7
Guest editorial Issues of context in information retrieval (IR): an introduction to the special issue Abstract The subject of context has received a great deal of attention in the information retrieval (IR) literature over the past decade, primarily in studies of information seeking and IR interactions. Recently, attention to context in IR has expanded to address new problems in new environments. In this paper we outline five overlapping dimensions of context which we believe to be important constituent elements and we discuss how they are related to different issues in IR research. The papers in this special issue are summarized with respect to how they represent work that is being conducted within these dimensions of context. We con- clude with future areas of research which are needed in order to fully understand the multidimensional nature of context in IR. Ó 2002 Elsevier Science Ltd. All rights reserved. Keywords: Context in information retrieval; Interactive information retrieval 1. Introduction to the topic Investigations of different aspects of context have been central concerns in studies and theories of human information behavior, information seeking processes and information retrieval (IR) interactions for some time now, as theorists and researchers have moved away from decontex- tualized views of IR toward more use-centered and cognitive viewpoints. At this point, it is generally recognized that IR is an inherently interactive process, which occurs within multiple, overlapping, contexts that inform, direct or shape the nature of this interaction. In other words, information seeking, use and evaluation take place within multidimensional contexts, which can be analyzed from multiple levels. One can see these developments in the theoretical models of Saracevic (1997), Belkin (1996), Ingwersen (1996) and others. All of these models of IR make ‘‘context’’, sometimes interchangeably used with the concept of ‘‘situation’’, a central variable or level of analysis in the IR process. The contexts within which a person seeks information consist of cognitive, social and other factors related to a person’s tasks, goals and intentions, which precipitate the information seeking episode(s). At another level of analysis, understanding context within the IR interaction itself is important. It is within the dialogue between user and system that elicitation of user context models is revealed, and queries disambiguated. This dimension of context can relate to user Information Processing and Management 38 (2002) 605–611 www.elsevier.com/locate/infoproman 0306-4573/02/$ - see front matter Ó 2002 Elsevier Science Ltd. All rights reserved. PII:S0306-4573(01)00054-1

Issues of context in information retrieval (IR): an introduction to the special issue

Embed Size (px)

Citation preview

Page 1: Issues of context in information retrieval (IR): an introduction to the special issue

Guest editorial

Issues of context in information retrieval (IR): an introductionto the special issue

Abstract

The subject of context has received a great deal of attention in the information retrieval (IR) literatureover the past decade, primarily in studies of information seeking and IR interactions. Recently, attention tocontext in IR has expanded to address new problems in new environments. In this paper we outline fiveoverlapping dimensions of context which we believe to be important constituent elements and we discusshow they are related to different issues in IR research. The papers in this special issue are summarized withrespect to how they represent work that is being conducted within these dimensions of context. We con-clude with future areas of research which are needed in order to fully understand the multidimensionalnature of context in IR. � 2002 Elsevier Science Ltd. All rights reserved.

Keywords: Context in information retrieval; Interactive information retrieval

1. Introduction to the topic

Investigations of different aspects of context have been central concerns in studies and theoriesof human information behavior, information seeking processes and information retrieval (IR)interactions for some time now, as theorists and researchers have moved away from decontex-tualized views of IR toward more use-centered and cognitive viewpoints. At this point, it isgenerally recognized that IR is an inherently interactive process, which occurs within multiple,overlapping, contexts that inform, direct or shape the nature of this interaction. In other words,information seeking, use and evaluation take place within multidimensional contexts, which canbe analyzed from multiple levels. One can see these developments in the theoretical models ofSaracevic (1997), Belkin (1996), Ingwersen (1996) and others. All of these models of IR make‘‘context’’, sometimes interchangeably used with the concept of ‘‘situation’’, a central variable orlevel of analysis in the IR process.

The contexts within which a person seeks information consist of cognitive, social and otherfactors related to a person’s tasks, goals and intentions, which precipitate the information seekingepisode(s). At another level of analysis, understanding context within the IR interaction itself isimportant. It is within the dialogue between user and system that elicitation of user contextmodels is revealed, and queries disambiguated. This dimension of context can relate to user

Information Processing and Management 38 (2002) 605–611www.elsevier.com/locate/infoproman

0306-4573/02/$ - see front matter � 2002 Elsevier Science Ltd. All rights reserved.

