Classifying the Small Text into Multiple Sentiment Labels Using Enhaned Features and Machine Learning Technique

Abstract

Social media and e-commerce is become one of the most popular applications. Almost every person has an account on social media and e-commerce. In both platforms, a small size of text is used for expressing opinions by users on a topic or a product. User opinion is one of the sources to understand how a user interacts with a social media post or a product in the e-commerce platform. In this paper, the e-commerce product reviews have been taken into consideration. Product reviews are also playing a vital role in the orientation of the buyer’s decision toward product purchasing. In this condition, fake or spam reviews may negatively impact the buyer’s decision Thus spam reviews can be a reason for the loss of credibility of the e-commerce platform and also impact buyers. Thus the main aim of the presented work is to identify spam reviews using Machine Learning (ML) technique. The proposed model includes an overview of a modified Term-frequency and inverse document frequency (TF-IDF) to select features from the review text. Then, k-means clustering is used to assign initial class labels to the review text. Finally, for assigning the final classes to the review text a neural network has been used. The experimentation has been conducted on the Amazon product review dataset, which is used for spam review classification tasks. The results were measured and compared, which shows improvement in prediction accuracy as compared to similar spam classification models.