This Small Business Innovation Research (SBIR) Phase I project will apply recent advances in knowledge discovery to bridge the gap between what is known about an Internet viewer and what is done with this knowledge to improve user experiences and business outcomes. Recent machine learning research has shown that latent group models perform extremely well compared to other relational probabilistic models (such as the more traditional relational Bayesian networks) in most problem categories. This research will investigate if latent group models can help a publisher make better publishing decisions. Online publishers operate in an environment of massive quantities of input data from disparate sources, non-homogeneous attribute data, and a business requirement for computation agility. Solving this problem will require advances in modeling, algorithmic, and implementation technologies. Today, online content publishers aggregate enormous volumes of data about their viewers from their web logs, registration systems, third-party web analytics providers, and ad serving systems. Mostly, these systems operate independently with a primary focus on describing what has happened. For example, a web site analytics package can best describe how many visitors came to this page yesterday, while an ad management system accurately counts how many ads were served on this section last month. Through analysis these tools can provide information used primarily for medium to long-range planning. None of these tools assist a publisher answer the question, "what does this viewer want from my site on this page at this point in time?" Answering this question is the key to unlocking a new path to growth for the online content publisher. If the publisher can anticipate the needs of its users, it can better hone its content and navigation to the specific needs of its diverse audience. This in turn leads to improved viewer satisfaction and more time spent, consuming more content from the content publisher
