An Approach to Overcome Flaming Problem of Social Networking Sites Using Various Classifiers

Author Name(s): Dr. Avinash J. Agrawal, A.V. Zadgaonkar
Author Email: agrawalaj@rknec.edu

Abstract

Natural language Processing (NLP) is a field of artificial intelligence and computational linguistics concerned with interaction of humans with machines. Social Networking sites (SNS) is a widely used communication tool today. Sometimes it gives rise to the problem of flaming. A flaming incident is triggered by comments and action made by user in SNS that ends up damaging reputation or causing negative impact on the target party. The paper proposes an architecture to overcome this problem. The training dataset used contains a total of 17282 tweets already annotated as positive, negative and neutral. The test dataset that has been used contains a total of 4382 tweets already annotated as positive, negative and neutral. Result of machine learning techniques used in the proposed architecture are reported and discussed in this paper.

Keywords Natural Language Processing, Flaming, Naïve Bayes Classifier, Maximum Entropy Classifier, Feature Selection

Introduction

Social networking sites have achieved huge popularity in the last decade by making the transmission of information [8][9] across the globe very easy for everyone. Flaming [1] is one of the current hazards of the online communication. The flaming is caused due to negative comments made by user that can have a huge negative impact on the target. So, in order to overcome this problem, sentiment analysis techniques can be used. In this paper, an architecture has been proposed that describes various techniques for finding polarity (positive, negative and neutral) of the input document. A training set of 17282 tweets already annotated (positive, negative and neutral) was preprocessed [4] by applying tokenization, case folding to lower case, stop word removal and stemming to construct a sentiment dictionary. This dictionary contains the words and positive, negative and neutral frequencies. The dictionary is context dependent as it depends on the training dataset. This is the problem of text classification [1]. Then this dictionary is used for assigning polarity of the input document. The proposed approach describes various method like Naïve Bayes Classifier[1][2], Maximum Entropy [3] and Feature Selection based on Entropy[1] method, that can use the constructed dictionary for sentiment analysis of input. Sentiment Analysis can help us to know the polarity of a given document. First step is to apply pre-processing to the labeled dataset and construct sentiment polarity dictionary for the target company. Second step is to pre- process input document (comment) and then classify its sentiment using constructed dictionary, Classifiers and Feature Selection based on Entropy method.

 

Conclusion

Classification using the Naive Bayes Classifier has minimum accuracy. Its accuracy improves by using the feature selection method. The classification accuracy of Maximum Entropy Classifier is better than the Naive Bayes Classifier but still less than the Naive Bayes Classifier with Feature Selection based on Entropy Method whose accuracy is further less than the Maximum Entropy Classifier with Feature Selection based on Entropy Method. So it is found out that maximum Entropy Classifier with Feature Selection based on Entropy Method is best for classification of a given input tweet and hence is best for the detection of flames.

413 total views, 1 views today

Download PDF File

About the author: admin