Allen Luniewski

Information organization using Rufus

Computer system users today are inundated with a flood of semi-structured information, such as documents, electronic mail, programs, and images. Today, this information is typically stored in filesystems that provide limited support for organizing, searching, and operating upon this data, all operations that are vital to the ability of users to effectively use this data. Database systems provide good function for organizing, searching, managing and writing applications on structured data. Current database systems are inappropriate for semi-structured information because moving the data into the database breaks all existing applications that use the data. The Rufus system attacks the problems of semi-structured information by using database function to help users manage semi-structured information without requiring that the users information reside in the database.

A design for fine-grained access control in Melampus

The authors believe that access controls for object-oriented systems should be fine-grained and thus apply to individual methods of individual objects. Efficiently supporting this approach is a challenging tasks, because (at least conceptually) a check is done on every method invocation. Their design uses access control lists and exploits virtual memory facilities to make these checks run fast. The costs include an extra level of indirection for method invocation and per-user storage for preprocessed access control information.<<ETX>>

An extensible classifier for semi-structured documents

In this paper, we present a vector space classifier for determining the type of semi-structured documents. Our goal was to design a high-performance classifier in terms of accuracy (recall and precision), speed, and flexibility. The ability to dynamically extend a classifier with userspecific classes is crucial for many applications. Unfortunately, the training data of existing classes is often not available, such that the extended classifier is imprecise a+ a result. We focus on this issue. First, we evaluate how to create class abstracts that can be used as training data replacement. Second, we introduce relevance feedback learning strategies to overcoming the remaining classifier flaw.