Chi-Squared Based Text Convolutional Neural Network for Fake Consumer Review Classification

The spam reviews misguide the decision makings of consumers and affected fair trading in online markets. The previous researches in fake review detection majorly concentrated on classification, but has less accuracy because of irrelevant features are used in classification. In this research, Chi-squared based Text Convolutional Neural Network (CSM based TCNN) is proposed for fake review classification. The Yelp dataset is used for fake review classification and it is pre-processed by tokenization and stemming. The POS features and Bag of Words are used for feature extraction phase which extract relevant feature and the relevant features are selected by using chi-squared technique that selected the required features for classification. Then, the classification is performed by Text- Convolutional Neural Network (TCNN) method with high accuracy. The evaluation of proposed method is performed with different metrics of accuracy, precision, recall and f1-score. The proposed algorithm has achieved 98.45% accuracy, 95.62% precision, 96.17% recall and 95.88% f1-score that is better than other existing algorithms like Convolution Neural Network and Long Short-Term Memory (CNN-LSTM).