Identification of significant features and machine learning technique in predicting helpful reviews

Abstract

Consumers nowadays rely heavily on online reviews in making their purchase decisions. However, they are often overwhelmed by the mass amount of product reviews that are being generated on online platforms. Therefore, it is deemed essential to determine the helpful reviews, as it will significantly reduce the number of reviews that each consumer has to ponder. A review is identified as a helpful review if it has significant information that helps the reader in making a purchase decision. Many reviews posted online are lacking a sufficient amount of information used in the decision-making process. Past research has neglected much useful information that can be utilized in predicting helpful reviews. This research identifies significant information which is represented as features categorized as linguistic, metadata, readability, subjectivity, and polarity that have contributed to predicting helpful online reviews. Five machine learning models were compared on two Amazon open datasets, each consisting of 9,882,619 and 65,222 user reviews. The significant features used in the Random Forest technique managed to outperform other techniques used by previous researchers with an accuracy of 89.36%.