基于改进型LSTM的文本情感分析模型研究  

Research on Text Sentiment Analysis Model Based on Improved LSTM

作　　者：罗正军[1] 柯铭菘 周德群[1] LUO Zheng-jun;KE Ming-song;ZHOU De-qun(Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China)

机构地区：[1]南京航空航天大学,江苏南京210016

出　　处：《计算机技术与发展》2020年第12期40-44,共5页Computer Technology and Development

基　　金：国家自科项目(71573121);中央高校基本科研业务费专项资金资助(NR2016016,NJ20140028)。

摘　　要：文本情感分析是自然语言处理领域的一大研究方向。文本情感分析本质上属于文本二分类问题,问题的核心是将一段文本所表达的情感分为正向和负向两类。传统的文本分类算法在进行文本情感分析时,不能很好地考虑到词与词之间的关联性以及词语之间的极性转移。针对LSTM神经网络模型在文本情感分析中的不足,设计并提出了基于改进型LSTM的文本情感分析模型。为了降低在原始LSTM模型中采用随机梯度下降法进行参数更新所带来的不确定性,提出一种基于向量空间的伪梯度下降法。在迭代过程中,为了减轻模型准确率的振荡现象,提出带有修正项的二元交叉熵损失函数,使改进后的模型有选择性地针对分类模糊的数据进行更新。实验结果表明,改进后的模型在分类正确率以及迭代效率上有所改进。Text sentiment analysis is a major research direction in the field of natural language processing.Text sentiment analysis is essentially a text binary classification problem.The core of the problem is to divide the sentiment expressed by a text into two categories:positive and negative.The traditional text classification algorithm cannot well take into account the association between words and the polarity transfer between words when performing text sentiment analysis.Aiming at the shortcomings of LSTM neural network model in text sentiment analysis,we design and propose a text sentiment analysis model based on improved LSTM.In order to reduce the uncertainty caused by the stochastic gradient descent method for parameter updating in the original LSTM model,a pseudo gradient descent method based on vector space is proposed.During the iterative process,in order to reduce the oscillation of the accuracy of the model,a binary cross-entropy loss function with a correction term is proposed,so that the improved model could be selectively updated for the fuzzy data.The experiment shows that the improved model has improved classification accuracy and iteration efficiency.

关 键 词：文本情感分析 机器学习 长短期记忆模型 梯度下降 损失函数 

分 类 号：TP315[自动化与计算机技术—计算机软件与理论]