Customer Review Analysis by Hybrid Unsupervised Learning Applying Weight on Priority Data

Abstract

This paper describes a method which deals with the assurance of E-commerce product quality by analyzing the customer reviews using unsupervised machine learning algorithms. The increasing E-commerce business involves people more engaging in online shopping though the assurance of the product quality is a big concern. For quality products, people now depend on the other customers’ reviews such as comments, emotions, hashtags, etc. The purpose of this study is to automatically extract the polarity of reviews in Bangla as positive or negative sentiment conducted by customers to unleash online product quality. Here, we have tried to implement several unsupervised machine learning techniques for clustering the customers’ reviews such as K-Means, Density-Based Spatial (DBSCAN), Mean Shift, Agglomerative, ensemble approach combined with different clustering, etc. It is observed from the experiments that the proposed method provides 98% accuracy for the experimental corpus by employing PWWA (Priority Word Weight Assignment) in data processing stage and an ensemble clustering algorithm, K-Means and Agglomerative Hierarchical approach, in post-processing stage.