LAUSR

: Customer reviews has a significant role in sale of the product on the e-commerce website and this influences other customers. The fake review affects the trust between the user and seller in the e-commerce website. The existing techniques in detection of fake review have disadvantage of overfitting problem and vanishing gradient problem. The Semantic Feature selection – Bidirectional Long Short Term Memory (SF-BiLSTM) is applied to increase the fake review detection efficiency. The autoencoder based semantic feature selection technique maps the features based on decision tree technique to select the relevant features. This technique helps to learn the features importance related to neighbourhood features that helps to increases the efficiency. The proposed SF-BiLSTM method has the advantage of applying the emotion recognition method in the fake review detection. The four datasets such as Amazon Review, Yelp, Restaurant, and Hotel were used to test the performance of classification. The semantic feature selection technique reduces overfitting problem in existing Convolution Neural Network classification and the mapping of features helps to increase model learning performance. The proposed SF-BiLSTM model has 99.2 % accuracy when compared to the existing methods in fake review detection.