融合深度学习与机器学习的在线评论情感分析  

Online Comment Sentiment Analysis Based on Deep Learning and Machine Learning

作　　者：刘晓彤 田大钢[1] LIU Xiao-tong;TIAN Da-gang(School of Management,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学管理学院,上海200093

出　　处：《软件导刊》2019年第2期1-4,共4页Software Guide

基　　金：沪江基金项目(A14006);上海市一流学科项目(S1201YLXK)

摘　　要：情感分析可以帮助商家了解客户喜好从而生产出满意度更高的商品,也可以监督网上舆论等。为此,基于传统机器学习方法,加入深度学习模块,对在线评论进行情感分析与对比。在词向量训练模块中引入Word2vec模型,用高维向量表示词语、句子,既可防止过度拟合问题,又可减少训练参数个数,提高训练效率。将得到的句向量作为输入代入机器学习模型(MLP、SVM、朴素贝叶斯等)与深度学习模型(CNN、LSTM、BILSTM等),比较实验结果,提出优化方向。结果表明,基于深度学习的情感分析模型准确率明显高于单一机器学习模型,但是深度学习需要大量语料,对实验机器要求也较高,很难完全展现其魅力。Sentiment analysis is very important,it can help merchants understand the preferences of customers so that they can produce more satisfying goods,and it can also supervise online public opinion.This paper is mainly based on the traditional machine learning method to give the results and do the comparison by employing the deep learning module.A total of two modules can be divided.First,the word vector training module introduces the Word2vec model,and uses high-dimensional vectors to represent words and sentences.Here,the pre-trained Word2vec model is introduced,which not only prevents the over-fitting problem,but also reduces the number of training parameters and improves the training efficiency.The second is to enter the obtained sentence vector as the input into the ma chine learning model(MLP,SVM,Na?ve Bayes,etc),deep learning model(CNN,LSTM,BILSTM,etc),compare the experimen tal results,and propose the optimization direction.The accuracy of sentiment analysis models based on deep learning is significantly higher than that of a single machine learning model,but deep learning requires a large amount of corpus,and the requirements for ex perimental machines are relatively high.It is difficult to demonstrate its charm fully.

关 键 词：情感分析 深度学习 机器学习 Word2vec模型 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]