基于CNN-BiLSTM-HAN混合神经网络的高校图书馆社交网络平台细粒度情感分析  

Fine-grained Sentiment Analysis of Social Network Platform of University Libraries Based on CNN-BiLSTM-HAN Hybrid Neural Network

作　　者：李博[1] 李洪莲[1] 关青 刘杨[1] LI Bo;LI Honglian;GUAN Qing;LIU Yang(Library of Harbin University of Commerce,Harbin 150028)

机构地区：[1]哈尔滨商业大学图书馆,哈尔滨150028

出　　处：《农业图书情报学报》2022年第4期63-73,共11页Journal of Library and Information Science in Agriculture

基　　金：国家社会科学基金青年项目“中美公共图书馆法人治理结构比较研究”(19CTQ005)。

摘　　要：[目的/意义]从高校图书馆社交网络平台用户评论数据挖掘角度出发,对用户评论情感极性进行细粒度分析,为高校图书馆了解用户真实情感倾向并提升服务质量提供科学依据。[方法/过程]以国内高校图书馆社交网络平台用户中文评论数据为研究对象,通过Tensor Flow深度学习框架,利用Keras人工神经网络库,将卷积神经网络(Convolution Neural Network,CNN)和双向长短时记忆网络(Bidirectional Long Short Term Memory,BiLSTM)结合,并引入层次化注意力机制(Hierarchical Attention,HAN),构建基于CNN-BiLSTM-HAN混合神经网络的情感分析模型。[结果/结论]利用真实高校图书馆社交网络平台用户评论数据集进行实验,本文方法取得93.38%的准确率,结果表明本文模型的有效性。模型较为复杂,导致模型训练时间上较长,方法模型的普适性有待进一步检验,表情符号信息没有得到有效利用,参数设置尚需进一步研究。[Purpose/Significance] From the perspective of data mining of user comments on a social network platform of a university library, the sentiment polarity of user comments is analyzed in a fine-grained way. It provides scientific basis for a university library to understand the real sentiment tendency of its users and improve its service quality. [Method/Process] This paper takes the Chinese comments data of social network platform users of domestic university libraries as the research object. Through the TensorFlow deep learning framework, we used Keras artificial neural network library, combined convolution neural network and bidirectional long short term memory network, introduced hierarchical attention mechanism, and constructed sentiment analysis model based on CNN-Bi LSTM-HAN hybrid neural network. [Results/Conclusions] The experiment is carried on by using the data set of user comments on the real social network platform of a university library. The accuracy of this method is 93.38%, and the results show that the model is effective. The model is more complex, as a result, the training time of the model is longer, the universality of the method model needs to be further tested., Emoticons are not used effectively, and the parameter setting needs further study.

关 键 词：高校图书馆 社交网络平台 卷积神经网络 双向长短时记忆网络 层次化注意力机制 情感分析 

分 类 号：G203[文化科学—传播学]