three weeks ratings for both audience and netizen groups of each movie were collected from Naver platform; Customers’ online reviews, known as Online Word of Mouth (WOM), have become an important source of information to prospective customers for their potential purchasing decision. However, these reviews cannot be fully trusted due to the presence of fake reviews. Movie reservation platforms suffer more from those fake reviews, and some genre experienced a terror of fake reviews. Therefore, one would question about the contents that drive people to post fake reviews, the way such posts evolve during the projection period in theaters and about their impact on the performance of the movies. This dissertation aims to provide empirical answers to these questions. To address the first question, we collected for 107 movies released between 2014 and 2017 contents descriptions from online Korean newspapers and first week ratings from Naver platform having both real reservation group from the platform and non-reservation group. Text mining technique (Topic Modeling) was applied on the contents descriptions and resulted in 16 different contents categories. That approach was necessary because the original classifications provided by the platform are poorly linked to their contents. To investigate the impact of the defined contents categories on fake reviews posting behavior, we first applied the Difference in Difference (DID) method proposed in the literature to estimate fake reviews in Naver reviews by subtracting the ratios of audience extreme ratings from the ratios of netizen extreme ratings. The Ordinary Least Square estimation showed that in the Korean context, fake reviews are sensitive to some movie categories such as political, historical and war contents.
The second part of the dissertation focuses on the dynamic of the fake reviews and their impact on movies sales. To this purpose, we constructed a dataset of 392 movies released between 2014 and 2018 on Korea; performance information were collected from the Korean Film Council website. Using System GMM to estimate our dynamic empirical models, analysis results on the dynamic of fake reviews (which were estimated using the DID approach) show that negative and positive fake reviews are highly linked with each other where contemporaneous negative (positive) fake reviews will generate more positive (negative) fake reviews in the following day; however, contemporaneous positive and negative fake reviews are negatively correlated with each other. This posting dynamic can be explained by the Spiral of Silence theory, as each opinion’s sides try to post negative or positive reviews in an attempt to influence the platform users’ opinions about a quality of a movie. Given this behavior, do fake reviews really matter? Using three weeks daily ratings of Naver audience and netizen groups, we use again System GMM estimator for dynamic panel data analysis to estimate the differential effects of real audience ratings and netizen ratings on daily and cumulative sales with our dataset. The results show that only audience ratings have predicting power on daily movies sales, reflecting users’ awareness regarding the existence of fake reviews and their trust in only genuine information when available. The findings of this dissertation provide valuable contributions for academicians and practitioners.