PII: S0306-4573(01)00054-1

Page 2: Issues of context in information retrieval (IR): an introduction to the special issue

models of their tasks and goals, the relevance judgment behaviors related to them, or, at a moremicro-level, the linguistic analysis of properties of queries themselves.

Despite this growing attention to issues of ‘‘context’’ in IR, there exists no single definition ofwhat the concept entails, or what the meaningful constituent elements of context are which haveimportant relationships to IR processes. In recent writing, Dervin (1997) labeled context as an‘‘unruly beast’’ because of the difficulties involved in gaining methodological control over it.Vakkari (1997, p. 460) addressing the first International Conference on Information Seeking inContext (ISIC) noted that quite often central concepts such as ‘‘situation’’ and ‘‘context’’ were‘‘most commonly used without taking much trouble in seeking their meaning.’’ Conceptual claritybetween the concepts of ‘‘situation’’ and ‘‘context’’ are often especially murky (Cool, 2001).Sonnenwald (1999) has taken a stab at disambiguating these twin concepts by characterizingcontexts as embedded within situations. Allen (1997) further distinguishes between situations andcontexts by describing contexts as the larger, ‘‘socially defined’’, settings such as work and taskenvironments, within which different situations take place. Despite the ill-defined nature of‘‘context’’ ongoing research has attempted to further specify its meaningful dimensions and theirdynamics in the IR process.

The goal of this special issue is to present theoretical and empirical work which addresses someaspect, or level, of context understanding in IR. In order to better frame this work, we can firstidentify what seem to us to be the salient levels of context for understanding issues in IR. Here wedescribe four overlapping and related levels of context that are relevant to IR. For each of theselevels we give a brief description, with examples of work that has been done in this area. It isbeyond the scope of this introduction to completely review all of the literature that might berelevant to these levels.

1.1. Information environment level

Context can be construed as the information environment within which information behaviorstake place. Some concrete examples might be institutional, organizational or work task settings.The well-known work of Taylor (1991) on ‘‘information use environments’’ represents an earlystatement about this level of context. Research within this contextual level explores the social andenvironmental factors that influence human information behaviors, including information seekingand IR interactions At this level of analysis, information environments have also been looked atin terms of channels of information and communication (Sonnenwald & Pierce, 2000; Spink &Cole, 2001).

1.2. Information seeking level

The issue of context in IR has been investigated most thoroughly at this level of analysis. Theinformation seeking context level includes the goal(s) that a person is trying to achieve, or someproblem resolution task that influences the IR interaction level. Early models which have ap-proached IR from this cognitive level of context have focused on the nature of tasks, goals andintentions as they relate to a person’s problematic situation (Schutz & Luckmann, 1973), oranomalous state of knowledge (Belkin, 1980). This level is more directly related to IR. Here we canaddress concerns such as where in the search process a person is with respect to the problem at

606 Guest editorial / Information Processing and Management 38 (2002) 605–611

Page 3: Issues of context in information retrieval (IR): an introduction to the special issue

hand, and the information seeking behaviors associated with this (Wilson, 1999). This also in-cludes information use or avoidance, with a variety of information resources. In other words,given some problematic situation, or task/goal environment, we look at context in terms of theinformation seeking behaviors people engage in.

1.3. IR interaction level

More recently, context has been examined in terms of the interactive space itself. The IR in-teraction level of context explores the user–system interaction within search sessions. The earlywork of Suchman (1987) on situated action theory as an explanation for information interactionbehavior and the recent investigation by Xie (2000) of the ‘‘planned vs. situated action’’ frame-work for understanding IR interactions are relevant works at this level of analysis. The IR in-teraction level investigates the role of feedback (Spink & Losee, 1996), relevance judgments asrelated to characteristics of users’ situations, as identified early on by Schamber, Eisenberg, andNilan (1990), multidimensional search strategies (Belkin, Marchetti, & Cool, 1993) and otherspecific search behaviors that occur in the context of IR interaction during information seekingprocesses. At this level of analysis, questions have recently been asked about successive or mul-tiple searching behaviors and how they represent significantly different IR contexts (Lin & Belkin,2000; Spink, Bateman, & Greisdorf, 1999).

