Computer Science Theses and Dissertations

Permanent URI for this collectionhttp://hdl.handle.net/1903/2756

Browse

Filters

1985 - 198911990 - 19942

Settings

Start date: 1980End date: 1999

Search Results

Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    Restructuring Textual Information for Online Retrieval
    (1985) Koved, Lawrence; Shneiderman, Ben; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md)
    Two experiments were conducted to evaluate two styles of online documents. The first experiment compared paper manuals to online manuals using two different database structuring techniques - a sequential (linear) structure and a tree structure. People using the paper manuals were faster at solving problems than the people using the computer manuals. No differences were found between the linear and tree structures, or in accuracy of problem solutions. In a subjective evaluation of user preferences, the computer manuals were rated as better and more organized than the paper manuals. The second experiment compared two methods of retrieving online information that allowed the reader to specify the attributes needed to guide the information retrieval process. The first manual recorded the attributes entered by the reader via menus, and material in the manuals not relevant to the current search was pruned from the search space. The second manual did not record the menu selections, and the readers repeatedly entered the attributes several times in order to complete the task. The manual that recorded the attributes allowed the readers to work over twice as fast and was pref erred over the other manual. A theoretical foundation is presented for the underlying online documentation used in the experiments. The user's traversal through the database is presented as a graph search process, using a production system. The results of the experiments and their theoretical foundations are evaluated in terms of the impact they might have on future online document storage and retrieval systems.
  • Thumbnail Image
    Item
    Adaptive Database Systems Based On Query Feedback and Cached Results
    (1994) Chen, Chung-Min; Roussopoulos, Nick; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md)
    This dissertation explores the query optimization technique of using cached results and feedback for improving performance of database systems. Cached results and experience obtained by running queries are used to save execution time for follow–up queries, adapt data and system parameters, and improve overall system performance. First, we develop a framework which integrates query optimization and cache management. The optimizer is capable of generating efficient query plans using previous query results cached on the disk. Alternative methods to access and update the caches are considered by the optimizer based on cost estimation. Different cache management strategies are also included in this framework for comparison. Empirical performance study verifies the advantage and practicality of this framework. To help the optimizer in selecting the best plan, we propose a novel approach for providing accurate but cost-effective selectivity estimation. Distribution of attribute values is regressed in real time, using actual query result sizes obtained as feedback, to make accurate selectivity estimation. This method avoids the expensive off-line database access overhead required by the conventional methods and adapts fairly well to updates and query locality. This is verified empirically. To execute a query plan more efficiently, a buffer pool is usually provided for caching data pages in memory to reduce disk accesses. We enhance buffer utilization by devising a buffer allocation scheme for recurring queries using page fault feedback obtained from previous executions. Performance improvement of this scheme is shown by empirical examples and a systematic simulation.
  • Thumbnail Image
    Item
    Treemaps: Visualizing Hierarchical and Categorical Data
    (1993) Johnson, Brian Scott; Shneiderman, Ben
    Treemaps are a graphical method for the visualization of hierarchical and categorical data sets. Treemap presentations of data shift mental workload from the cognitive to the perceptual systems, taking advantage of the human visual processing system to increase the bandwidth of the human-computer interface. Efficient use of display space allows for the simultaneous presentation of thousands of data records, as well as facilitating the presentation of semantic information. Treemaps let users see the forest and the trees by providing local detail in the context of a global overview, providing a visually engaging environment in which to analyze, search, explore and manipulate large data sets. The treemap method of hierarchical visualization, at its core, is based on the property of containment. This property of containment is a fundamental idea which powerfully encapsulates many of our reasons for constructing information hierarchies. All members of the treemap family of algorithms partition multi-dimensional display spaces based on weighted hierarchical data sets. In addition to generating treemaps and standard traditional hierarchical diagrams, the treemap algorithms extend non-hierarchical techniques such as bar and pie charts into the domain of hierarchical presentation. Treemap algorithms can be used to generate bar charts, outlines, traditional 2-D node and link diagrams, pie charts, cone trees, cam trees, drum trees, etc. Generating existing diagrams via treemap transformations is an exercise meant to show the power, ease, and generality with which alternative presentations can be generated from the basic treemap algorithms. Two controlled experiments with novice treemap users and real data highlight the strengths of treemaps. The first experiment with 12 subjects compares the Macintosh TreeVizTM implementation of treemaps with the UNIX command line for questions dealing with a 530 node file hierarchy. Treemaps are shown to significantly reduce user performance times for global file comparison tasks. A second experiment with 40 subjects compares treemaps with dynamic outlines for questions dealing with the allocation funds in the 1992 US Budget (357 node budget hierarchy). Treemap users are 50% faster overall and as much as 8 times faster for specific questions.