Abstract:
Delivering personalized services to subscribers is a relatively new challenging area for GSM content providers. One of the most effective ways to personalize mobile services is to make use of the location of the mobile device and to provide location based services. Example location based services are receiving alerts about a traffic jam, notification of a sale on gas, issuing discount coupons to customers nearing a store and providing information about playing movies in cinemas around. Applications can also be developed that will inform users about other users in close vicinity that have matching profiles, this type of application for example can be used for friend finding. However tracking users is a very expensive operation. The goal is to decrease the number of location queries sent to GSM operators when tracking users. A solution is presented to track users cost efficiently by predicting future locations of users. The algorithm proposed extracts movement patterns of subscribers and uses these patterns for location prediction. Two types of patterns, stationary and movement, are mined from movement data and two types of predictions, point and path, are used for predicting location. Real location data of many users provided by a GSM operator are used to evaluate the algorithm. Simulations are also performed with artificially generated log data. Results demonstrated that the number of location queries is reduced significantly while keeping error rate in an acceptable level.