1.4. Query level

This linguistic level of context explores IR system performance on user queries, including querydisambiguation as terms in the query are interpreted correctly or incorrectly by the system ac-cording to the context assigned to them. A classic problem in IR is the retrieval of documents thathave the right terms, but in the wrong context, and this remains a central research concern that isincreasingly important in the Web environment, where user queries are typically short and usermodels are insufficiently generated to disambiguate them. Lawrence (2000) provides a nice in-troduction and overview to these problems of assessing context in Web queries. A number oftechniques are being experimented with in order to address IR problems at this linguistic level ofanalysis, such as the statistical language modeling technique of Ponte and Croft (1998), localcontext analysis (Xu & Croft, 1996), and some of the term suggestion devices reported in therelevance feedback studies conducted by Belkin et al. (2000). A remaining problem to be ad-dressed at this level is the development of appropriate elicitation techniques to disambiguate userqueries through better representations of their context.

2. Papers in this issue

Each of the papers in this special issue provides some insight into issues associated with levels ofcontext in IR as discussed above. Ng renews the debate over ‘‘situated’’ and ‘‘planned’’ actionmodels of behavior in IR, bringing in a fresh perspective from phenomenology. Contributingto our better understanding of both information seeking and IR interaction levels of context

Guest editorial / Information Processing and Management 38 (2002) 605–611 607

Page 4: Issues of context in information retrieval (IR): an introduction to the special issue

analysis, this paper reports on the results of an empirical study of planned vs. situated action inone specific IR task. Using graduate students as subjects, Ng looks at the strength of a priori plansas opposed to what he terms ‘‘plan deviation’’ over the course of changes in a problem solvingsituation. Results of this research lead to a rethinking of the situated action model of IR inter-action that is widely in play, and confirm that both plans and situations are important constituentelements of context in IR.

The interaction between the information environment and IR interaction levels of context isdemonstrated in the paper by Wolfram and Xie. In this work, the authors take a look at differenttypes of information use environments, specifically comparing traditional IR systems and whatthey term ‘‘general audience digital libraries’’ (GADLs) and the information interactions asso-ciated with them. They argue that there is a lack of empirical information on the subject of howusers interact with the Web and Web-based services such as the GADL. The particular digitallibrary they investigate is found to contain a dual context for users, in which some traditional IRmechanisms and functionalities are embedded within a framework of convenience of access andease of use that is familiar to users from Web IR. Their findings demonstrate the influence of thisdual context on searching behaviors and the authors conclude with recommendations for designand evaluation of GADL services from this perspective.

Bilail and Kirby present findings that further our understanding of context at the IR interactionlevel. They explore some of the cognitive, affective and physical behaviors demonstrated by juniorhigh school and graduate school students as they perform searching tasks using the Yahooligans!Web search engine. The structure of the search engine itself, in particular the reliance on keywordsearching, along with developmental differences between younger and older students were foundto be important elements of context within this experiment. In this study, searchers’ ability tounderstand the functional structure of Yahooligans! can be seen as context understanding.Context understanding and misunderstanding were important elements in the ability of studentsto recover from ‘‘breakdowns’’ in the search process. The ability to recover from ‘‘breakdowns’’,or to regain context understanding, was associated with greater retrieval performance in the factfinding task assigned to the children. Recovery of context understanding was found to be greatestamong older students.

Also exploring context at the level of IR interaction, combined with that of informationseeking, Quiroga and Mostafa investigate the role of context in relevance feedback assessments,and how the identification of important contextual elements can be used to inform the design ofmore effective information filtering systems. They report results of an experiment in whichgraduate students interacted with the SIFTER filtering system, under three different conditions ofprofile acquisition. Each condition captured different amounts of contextual knowledge about theuser. Analysis of the data showed that relevance feedback alone was less effective in terms offiltering performance as a profile-generating mechanism than the condition in which users coulddirectly specify the profile. The additional information specified by the subjects in this studycontained situational and contextual information that was not captured by relevance feedbackmechanisms alone, leading the authors to conclude with suggestions for ways in which to captureadditional dimensions of context in user profiles and how to include them in document repre-sentation and relevance feedback mechanisms.

