Abstract: |
An event history is a collection of events that have occurred in an event-based system over a period of time. There can be various types of events, among which are the temperature changes and power demands in a power management system, client requests for data items in a broadcast system, price increase of a stock in stock market, and so on. There is a lot of interesting information that can be extracted from an event history via data mining techniques. Our purpose in this thesis is to propose methods for extracting this useful information in the form of event sequences and event associations from a single or two correlated event histories. We also aim to show how the results of the mining process can be used for active and mobile data management. The results of the mining process demonstrates the relationships among the events which are generally captured as associations or sequences. The relationships among the events are shown to be a useful tool to enhance an event-based system via event organization, predictive event detection, and proactive rule execution.
We consider the mining of both a single event history and two correlated event histories. We first propose a method for mining binary relationships from a single event history. The binary relationships among events are used to organize the events into related groups of event. This organization is important because the number of events in an event-driven system may become very high and unmanageable. The groups of related events and the relationships among the events are exploited for predictive event detection and proactive rule execution in active database systems. We also consider the mining of two correlated event histories which are disjoint and the events in one history are related to the events in the other history. We describe how we can efficiently extract associations among the events spanning different event histories, which we call cross associations.
We have chosen data broadcast in mobile computing environments as a case study for active data management. One of the important facts in mobile computing environments with wireless communication medium is that the server-to-client (downlink) communication bandwidth is much higher than the client-to-server (uplink) communication bandwidth. This asymmetry makes the dissemination of data to client machines a desirable approach. However, the dissemination of data by broadcasting may induce high access latency in case the number of broadcast data items is large. Our purpose is to show how active data management can be used to improve mobile data management through broadcast data organization and prefetching from the broadcast medium. In order to achieve this, the client requests of data items at the server are considered as events and the chronological sequence of items that have been requested by clients is considered as an event history. We call the event history in broadcast medium a broadcast history. The first step in our work is to analyze the broadcast history to discover sequential patterns describing the client access behavior. The sequential patterns are used to organize the data items in the broadcast disk in such a way that the items requested subsequently are placed close to each other. Then we utilize the predictive event detection techniques to improve the cache hit ratio to be able to decrease the access latency. Prediction of future client access behavior enables clients to prefetch the data from the broadcast disk based on the rules extracted from the broadcast history.
|
