Multiple facets of personalization


Panel Overview and Motivation

Personalization of information access can involve customization of information, its presentations and interaction style. The need for personalization of interaction in information systems such as Digital Libraries (DLs) lies in two major areas: one having to do with performance of DLs, understood primarily as the goodness of search results; the other with the user's experience in interacting with the DL. For some time, important segments of the DL and information retrieval (IR) research communities have agreed that major improvement in the performance of DLs will now come primarily through taking increasing account of the users of DLs, and the contexts and situations in which they find themselves.

User's interactions with DLs at the moment follow the classic model of “one size fits all”; that is, there is typically no attempt to adapt the manner in which the user can interact with the DL by taking account of a particular user's characteristics, situation, goals and other contextual factors. It is however clear that different users of DLs (or even the same user at different times) will prefer to engage with the DL in quite different ways, depending upon their cognitive abilities, their previous experience with DLs or with the topic at hand, their preferences in information seeking for browsing or for searching, and so on. The ability of a DL to adapt to such differences could lead to a qualitatively enhanced user experience for all users, since they would not have to adapt to the one mode of interaction which a typical DL provides. But, once again, although some research has been done on various aspects of such personalization, for the most part, the situation remains one of the user adapting to the DL, rather than vice versa.

This panel directly addresses the issue of personalization of interaction with information within a DL, by considering a small number of facets of personalization that either have been shown or are predicted to affect or influence information seeking behavior. Belkin (2006) suggests that personalization of interaction with information can be characterized according to evidence associated with the following facets:

  • Relevance/usefulness/interest

  • Task

  • Problem state

  • Personal characteristics

  • Personal preferences

  • Context/situation.

The focus of the panel will be on combining evidence for personalization of information access. Panelists will talk about different personalization facets they have investigated, and address the issue of interaction among the facets (either theoretically or empirically). Examples of the different facets of personalization that could be addressed are user interest, intention, cognitive differences, task, and domain/task knowledge. The discussion will be led by the moderator.

Panelists' Viewpoints

The Person's Knowledge by Susan Gauch

There are many different possible sources of information about a user's interests and many different technologies possible for collecting that information. Implicit collection techniques have the advantage that they place no burden on the user, but they may have drawbacks in terms of the accuracy of the information collected and their possible infringement on the user's privacy without the user's knowledge. Implicitly created profiles also have the advantage of adapting over time, rather than being static, but we need to understand better how to identify, and exploit, short-term interests related to the current task and long-term, general user interests. Users' interests may be represented by their web browsing histories, their search histories, the contents of documents on their local hard drive, and/or their currently open windows. Open challenges remain on how and when to use (and how to combine) these various potential sources of information, how to use them to build an accurate user profile, how to use the profile to assist the user in their tasks, and how to keep a profile up to date over time. As user profiles become used in more and more personalized systems, we need some way to create portable profiles that can be created once but shared by multiple applications.

The Workplace Situation (Context/Situation/Task facet) by Luanne Freund

Information access in the workplace is shaped by situational factors characteristic of different professional work settings. Chief among these factors are: expertise, professional roles, work tasks and information tasks. Evidence from a study of workplace information access will be presented to show that variation in these factors influences the nature of information that is considered to be useful and the extent to which it is accessible. To find information that meets their situated needs, searchers adopt strategies such as seeking out document genres that are well-suited to particular tasks. By modeling these patterns of contextualized information searching, we can design systems capable of retrieving results tailored to a searcher's situation. This will be demonstrated in the form of a prototype system that exploits relationships between tasks and genres to customize search results.

Cognitive Differences (Personal characteristics facet; Task facet) by Jacek Gwizdka

People differ with respect to their information processing ability and their preferred cognitive style. These differences affect how they interact with information search systems. I argue that personalization should take into account a whole range of factors, including the person's cognitive abilities. In the world of scarce attention, a system that does not match cognitive abilities may require extra cognitive processing and impose an unnecessary cognitive load. This extra load may prevent the person from completing their information tasks and may even lead to the system avoidance or abandonment. I will present some findings that demonstrate the effects of the cognitive differences among people on their execution of information tasks.

Information Relevance and User Intention (Relevance/usefulness/interest facet) by Susan Dumais

Most retrieval systems are designed to support people in finding new information. Yet many tasks that information workers, students and researchers conduct involve re-finding previously seen information. For example, 40% of the web queries that individuals submit are queries that they have issued before. Similarly, 60–80% of web pages that people visit are pages that that have seen before. I describe prototypes that we have developed to better support people in both finding and re-finding, using adaptive methods for query specification, ranking and results presentation.

Panel structure

  • 1Introduction to the topic: 10 minutes, by the moderator (Belkin);
  • 1Four presentations by the panelists on work done, highlighting the main facet in their investigations, and facets affecting its use for personalization: each 10 minutes;
  • 1Discussion amongst the panelists based on questions posed to them by the moderator, both in advance of the panel, and based on the presentations, with contributions/questions from the audience: 30 minutes;
  • 1Summary discussion from the panel, led by the moderator: 10 minutes.


