Enhanced Classifier Model for Detecting Spam Reviews Using Advanced Machine Learning Techniques

Abstract

Now a days lot of people, uses e-commerce websites for purchasing the products. Before purchasing any product once, the reviews will be checked by the user. In this aspect, mainly we are focusing on the fake review detection using classification of those reviews by the polarity score index. Traditional Naive Bayes and its variants (Multinomial, Bernoulli model) are used in estimating the initial weights of the features in classifying the sentiments. The posterior knowledge of NB algorithm is used in finalizing the initial value of each term present in the reviews. Later, Gradient ascent technique is used to estimate the exact weights of the attributes towards improving the likelihood of the classifier. The use of Gradient ascent is finalizing the exact weight of each term towards improving the likelihood of text classifier. The proposed work has its advantages in estimating the intensity in sentiment present in review based on context available to determine fake or not using polarity score. The design of novel supervised text classifier mainly deals with extracting impactful features from text based on context. The outcome from the present work is made in use with context-based sentiment analysis. It avoids the use of dictionaries and updating the dictionary based on context and it save lots of time in processing hung number of reviews.