Sentiment analysis for various SNS media using Naïve Bayes classifier and its application to flaming detection

Abstract

SNS is one of the most effective communication tools and it has brought about drastic changes in our lives. Recently, however, a phenomenon called flaming or backlash becomes an imminent problem to private companies. A flaming incident is usually triggered by thoughtless comments/actions on SNS, and it sometimes ends up damaging to the company's reputation seriously. In this paper, in order to prevent such unexpected damage to the company's reputation, we propose a new approach to sentiment analysis using a Naïve Bayes classifier, in which the features of tweets/comments are selected based on entropy-based criteria and an empirical rule to capture negative expressions. In addition, we propose a semi-supervised learning approach to relabeling noisy training data, which come from various SNS media such as Twitter, Facebook, blogs and a Japanese textboard called '2-channel'. In the experiments, we use four data sets of users' comments, which were posted to different SNS media of private companies. The experimental results show that the proposed Naïve Bayes classifier model has good performance for different SNS media, and a semisupervised learning effectively works for the data consisting of long comments. In addition, the proposed method is applied to detect flaming incidents, and we show that it is successfully detected.

Data properties and the performance of sentiment classification for electronic commerce applications

“…In the previous research, Naïve Bayesian (Kang et al 2012;Yoshida et al 2014), support vector machine (Mullen and Collier 2004) and decision tree (Sui et al 2003) have been used for implementation of sentiment classification and these algorithms are reported to be effective for sentiment classification in the literature (Forman 2003;Dhillon et al 2003;Sebastiani 2005;Wan et al 2012;Ur-Rahman and Harding 2012). In this study, we also use these three algorithm for the experiment of impact of data properties on sentiment classification performance.…”

Proposed Techniques to Remove Flaming Problems from Social Networking Sites and outcome of Na

“…It has made the transmission of information across the globe very easy for everyone. But it has also brought with it a problem called flaming [1] which are one of the current hazards of online communication. They are caused due to negative comments made by users that can have a huge impact on the target in a negative way.…”

“…In this paper an architecture has been proposed that describes various techniques for the purpose of finding polarity (negativity, positivity or neutrality) of the input document. A training set of 17282 tweets already annotated(as positive, negative or neutral) was preprocessed [4] by applying tokenization, stop words removal and stemming to construct a sentiment dictionary [1]. This dictionary contains the word and its positive, negative and neutral frequencies.…”

“…The dictionary is context dependent as it depends on the training dataset. This is a problem of text classification [1]. Now this dictionary is used for the assigning polarity to the input document.…”

“…Now this dictionary is used for the assigning polarity to the input document. The proposed approach describes various techniques, like Naïve Bayes Classifier [1][2], Maximum Entropy Classifier [3] and Feature Selection based on Entropy [1] method, that can use the constructed dictionary for sentiment analysis of input. This paper also describes in detail, Naïve Bayes Classifier which was trained and tested.…”