Do online review readers react differently when exposed to credible versus fake online reviews?
利用韩国Naver网站不同评论发布政策产生的独特数据，区分可信与虚假评论，通过文本分析发现虚假评论内容更肤浅，且不同可信度对票房收入影响不同。
Marketing research on online reviews has attempted to understand the antecedents and consequences of review manipulation. Building on the elaboration likelihood model (ELM), this study deploys a rare dataset that allows distinguishing credible from less credible (and likely fake) online reviews by means of the online review posting policy adopted by the movie review website Naver.com. We use text analysis entailing word embedding and topic modelling techniques such as Latent Dirichlet Allocation, to capture content depth across different types of online reviews (credible vs manipulated). Furthermore, we explore how differences in the textual content of credible vs manipulated online reviews affect customer purchase decisions. Our results highlight that less credible reviews tend to contain more superficial information compared to more credible reviews, and that different levels of source credibility lead to distinctively different impacts of online reviews on box office revenue.