Aspect-Based Sentiment Analysis of Amazon Reviews for Fitness Tracking Devices

Abstract

The year 2012 marked the birth of a new class of wireless wearable fitness trackers (e.g., Fitbit One) that track daily activity from the count of steps taken and calories burned to stairs climbed and sleep patterns. As the recent trend in research extends the use of these devices to a broader range of applications, questioning the reliability and accuracy of these devices became much more legitimate. In this research, we assess the public opinion on these devices through utilizing novel sentiment analysis techniques to build a fully automated aspect-based sentiment summarizer that transfers the sheer amount of Amazon reviews of these products to a user-friendly summary. Product features are extracted using the text of reviews, the description and features sections on Amazon. Another approach is also proposed that extracts the names of competing products and compares their reviews to separate the features from the other common nouns. To enhance sentiment classication, the system combines two sentiment lexicons, handles complex negation types through parsing while handling semantic relations, and assigns the sentiment tothe proper product and feature. The proposed summarizer’s components generally outperform the state-of-the-art methods with notable improvements in detecting product features, competing products and negation and can easily generalize to other domains.