Application-Embedded Retrieval from Distributed Free-Text Collections

Abstract

A framework is presented for application-embedded information retrieval from distributed free-text collections. An application’s usage is sampled by an embedded information retrieval system. Samples are converted into queries to dis-tributed collections. Retrieval is adjusted through sample size and structure, anydata indexing, and dual space feedback. The framework is investi-gated with a retrieval system embedded in a text processor. The Problem The integration of query generation and feedback challenges information retrieval (IR) technologies. Rel-evance feedback (Aalbersberg 1992; Robertson and Walker 1997), while useful to information science pro-fessionals (Spink and Saracevic 1997), is frequently in-appropriate for lay users. Many users are unable to ade-quately state their information needs in queries (Burke, Hammond, and Young 1996). The constant, explicit so-licitation of relevance judgments interferes with many users ’ information-seeking behaviors (Kulyukin 1998c). Feedback utilization in the query space alone prohibits retrieval adjustments over multiple interactions (Ku-lyukin 1998a; Kulyukin, Hammond, and Burke 1998). The focus on maximizing the number of relevant docu-ments in every retrieval is inconsistent with many users’ information needs (Jansen et al. 1998). The approach to the integration of query generation and feedback henceforth presented is called application-embedded distributed IR. The approach is implemented in an embedded IR (EMIR) system for text proces-sors. The objective is a technology for building non-intrusive, feedback-sensitive IR tools that users embed into their applications to tap into and monitor informa-tion sources, while engaged in routine usages of those applications. The technology is motivated by a growing number of sources with a wealth of data, but few tools to timely put it to use.