Working directly at the contextual level of IR interaction, Choi and Rasmussen investigatecriteria affecting users’ relevance judgments of images in an American history database. Building

608 Guest editorial / Information Processing and Management 38 (2002) 605–611

Page 5: Issues of context in information retrieval (IR): an introduction to the special issue

upon earlier seminal work on situational and contextual aspects of relevance judgments, theymove forward with a study designed to investigate important criteria used in the context ofimage retrieval. An important finding from this study is that while topicality was found to be animportant criterion in relevance judgment of images, other factors directly related to this newIR environment, such as image quality and clarity, were important. Results of the studyare then discussed in terms of how to create better IR contexts within which images may beretrieved.

The paper by He, Goker and Harper is a representation of work at the query level ofanalysis. Web searching is generally characterized by short queries, with relatively little con-textual information about users, making effective information retrieval difficult. In this paper,the authors focus explicitly on how contextual information from users can be used to improveIR performance at the system level. This paper confines itself to an analysis of Web searches,and raises issues of context in IR within that environment. Context here is automaticallygenerated on the basis of analysis of users’ search logs over the course of a search session. Thisstatistical treatment of context is representative of the recent work that is focusing on automaticmethods for query disambiguation as a way of addressing issues of query level context. Theauthors address the problem of how to automatically increase the amount of contextual in-formation about users in order to improve retrieval performance by using two principle sourcesof evidence obtained from the log sessions: time interval and search pattern data, which are usedto infer contextual information about users. Results of this experiment indicate that theseprovide good sources from which to identify and analyze user contexts, resulting in better re-trieval performance.

3. Summary

The papers in this special issue address different levels of context, some of them multiple andoverlapping. In previous discussions of this topic – ‘‘context in IR’’ – emphasis has been primarilyon information seeking processes and behaviors, although the theoretical literature in this areahas focused on the multidimensional nature of context which is important to understand atmultiple points in interactive IR. Here, we have taken a step at further identifying multiple levelsof context and presenting current work in these directions.

The future for advancing research addressing issues of context in IR lies with increasingdepth and integration. First, we need a deeper understanding of the important contextual el-ements that encompass IR processes and generalizeable process models across situations. Sec-ond, we need further integration of IR and information seeking models within the broaderhuman information behavior context. Helping users resolve their information problems hasalways been a goal of IR research. Integrative models will work to help people develop theirinformation seeking behaviors into effective searches, involving query formulation and refor-mulation, and to effectively formulate relevance judgments. A relatively new area of researchwith respect to context focuses on deriving contextual information from user queries themselves.Work in this area has been conducted primarily in the Web environment, but it has generalapplication to interactive IR more generally and we can expect more work in this area in futurestudies.

Guest editorial / Information Processing and Management 38 (2002) 605–611 609

Page 6: Issues of context in information retrieval (IR): an introduction to the special issue

References

Allen, B. (1997). Information needs: a person–situation approach. In P. Vakkari, R. Savolainen, & B. Dervin (Eds.),

Information seeking in context: Proceedings of an international conference on research in information needs, seeking

and use in different contexts (pp. 111–122). London: Taylor Graham.

Belkin, N. J. (1980). Anomalous states of knowledge as a basis for information retrieval. Canadian Journal of

Information Science, 5, 133–143.

Belkin, N. J. (1996). Intelligent information retrieval: whose intelligence?. In J. Krause, M. Herfurth, & J. Marx (Eds.),

Herausforderungen an die Informationswirtschaft. Informationsverdichtung, Informationsbewertung und Datenvisual-

isierung. Proceedings of the Fifth International Symposium for Information Science (ISI ’96) (pp. 25–31). Konstanz:

Universit€aatsverlag Konstanz.

