Publication

Abstract : Opinions for any product, topic or organization are provided by users on Social networking and E-commerce sites. These opinions have high influence on other users’ purchasing decisions. Spammers post fake opinions so that their organization’s product can be promoted and competitor’s product can be demoted. There is need for detecting spam opinions so that users can have fair information about any product. Review text and reviewer behavior are factors considered to detect spam opinions. Traditional machine learning techniques such as SVM, Logistic Regression and Naive Bayes are applied to distinguish spam opinions from original reviews, but accuracy is not adequate. Furthermore, traditional approaches use only bag-of-words representation which do not consider context of words. In this research work, reviews text are observed and fed into Convolutional Neural Network (CNN) model. Convolutional layers, number of hidden layers, dense layers and filter size are set to build model. Further, essential text features are included in CNN model to improve accuracy. GloVe word embedding is used in proposed model. Experiment analysis on Ott dataset proves that accuracy is improved significantly by proposed model as compared to traditional machine learning approaches.