مدل اثرات تصادفی برآورد اثربخشی تبلیغات در بازارهای آنلاین

This paper presents an application of the Bayesian Markov Chain Monte Carlo (MCMC) used to select cost-effective ad spots in online marketplaces. Due to the rise of electronic commerce, the online advertising industry which is highly complex undergoes rapid changes. And there are plenty of studies that keep coming up with the similar methodologies to predict click-through rates for ad spots. Previous research has mainly considered the following models: a logistic regression model and a binomial model connected by a linear link function. However, it is problematic to directly apply the existing online advertising effect models to the click-through data of online marketplaces. Because generally a click-through rate is fairly low so that a small change in its rate might give a somewhat larger prediction error in terms of click-throughs. We propose a Bayesian Poisson-gamma model to predict click-throughs instead of their rates and further extend to incorporate random effects in order to account for heterogeneity of variance between keywords. Our results may help guide online advertisers in decision-making.

As the Internet usage continues to surge, the online advertising industry grows dramatically. For instance, the volume of the online advertising market in South Korea skyrocketed from $6.625 billion in 2005 to $12.409 billion in 2007 while the volume of whole advertising market only increases from $70.539 billion to $79.897. It means that not only the portion of online advertising market surged but also the importance of the online advertising market soared. In the next few years, the size of the online advertising market can overtake the TV advertising one. Increasing Internet penetration rate and the ones’ usage and surging e-commerce transaction are the major causes that lead rapid growth of the online advertising market. In this circumstance, it can be said that now Internet has become the crucial medium for advertising such as traditional medium (e.g., TV, radio, newspaper, magazine). Recently the support of database allows researcher to analyze the data at the individual level and it become a new fashion of analyzing advertising effect at the personal level. It is no surprise that there has been a lot of research conducted on measuring the effectiveness of online advertising at the personal level (e.g., Chatterjee et al., 2003 and Manchanda et al., 2006). Especially a traditional sales-advertising model and banner advertising model have been developed and applied to estimate the online advertising effect. When e-commerce is in its infancy, people are not accustomed to buying the product in online mall and afraid of being deceived especially when they are going to buy products from unknown sellers (e.g., C2C e-market place). However, as the volume of transactions in online market increases, people get used to shop in the online mall and the size of the one gets bigger and bigger. Especially in the Internet, “open market” has emerged as a place where people can register easily and do their business (e.g., eBay), and we name the online mall such as eBay an “online marketplace”.

This study introduced the Poisson-gamma model to the online advertising effect model and developed an improved model with a time-decaying effect that is applicable for the click-through data in the online marketplace. The proposed model has been experimentally proved to have advantages over the existing advertising effect models. It captured the effectiveness of each advertisement and showed significantly diminished variances compared to the previous models. This study made important theoretical contributions as follows: First, it is the first study that has adopted the Poisson-gamma model to investigate the effect of online advertising. A key contribution of this research is the development of a model that can effectively capture the effectiveness of online advertising in the online marketplace where many buyers and sellers exist and the purchase decision is made within a very limited number of visits compared to other e-marketplaces. Second, the proposed model can incorporate click-through data and is able to give inferences about advertising effects on customers who visit only once before they drop out. Third, this study has introduced the time-decaying effect based on previous research and developed a Poisson-gamma model employing that effect for click-through data. It enables the Poisson-gamma model to capture the consumers’ gradual forgetfulness of the advertisement.