Mapping Distinct Source and Target Domains on Amazon Product Customer Critiques with Cross Domain Sentiment Analysis

Sentiment Analysis is the computational investigation of individuals’ conduct, inclinations, judgment, and assessments about people, issues, elements, points, occasions, items as well as their characteristics. In addition, the Internet has become the most significant spot for communicating opinions about items and administrations, just as for remarking on social issues and executive blueprint. The analysis of this kind of information is exceptionally valuable for an entire scope of useful applications, yet it is challenging as well. And the indispensable obstacle in sentiment analysis is that it is highly domain-centralized. Hence, a model that performs satisfactorily in one domain might not perform in another. This work aims to use cross-domain sentiment classification using Machine Learning on an Amazon product dataset to try to overcome these challenges and build on and attempt to improve the previous work carried out. The dataset is first preprocessed to clean the noisy data, fill the missing values, etc, and then normalized according to the requirements. It then uses feature extraction to develop the labeled feature vectors from the source domain and train a model which allows it to pick important features in a comprehensive manner. Next, it uses the trained classifier to classify reviews in the target domain. The proposed method is expected to be a significant improvement and aims to generalize the underlying method to solve other types of domain-dependent tasks in the future.