A framework for non parametric naïve bayes classification using opinion mining

Abstract

Opinion mining has gained much attention with the rapid growth of newlinesocial media. Polarity classification (Fersini et al. 2014) is a task of opinion newlinemining in which decisions are taken based on customer reviews and survey newlineresponses. The task of polarity classification is to classify the text document into newlinepositive or negative (Bijal et al. 2015 and Janardhana et al. 2015). This is newlineachieved by implementing the machine learning methods of classification. newlineGenerally, the opinion dataset contains a large number of features that operates newlineon a higher dimension. If all those features are considered, then it leads to poor newlineaccuracy of the classifier. Therefore, the dimension of the data must be reduced newlinebefore building the classifier model, which is carried out by transforming the newlinehigher dimension data into lower dimension by considering only the intrinsic newlineinformation of the data. The reduction of dimension can improve the robustness newlineof the classifier and reduces the time, and computational complexity. To classify newlinesuch a large volume of opinion dataset, the supervised machine learning newlinetechnique the Naive Bayes - Kernel Density Estimation (NB-KDE) is proposed. newlineThe classification of opinions can be divided into four stages. The newlinefirst stage involves pre-processing of the opinions. Generally, the text documents newlinecontain a large volume of data in which most of the words are irrelevant to the newlinecontent. Therefore, pre-processing is needed while classifying the text newlinedocuments. To handle the pre-process task efficiently, this system proposes the newlineStringToWordVector filter that has a number of parameters like stemming, newlinestopword removal and tokenizer. Tokenization (Muhammad et al. 2016) is the newlineact of dividing the strings into a number of tokens like words, symbols, and newlinephrases. It is the conventional method of text analysis to generate a basic unit of newlinewords. In WEKA, the WordTokenzier is used as a simple Tokenizer, which newlinegives the output as tokens. Stopwords are filtered out before classifying the text. newlineThe stopwords (Janardhana et al. 2015) refer to the most common words that do newlinenot have any meaningful information. The Rainbow technique is used to remove newlinethe stopwords in the proposed system newline