Date of Award
2018
Degree Type
Thesis
Degree Name
Master of Science in Computer Science
Department
Computer Science and Statistics
First Advisor
Lisa DiPippo
Abstract
Nowadays, online reviews have become an important source of opinions that people refer to while making decisions. For instance, there are more and more people who refer to Yelp reviews to judge the quality of services that are provided by local businesses. Due to the popularity and guidance of online reviews, many reviews have been imposed for the purpose of either promoting or downgrading target services. Yelp develops its own automatic review recommendation algorithm, which has marked many suspicious reviews as Not-recommended Reviews. Yelp has automatically grouped its online reviews in two different categories, and it is a common question “What are the differences between Not-recommended Reviews and Recommended Reviews?”. One of the goals in this thesis is to explore the differences. Particularly, it employs the Text, one of the most important components of an online review, to develop six different sentiment features, i.e., Strong Positive, Strong Negative, Ordinary Positive, Ordinary Negative, Ordinary, and Strong, and study the differences in terms of sentiment between recommended reviews and not-recommended reviews. It has been found that not-recommended reviews usually contain more polarized (positive or negative) words.
In addition, online reviews are posed for services and products randomly. Generally, the reviews for a service/product are evenly distributed in their lifespan. However, it has been reported in the Amazon system that there are time periods where the reviews for some products are bursty. Put in other words, there are sudden concentrations of reviews in certain time periods. Another goal in this thesis is to investigate review bursts on Yelp. First, it is to explore the Date component of a review to develop the Density of Burstiness for the reviews of a business. Second, the normalized burstiness density has been introduced to select Density Periods, where reviews are mostly concentrated. It has been found that Yelp reviews have the following concentration observations, (1) the maximum burstiness density values for density periods vary significantly; (2) the review bursts often occur at the beginning days of the reviews’ lifespan; (3) some restaurants have multiple density periods.
Recommended Citation
Li, Na, "SENTIMENT FEATURES FOR YELP NOT-RECOMMENDED ONLINE REVIEWS STUDY" (2018). Open Access Master's Theses. Paper 1281.
https://digitalcommons.uri.edu/theses/1281
Terms of Use
All rights reserved under copyright.