| Online user interaction is a topic of considerable current interest, both from a research as well as from a practical perspective. Virtually all online user interaction technologies in use today (e.g., personalization and customer relationship management software) are based on the notion of storing as much historical customer session data as possible, and then querying this data store in order to react to customers (e.g., offering a discount on an item that the user has shown interest in). The holy grail of online user interaction is an environment where fine-grained, detailed historical session data can be queried based on current online navigation patterns for use in formulating near real-time responses. Unfortunately, most existing online user interaction technologies are unable to scale to support the high user loads and large volumes of customer data that are typical of many e-commerce sites today. Providing true online user interaction requires that data be retrieved from large persistent databases within subsecond time frames, and typically this must be done under heavy user loads. Thus, the primary bottleneck lies in the underlying database systems—existing database systems cannot effectively support these requirements.; This research attempts to present an approach to perform true online user interaction. The proposed framework entails: (1) observing specific instances of online behavior, (2) correlating this specific behavior with the vast amounts of historical behavior collected over time, and (3) reacting to the user. Our solution approach consists of two key ideas: (1) a data warehouse to store historical behavior, and (2) rule caching to track online behavior and correlate with the historical data. An implementation of an online user interaction system based on the proposed framework is presented, along with a set of performance results, which indicate that the system is indeed capable of providing near real-time responses, even under heavy user loads. |
