Modeling word-of-mouth: online reviews and movie industry

Abstract

This thesis consists of three studies investigating word-of-mouth(WOM) and corresponding strategies for consumers and companies. There are recent developments in literature of word-of-mouth (WOM). First, the researches on the firm-created WOM are emerging. Because the importance of WOM activity is well-known for firms, they put a lot of efforts to create or facilitate the positive WOM for their products nowadays. These “promotional” reviews are interesting new phenomena. Second, the needs to decompose the dimensions of WOM rise in literature. The valence and volume of online reviews are most commonly used dimensions of WOM. Some researchers find that further decomposition of dimensions helps to identify the role and effects of WOM. Our studies contribute to the recent literature of WOM theoretically, empirically and methodologically. In chapter 2, we develop a new method incorporating Bayesian outlier detection and data augmentation techniques. The proposed method enables us to investigate phenomena involving unobservable independent variables such as promotional reviews. The major challenge of researches on the firm-created WOM arises from the fact that the firms hardly reveal their efforts on promotional WOM activities. Instead of direct observation, we can indirectly infer the promotional reviews through abnormal review posting behaviors. When firms try to manipulate online reviews, they inevitably need to post a large volume of reviews, and we can observe abnormal volume of reviews on the incidence. Our method detects these outliers using Bayesian outlier detection technique, and then includes them as data for further modeling under the data augmentation framework. The method can be applied for similar research environment that includes unobservable, yet indirectly inferable, independent variables. In chapter 3, we examine the effects of online and offline word-of-mouth on box-office sales. We model online and offline word-of-mouth using the Bayesian learn...