When Are Negative Online Reviews More Helpful Than Positive Reviews? A Multi-method Investigation Into Millions of Online Hotel Reviews

• Main Author: Lee, Sanghyub John (Author)
• Other Authors: Kim, Jungkeun (Contributor), Marshall, Roger (Contributor)
• Format: Others
• Published: Auckland University of Technology, 2020-06-04T04:07:34Z.
• Subjects: Big data; Hotel online customer reviews; Electronic word-of-mouth; Negativity bias and loss aversion; Machine learning model; Deep learning; Thesis
• Online Access: Get fulltext

 The role and volume of electronic word-of-mouth (e-WOM) online reviews worldwide are increasing rapidly so consumers, particularly in the tourism industry, may suffer from information overload. This, in turn, may impact on driving consumer behaviour. Thus, understanding factors that influence which reviews are perceived as helpful may be important for vendors in the tourism industry particularly in the hotel industry. This thesis suggests negativity bias and loss aversion as a theoretical anchor to illustrate the impact of star ratings on the perceived helpfulness of hotel reviews. Since prior research results appear to be diverse, there is a need to find which systemic moderators elicit different outcomes. This research seeks to provide a significant moderating role of reviews' differences such as consumer scepticism, and systematic information processing. A quantitative approach via big data consisting of over two million online hotel reviews was adopted to address the inconsistent results. This research offers an enhanced predictive effect instead of small sample sized surveys used in prior studies. By deploying spatial regression discontinuity design between one-sided and two-sided reviews on Booking.com, as well as Agoda.com and Booking.com reviews, I proposed and validated the moderating role of consumer scepticism. This is expressed as 'too good to be true'. To make the results of the analysis more robust and addressing a small statistical effect size, the effect of the independent variables is not only measured in the statistical methods (regression and PROCESS macro) but also in traditional (bi-logistic regression) and new machine learning techniques (deep-learning). The findings given in this work could offer pivotal implications for academics. They could additionally: (1) provide systemic moderators that elicit different outcomes; (2) illustrate a negative association between the review valence and the perceived helpfulness of the reviews; (3) document that when the level of consumer scepticism and heuristic information processing decreases, then negativity bias and loss aversion also weaken or are eliminated; (4) offer extensions for the broader research stream of e-WOM; and (5) extend the stream of research that utilizes big data with machine learning techniques. Limitations and directions for future research are discussed in the closing chapter.