Social Media and the Diffusion of an Information Technology Product

The expansion of the Internet has led to a huge amount of information posted by consumers online through social media platforms such as forums, blogs, and product reviews. These text data are useful especially when numeric sales data are not enough, as is typically the case with new product diffusion. This study proposes a diffusion model that accommodates pre-launch social media information and combines it with post-launch sales information in the Bass model to improve the accuracy of sales forecasts. The model is characterized as the extended Bass model, with time varying parameters whose evolutions are affected by the consumer's communications in social media. Specifically, we first extract information from social media to build variables, such as the number of positive and negative comments, and also latent topics. These data are fed as key parameters in the diffusion model's evolution process for the purpose of plugging the gap between the time-invariant key parameter model and that of observed sales. We examine several models using text analysis techniques, e.g., sentiment analysis for counting numbers of positive and negative comments and topic analysis by topic model to extract relevant topics. These results are then compared with the conventional Bass model using only post-launch sales data. An empirical study of the first-generation iPhone during 2006 and 2007 shows that the model using additional variables extracted from sentiment and topic analysis on BBS performs best based on several criteria, including DIC (Deviance Information Criteria), marginal likelihood, and forecasting errors of holdout samples. We discuss the role of social media information in the diffusion process for this study.