Panel Participants' Bios

Jacek Gwizdka (Panel Organizer & Contact Person) Assistant Professor, SCILS, Rutgers University, 4 Huntington St., New Brunswick, NJ 08901, USA Voice: 732-932-7500 ex.8236, E-mail:

Dr. Jacek Gwizdka is an Assistant Professor in the Department of Library and Information Science, within the School of Communication, Information and Library Studies, at Rutgers University. Dr. Gwizdka studies affects of cognitive differences among people on their interaction with information systems. He is interested in using rich interaction logging for unobtrusive identification of user states and characteristics. He currently works on the Personalization of the Digital Library Experience project (PooDLE: His research includes the study of email use and email message management, as well as work on interaction mechanisms for adding metadata to electronic notebooks and to collections of digital photos. He is a contributing author to the edited PIM volume. Dr. Gwizdka conducted research at Xerox PARC, Hewlett Packard Research Labs, and Fuji Xerox Palo Alto Labs. Dr. Gwizdka holds one patent relating to PIM. He holds a PhD in Human Computer Interaction from the Department of Mechanical and Industrial Engineering at the University of Toronto. For more information please visit: and∼jacekg/

Nicholas Belkin (Co-organizer & Moderator) Professor II, SCILS, Rutgers University, 4 Huntington St., New Brunswick, NJ, USA

Nicholas Belkin has been Professor of Information Science in the School of Communication, Information and Library Studies at Rutgers University since 1985. Prior to this appointment, he was Lecturer, and then Senior Lecturer in the Department of Information Science at The City University, London, from 1975. He is the recipient of the ASIST Outstanding Teacher and the ASIST Research Awards, and of the ASIST Award of Merit, as well as having served as Chair of the ACM SIGIR. His current research interests focus on the personalization of IR systems, on IR systems which support multiple information seeking strategies, and on the development of a theory of information retrieval as interaction with text. Current or recent projects include classification of interactions with information, design for IR systems which support multiple interactive information seeking strategies, studies in how users understand relevance feedback and ranking in IR, experiments in the combination of evidence for IR, interface design for IR systems, the use of language modeling techniques for developing user models for IR systems, and the personalization of the digital library experience (

Susan Dumais (Panel Participant), Principal Researcher, Context Learning and User Experience for Search (CLUES), Microsoft Research

Susan Dumais is interested in algorithms and interfaces for improved information retrieval, as well as general issues in and human-computer interaction. She joined Microsoft Research in July 1997, and works on a wide variety of information access and management issues, including: personal information management, web search, question answering, information retrieval, text categorization, collaborative filtering, interfaces for improved search and navigation, and user/task modeling. Prior to coming to Microsoft, she worked on a statistical method for concept-based retrieval known as Latent Semantic Indexing. Pointers to this work are on the Bellcore (now Telcordia) LSI page ( Dr. Dumais is, among other honors, a Fellow of the ACM, and the recipient of the NJASIST Distinguished Lectureship Award.

Luanne Freund (Panel Participant), Assistant Professor, School of Library, Archival and Information Studies, University of British Columbia, Vancouver, BC, Canada

Dr. Luanne Freund is an Assistant Professor in the School of Library, Archival and Information Studies at UBC. Her areas of research are human information interaction in digital environments; pragmatic and task-based approaches to information searching; digital document genres; and evaluation of interactive information retrieval. Her dissertation research, funded by the IBM Centre for Advanced Studies, focused on the relationship between tasks and document genres in workplace information retrieval. Current work focuses on access and use of digital government information and electronic reading in academic contexts. Further information is available at:

Susan Gauch (Panel Participant), Rodger S. Kline Leadership Chair, Professor and Head of the Department of Computer Science and Computer Engineering, University of Arkansas, USA

Dr. Gauch's primary research field is Intelligent Information Agents. She received her Ph.D. from University of North Carolina – Chapel Hill in 1990 where she developed an expert search assistant for an online full-text database. While a Senior Research Scientist with the Biological Knowledge Laboratory at Northeastern University, she explored the storage, retrieval, and user interface technologies necessary to present and navigate databases of technical literature. While at the University of Kansas, from 1993–2007, her research encompassed intelligent agents for information discovery and fusion from the World Wide Web (ProFusion), content-based searching of digital video libraries, a National Science Foundation-sponsored project on the application of corpus linguistics to the field of information retrieval, and another National Science Foundation project focusing on the use of ontologies of coordinating distributed information agents. In 2007, Dr. Gauch joined the University of Arkansas to become the Head of Computer Science and Computer Engineering. One currently funded project is investigating conceptual and personalized information retrieval within the context of the Citeseer archive of computer science literature. Another NSF project is exploring the use of statistical techniques to semi-automatically create an ontology for amphibian morphology.