基于Senti-PMU模型的文本情感分析  

Text Sentiment Analysis Based on Senti-PMU Model

作　　者：余亮 蒋玉明[1] YU Liang;JIANG Yu-ming(College of Computer Science,Sichuan University,Chengdu 610065)

机构地区：[1]四川大学计算机学院,成都610065

出　　处：《现代计算机》2020年第29期19-24,共6页Modern Computer

摘　　要：随着互联网技术的高速发展,互联网通讯产生的信息量也指数级地增加。这些庞大的数据资源库给情感分析提供了取之不竭的语料信息。如何利用这些文本数据训练出优秀的自然语言处理的网络模型是NLP的热点研究方向之一。提出一种由输入文本情感值调控的神经网络模型Senti-PMU。Senti-PMU是基于节制记忆单元(PMU)的神经网络,PMU从结果精度、模型复杂度、训练时长都要优于传统循环神经网络模型。Senti-PMU保留PMU模型的原有结构,加入潜在语义增强模型(MLSM)思路。通过对比情感词典SentiWordNet与输入文本,得到每个命名实体的情感值,根据情感值动态地调节输入文本与记忆单元隐藏信息的权重:情感值越大,则表明该命名实体对于整体文本情感极性的贡献越大,记忆单元需要对这部分文本保留更多的信息量;情感值越小,记忆单元则对该命名实体保留更少的信息量。实验表明,Senti-PMU在同种数据集上,结果精度要优于传统循环神经网络模型。With the rapid development of Internet technology,the amount of information generated by Internet communication has also increased expo⁃nentially.These huge data resources provide the inexhaustible corpus information for emotion analysis.How to use these text data to train excellent NLP network model is one of the hot research directions of NLP.In this paper,a neural network model Senti-PMU is proposed,which is controlled by the emotional value of input text.Senti-PMU is a neural network based on parsimonious memory unit(PMU).PMU is superior to the traditional cyclic neural network model in terms of result accuracy,model complexity and training time.Senti-PMU keeps the original structure of PMU model and adds the idea of memory enhanced late semantic model(MLSM).By comparing SentiWordNet with the input text,we can get the emotional value of each named entity,and dynamically adjust the weight of the information hidden in the in⁃put text and memory unit according to the emotional value:the larger the emotional value is,the greater the contribution of the named entity to the emotional polarity of the whole text is,and the memory unit needs to retain more information for this part of the text;the smaller the emotional value is,the memory unit needs to keep more information for this part of the text Memory cells retain less information for the named entity.Experimental results show that the accuracy of Senti-PMU is better than that of traditional cyclic neural network model on the same data set.

关 键 词：文本情感分析 PMU Senti-PMU MLSM SentiWordNet 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]