SIGIR 2007 Proceedings

Demonstration

ISKODOR: Unified User Modeling for Integrated Searching
Inst. of Computer Science III University of Bonn and Fraunhofer Institute IAIS Sankt Augustin, Germany

Melanie Gnasa

Inst. of Computer Science III University of Bonn Bonn, Germany

Armin B. Cremers

melanie.gnasa@ iais.fraunhofer.de

abc@iai.uni-bonn.de

Coll. of Info. Studies/UMIACS University of Maryland College Park, MD 20742 USA

Douglas W. Oard oard@umd.edu

ABSTRACT
iskodor integrates p ersonal collections, p eer search, and centralized search services. User modeling in iskodor fills three roles: discovery of sites with suitable information stores, context-based query interpretation, and automatic profilebased filtering of new information. Explanation and control are achieved through graphical depiction of sources, explicit feedback regarding utility, and explicit control over p eer association b ehavior and information sharing.

Query

Client

Extract Index Terms

Web Search Engine(s)

IIS System

Ranked List

Rating

Discover Communities

Categories and Subject Descriptors
H.3.3 [Information Storage and Retrieval]: Information Search and Retrieval

Communities

Extract Index Terms

Extract Index Terms

Personal Search Memory

General Terms
Algorithms, Exp erimentation

Figure 1: Integrated Searching with Iskodor grated search concept is similar to traditional information retrieval. The main difference arises from the unified user model which fills three roles: · iskodor helps the user identify and select p eers who share common interests. The client maintains a search history which can b e used to indicate the utility of prior search results. Communities are nominated automatically based on the search histories that their users choose to share; nominations accepted by the local user can then guide future p eer search. · iskodor ranks results from local search, p eer search, and web search based on context-dep endent query interpretation. The context for a query draws on the client's search history and on shared search histories by community memb ers. · iskodor p erforms automatic profile-based filtering for new information. This process aims to provide awareness of documents which have b een found by other community memb ers. All three roles of the unified user model are implemented in a working prototyp e in which a graphical user interface provides for explanation and control.

Keywords
Web search, explicit relevance feedback, collab orative search

1.

MOTIVATION

User modeling for information retrieval is an active research area to improve the effectiveness of information access. Two new research directions have b een explored in different settings. Lu and Callan [3] demonstrated that user modeling in p eer-to-p eer networks significantly improves the efficiency of federated full-text search without degrading its accuracy. In the context of web search, the exp eriments of Agichtein et al. [1] demonstrate a significant improvement of the accuracy of comp etitive web search ranking algorithms in a realistic environment by incorp orating user b ehavior information. Our goal is to integrate p ersonal collections, peer search, and centralized search services in order to improve web search accuracy. Thus, a user can b e made aware of information providers that have contributed to his or her search result, and he gets a p ersonalized access to large information storage. We refer to this integration as congenial web search [2].

2.

SYSTEM DESIGN

3. REFERENCES
[1] E. Agichtein, E. Bill, a. S. Dumais. Improving web search ranking by incorp orating user b ehavior information. In SIGIR, 2006. [2] M. Gnasa, M. Won, a. A. B. Cremers. Three pillars for congenial web searching--continuous evaluation for enhancing web search effectiveness. Journal of Web Engineering, 3:252­280, 2004. [3] J. Lu a. J. Callan. User modeling for full-text federated search in p eer-to-p eer networks. In SIGIR, 2006.

The architecture of the iskodor prototyp e integrates commercial web search engines in a virtual p eer-to-p eer network. All users are organized in the network as p eers and maintain a p ersonal collection. Figure 1 shows a flow diagram of an integrated search with iskodor. In general, the inteCopyright is held by the author/owner(s). SIGIR'07, July 23­27, 2007, Amsterdam, The Netherlands. ACM 978-1-59593-597-7/07/0007.

898