基于改进KNN的消费者评价信息情感分类研究  

Research on classifying sentiment of consumer evaluation based on modified KNN algorithm

作　　者：刘晓菲[1] 丁香乾[1] 石硕[1] 李林春 李忠态 

机构地区：[1]中国海洋大学信息科学与工程学院,山东青岛266100 [2]红塔烟草(集团)有限责任公司信息网络科,云南玉溪653100

出　　处：《微型机与应用》2014年第24期81-83,86,共4页Microcomputer & Its Applications

基　　金：国家科技支撑计划(2012BAF12B06);青岛市科技计划项目课题(12-4-1-9-gx)

摘　　要：面对大数据时代消费者评价的海量信息,为了识别消费者评价信息的情感倾向,及时掌握消费者的评价信息反馈,采用K-近邻(KNN)算法对消费者评价信息进行情感分类,但是该算法在文本分类过程中因文本特征向量的维度高,使得算法的时间复杂度和空间复杂度较高,计算的开销很大。针对这一问题,通过对获取信息的文本结构以及情感表达特点的分析,采用一种改进的KNN算法进行文本情感分类。在对消费者评价信息进行分类时,先由潜在语义分析算法对文本特征向量进行降维处理,然后利用加权KNN算法进行分类。实验结果表明,该方法在提高文本分类速度的同时保持了良好的分类效果。Facing the huge mountains of consumers evaluation in the age of big data, in order to identify consumers sentiment orientation of assessment information and get consumers evaluation information feedback in time, K-nearest neighbor（KNN）algorithm can be used to solve the problem of emotion classification of consumer evaluation information. But in the process of classification, feature vector dimension is so high, leading to high time complexity and space complexity of the algorithm and massive calculation. To solve the problem, by analyzing the structure of text information and the characteristics of the sentiment expression, a method based on modified KNN algorithm is proposed in this paper. When classifying consumers assessment information, firstly, it takes dimensionality reduction with Latent Semantic Analysis algorithm, then classifies with KNN algorithm. Experiment results show that tile method can effectively improve speed of sentiment analysis and keep a good classification effect.

关 键 词：大数据 特征降维 LSA算法 KNN算法 情感分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]