Recognizing trust in natural language in Amazon's online reviews

Abstract

The goal of this project is to build a system which could extract and classify ethotic statements (ethos relates to trustworthiness, credibility and reliability of seller) about Amazon’s service from a corpus of Amazon’s reviews. Until now processing and extracting ethos was done manually. With this project, we take the first step in automating the process of trust extraction from product reviews. The paper includes discussion on ethos extraction and has used natural language processing, machine learning and python to achieve the goals. Specifically we have used sentiment analysis, argument mining, python, supervised and semisupervised machine learning algorithms such as Naive Bayes and Maxent Classifiers. The contribution of this project is the development of two classifiers. One classifier that classifies sentences into ethos support and ethos attack and the other classifier that extracts ethotic statements from a corpus of Amazon reviews. These classifiers provide an initial solution to automatic ethos extraction.