Yelp review dataset sentiment analysis using machine learning techniques

Abstract

Today, internet review sites are becoming a significant criterion for users’ consumption habits on products and services, while being vital source of feedback for businesses. This project aims to present quick feedback on whether consumers are satisfied with businesses’ product and services, lessen the allocation of resources on information extraction towards these reviews, and provide a more agile environment for businesses, by automatizing the extraction of the information “whether the sentiment towards the business service or product is positive or negative” from textual data. The problem, binary classification out of textual data, is addressed through Yelp Company reviews dataset. Yelp is an internet review website, it enables users to review products, services, and businesses. Alongside with the text formatted restaurant reviews, star-rating is converted to 1 (positive) and 0 (negative). These values are obtained to provide the target column to predict the sentiment of the review text. 100,000 restaurant review records are used in 4 different machine learning algorithms to predict the binary classification problem of predicting whether the review sentiment is positive or negative. 2 neural networks one with pre-trained GloVe, SVM, and Logistic Regression models are used, and the success of these models is compared using F1-Score as a performance metric. These results are presented in the paper.