A text-mining based model to detect unethical biases in online reviews: a case-study of Amazon.com

• Autor(a) principal: Costa, Ana Rebello de Andrade da
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/15297
• topic: Online reviews; Text mining; Sentiment analysis; VADER; Comércio eletrónico; Criação de valor; Text mining; Satisfação do cliente

The rapid growth of social media in the last decades led e-commerce into a new era of value co-creation between the seller and the consumer. Since there is no contact with the product, people have to rely on the description of the seller, knowing that sometimes it may be biased and not entirely truth. Therefore, reviewing systems emerged in order to provide more trustworthy sources of information, since customer opinions may be less biased. The problem was, once sellers realized the importance of reviews and their direct impact on sales, the need to control this key factor arose. One of the methods developed was to offer customers a certain product in exchange for an honest review. However, in the light of the results of some studies, these "honest" reviews were proved to be biased and skew the overall rating of the product. The purpose of this work is to find patterns in these incentivized reviews and create a model that may predict whether a new review is biased or not. To study this subject, besides the sentiment analysis performed on the data, some other characteristics were taken into account, such as the overall rating, helpfulness rate, review length and the timestamp when the review was written. Results show that some of the most significant characteristics when predicting an incentivized review are the length of a review, its helpfulness rate and the overall polarity score, calculated through VADER algorithm, as the most important sentiment-related factor.