reviews

作　　者：岑咏华 李文敬[2] 李莉 CEN Yong-hua;LI Wen-jing;LI Li(Management School,Tianjin Normal University,Tianjin 300387,China;School of Economics&Management,Nanjing University of Science&Technology,Nanjing 210094,China)

机构地区：[1]天津师范大学管理学院,天津300387 [2]南京理工大学经济管理学院,江苏南京210094

出　　处：《情报科学》2021年第9期67-73,100,共8页Information Science

基　　金：国家自然科学基金项目“数据驱动的网络广告效应测评与投资决策研究”(71771122)。

摘　　要：【目的/意义】文本情感分类是近年来情报学领域的研究热点之一。已有研究大多关注针对目标文本的单一情感分类。本文旨在探索基于深度学习的电商评论信息多刻面情感分类方法。【方法/过程】提出一种基于Attention-BiGRU-CNN的多刻面情感分类模型,通过BiGRU和CNN获取上下文信息和局部特征,利用Attention机制优化隐层权重,以深度挖掘文本内隐语义和有效刻画多刻面情感。【结果/结论】在中文电商评论信息语料上的实验表明,相较于其他神经网络模型,本文方法可有效提高多刻面情感分类的准确度。【创新/局限】进一步丰富多刻面情感分类的方法途径,为深度挖掘电商评论信息以及优化产品和营销策略提供参考。本文语料主要基于单一类别电商评论信息,聚焦可归纳刻面的情感分类,进一步的研究可面向类别多元化、需通过深度学习提取刻面信息的更大规模语料展开。【Purpose/significance】Sentiment classification/analysis has become an important issue in information sciences,with substantial light shed on one overall sentiment for one textual target.The present research seeks to explore a deep learning based multi-aspect sentiment classification approach.【Method/process】An Attention-BiGRU-CNN model is contrived.The BiGRU and CNN layers aim at effectively capturing contextual information and local features,while the attention mechanism is incorporated to optimize the weights of hidden layers.The model is expected to essentially delve into the deep semantics of text for multi-sentiment knowledge.【Result/conclusion】Experiments on a Chinese e-commerce review corpus evidence the superiority of the proposed method over other neural networks based models,with effectively improved accuracy in multi-aspect sentiment classification.【Innovation/limitation】The research adds to literature of multi-aspect sentiment classification/analysis with methodological and practical implications for e-commerce product improvement and marketing optimization.Further effort could be made to investigate a more scalable multi-aspect sentiment classification approach for large and cross-domain corpus whose aspects is implicitly complex and should be extracted by dedicated deep learning approaches.

关 键 词：多刻面情感分类 电商评论信息 GRU CNN 注意力机制 

分 类 号：F713.36[经济管理—产业经济]