Back to search

Cornell University

Debiasing Online Reviews with Machine Learning

Abstract

dc:description.abstract

Online review aggregation platforms suffer from biases which can lead to distress to both the platforms and their consumers as the average rating crowd sourced on these platforms do not represent the correct perceived quality of the product or service. We look at the problem of polarization bias on Yelp and present the evaluation of an estimation model to determine the unbiased average rating. We explore how the Yelp's elite membership program helps in cutting down the bias. Our results propose that the average biased rating listed on platforms is correlated with the true unbiased rating and by including more information from online reviews as input features of our model we can get a reasonably well estimate of the average unbiased ratings. We propose and compare several predictive models to estimate unbiased average ratings and show how textual data can play a critical role in enhancing the predictive power. Our results can help review aggregation platforms to determine mechanisms to cut down the bias on their platforms and can benefit businesses and consumers on their platform to access a fairer metric for service quality.

Degree

thesis:*
Name thesis:degree_name
M.S., Management
Level thesis:degree_level
Master of Science
Discipline thesis:degree_discipline
Management
Grantor
Cornell University
Year dc:date.issued
2020

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Goel, Shivank
Chairs dc:contributor.chair
  • Chen, Li
  • Mankad, Shawn

Rights

Language dc:language.iso
en

Identifiers

dc:identifier.*
Dc Identifier Bibid
13254371
Dc Identifier Other
ProQuest Submission ID: 10864
ProQuest Publication ID: 27962175

Chain of custody

source
Harvested from
Cornell University
Base URL
ecommons.cornell.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Goel, Shivank. Debiasing Online Reviews with Machine Learning. Master of Science thesis, Cornell University, 2020. https://hdl.handle.net/1813/70258