KAVANAH: AN ACTIVE USER INTERFACE INFORMATION RETRIEVAL APPLICATION

摘要

This paper reports our implementation and evaluation of an active user interface in an information retrieval application called Kavanah. The goal of the active user interface is to improve the quality of information retrieval and to reduce the user's cognitive workload while searching for information. Our underlying concept is to dynamically construct the search queries based on a dynamic representation that captures user interests, preferences and searching context (as represented in a user ontology). Our approach to disaggregating the essential aspects of a user's intent for searching allows for focused multi-agent based construction and correction of the overall user model that captures the user's intent, thus promoting increased effectiveness and efficiency. We evaluate the effectiveness of the active user interface with commonly used metrics from the information retrieval community by measuring retrieval performance with and without the presence of an active user interface. Furthermore, we measure the ability to discover new knowledge by evaluating our dynamic online ontology construction. The evaluations use the Unified Medical Language System knowledge base as a test bed.