Multiclass Classifier Building with Amazon Data to Classify Customer Reviews into Product Categories
– E-commerce refers to the Electronic Commerce and defined as buying and selling of products over electronic systems such as the Internet. With the widespread use of the Internet, the trade conducted electronically (online) has grown extraordinarily. The E-commerce companies have a large database of products and a number of consumers that use these data. To address this data and information explosion, e-commerce stores are applying machine learning to identify and customize the product category information. Data scientists in this field are utilizing machine learning potential to build unmatched competitiveness in the market by finding purchase preferences, customer churn and product suggestions etc. Applying popular Machine Learning algorithms to huge datasets brought new challenges for the ML practitioners as traditional ML libraries do not support well processing of large datasets. So to address the issue, data and computation can be distributed to any Cloud Computing environment with minimal effort. Cloud computing paradigm turned out to be valuable alternatives to speed-up machine learning platforms. The work first discusses the machine learning and its importance in predictive analytics. Introduction to multiclass classification is presented. Few E-commerce classification frameworks and need of cloud platforms to analyze ever-growing E-commerce data are briefly surveyed then. Finally, this work proposes the predictive framework for E-commerce Product Classification which is developed over the Microsoft Azure Cloud. The proposed framework predicts the Product Category in a large E-commerce dataset released by a famous e-commerce company for a competition. The proposed classifier is build using ‘Multiclass Logistic Regression’ by choosing the optimal parameters. The results obtained by the proposed model are evaluated and presented in terms of accuracy. The work also demonstrates the use of leading cloud environment for machine learning. The results obtained in this research are promising and the dissertation also directs the future research work in the field.
Research Paper Link: Download Paper