Visual Information Retrieval Environments

Jin Zhang

School of Information Sciences
University of Pittsburgh

In this dissertation, two visual information retrieval models, the distance-angle-based visual retrieval tool and the angle-angle-based visual retrieval tool are presented. Within the visual retrieval environment, semantic relationships among documents and reference points are demonstrated, five traditional information retrieval models(distance model, cosine models, conjunction model, disjunction model and ellipse model) are interpreted, five new information retrieval models (advanced distance model, advanced cosine model, advanced conjunction model, advanced disjunction model and control-bar model) are developed, documents are browsed, three different metrics (L1, L2 and Lmax) are provided, three similarity measures (the distance-based similarity measure, the angle-based similarity measure and the distance-angle integrated similarity measure) are offered, reference points are generated and adjusted based on feedback information, and ambiguity is alleviated. A distance-angle integrated similarity measure, which takes the strengths of both a distance-based similarity measure and an angle-based similarity measure, is developed and integrated into the distance-angle-based visual retrieval environment. A prototype of the distance-angle-based visual retrieval tool, the DARE application for window, is implemented using Visual C++ (MFC) programming language.

The effects of the information retrieval models, the similarity measures, their interactions, the newly developed information retrieval models, the traditional information retrieval models, the two-reference-point-based information retrieval models vs. the one-reference-point-based information retrieval models, and the distance-based information retrieval models vs. the angle-based information retrieval models on the retrieval performance within the DARE are investigated.

Future research directions on the topics are addressed, which include implementing a new zoom function in the visual space, updating the current DARE version to a web-based version, comparing the retrieval performance between the DARE and a traditional information retrieval system and between the DARE and other visual retrieval tools.

Return to Seminar home page / ASIS home page.