Project Activities

Data stream processing

• Continual range query indexing for efficient stream processing, adaptive load shedding for stream joins

• Query indexing methods for managing continual range queries for location-based mobile services; evaluation of moving queries over moving objects

• Algorithms and systems for detecting bio-terrorist attacks via multiple distributed data channels from non-traditional sources, such as school attendance, work attendance, OTC drug sales, ER data, and insurance claims.
  

• Content-based predicate interval indexing for single attribute and multiple attributes, continual query monitoring, rule management, and XML document filtering.
  

• Intrusion detection rules monitoring
  

• Tools and techniques for web usage mining and analysis, personalization tools for e-commerce and ISP, collaborative proxy caching methods, Web infrastructure issues, web access billing methods, and load balancing issues among web content/proxy servers
  

• Mobile indexing and index broadcasting
  

• Buffer management
  

• Concurrency control, performance analysis, and buffer management

Software Prototypes Developed:

• A personalization system for corporate or ISP portal pages. Implemented in Java, it generates user dynamic profiles for personalized applications. It uses content-based collaborative filtering techniques to create user profiles, form user communities and make recommendations. The system analyzes user logs, fetches documents accessed and categorizes them. Each user is then described by a vector of document categories. Such user characterizations are then used to find user communities based on a projected clustering scheme. The log analysis and content categorization are run periodically off-line to capture dynamic user profiles, which are then used online for personalized applications.
  

• An Intelligent Recommendation Analyzer for ISP. It is a personalization tool that processes the proxy log and subscriber log of an ISP, generates the categories of documents that each subscriber browsed during each sign-on session, updates a subscriber's profile and then uses it to make personalized URL recommendations. This is useful for an ISP to effectively place banner ads on each subscriber's personalized page.
  

• An Intelligent Recommendation Analyzer consisting of a set of recommendation engines for E-commerce. It provides a set of APIs for an E-commerce application to construct personalized web contents. Various recommendation engines are implemented based on products, contents, and ratings as well as unique real-time Web access context.
  

• A Web usage mining tool employing user session inference algorithms and data mining technologies to derive user behavior in surfing a World Wide Web home site. Frequent user traversal paths and frequently visited group of pages by users are provided after identifying user sessions. The tool mines the user oriented statistics directly from the regular HTTPd log files. It does not require user registrations. The user oriented statistics provide a webmaster a set of rich information about client access behavior so that he/she can better organize a home page presentation. The tool has been incorporated into various IBM web server product offerings for different system platforms, including OS/390, AIX, HP UX, Sun Solaris, OS/2 and Windows 95/NT.