Belkin, N. J., Cool, C., Head, J., Jeng, J., Kelly, D., Lin, S. J., Park, S. Y., Rieh, S. Y., Savage-Knepshield, P., & Sikora,

C. (2000). Relevance feedback versus local context analysis as term suggestion devices. Rutgers’ TREC-8 Interactive

Track experience. In E. M. Voorhees, & D. Harman (Eds.), The eighth Text REtrieval Conference (TREC-8) (pp.

565–573). Washington, DC: GPO.

Belkin, N. J., Marchetti, P.-G., & Cool, C. (1993). BRAQUE: design of an interface to support user interaction in

information retrieval. Information Processing and Management, 29, 325–344.

Cool, C. (2001). The concept of situation in information science. In M. E. Williams (Ed.), Annual review of information

science and technology (Vol. 35) (pp. 5–42). Medford, NJ: Learned Information.

Dervin, B. (1997). Given a context by any other name: methodological tools for taming the unruly beast. In P. Vakkari,

R. Savolainen, & B. Dervin (Eds.), Information seeking in context: Proceedings of an international conference on

research in information needs, seeking and use in different contexts (pp. 13–38). London: Taylor Graham.

Ingwersen, P. (1996). Cognitive perspectives of information retrieval interaction: elements of a cognitive IR theory.

Journal of Documentation, 52(1), 3–50.

Lawrence, S. (2000). Context in web searching. Data Engineering, 23(3), 25–32.

Lin, S. J., & Belkin, N. J. (2000). Modeling multiple information seeking episodes. In D. H. Kraft (Ed.), Proceedings of

the 63rd annual meeting of the American Society for Information Science (pp. 133–147). Medford, NJ: Information

Today.

Ponte, J., & Croft, W. B. (1998). A language modeling approach to information retrieval. In W. B. Croft, A. Moffat,

C. J. van Rijsbergen, R. Wilkinson, & J. Zobel (Eds.), SIGIR ’98. Proceedings of the twenty-first annual

international ACM SIGIR conference on research and development in information retrieval (pp. 275–281). New York:

ACM.

Saracevic, T. (1997). The stratified model of information retrieval interaction: extension and applications. In C.

Schwartz, & M. E. Rorvig (Eds.), Proceedings of 60th annual meeting of the American Society for Information

Science. Medford, NJ: Learned Information.

Schamber, L., Eisenberg, M. B., & Nilan, M. (1990). A re-examination of relevance toward a dynamic situational

definition. Information Processing and Management, 26(6), 755–776.

Schutz, A., & Luckmann, T. (1973). Structures of the life world. Evanston, IL: Northwestern University Press.

Sonnenwald, D. H. (1999). Evolving perspectives of human behavior: contexts, situation, social networks and

information horizons. In T. Wilson, & D. K. Allen (Eds.), Exploring the contexts of information behaviour:

Proceedings of the 2nd international conference on research in information needs, seeking and use in different contexts

(pp. 176–190). London: Taylor Graham.

Sonnenwald, D. H., & Pierce, L. (2000). Information behavior in dynamic group work contexts: interwoven situational

awareness, dense social networks and contested collaboration in command and control. Information Processing and

Management, 36(3), 461–479.

Spink, A., Bateman, J., & Greisdorf, H. (1999). Successive searching behavior during information seeking: an

exploratory study. Journal of Information Science, 25(6), 439–449.

Spink, A., & Cole, C. (2001). Information and poverty: information seeking channels used by African–American low-

income households. Library and Information Science Research, 23, 45–65.

Spink, A., & Losee, R. (1996). Feedback in information retrieval. In M. E. Williams (Ed.), Annual review of information

science and technology (Vol. 31) (pp. 33–78).

610 Guest editorial / Information Processing and Management 38 (2002) 605–611

Page 7: Issues of context in information retrieval (IR): an introduction to the special issue

Suchman, L. (1987). Plans and situated actions: The problem of human machine communication. Cambridge: Cambridge

University Press.

Taylor, R. S. (1991). Information use environments. In B. Dervin, & M. J. Voight (Eds.), Progress in communication

sciences (Vol. 1) (pp. 217–255). Norwood, NJ: Ablex.

Vakkari, P. (1997). Information seeking in context: a challenging metatheory. In P. Vakkari, R. Savolainen, & B.

Dervin (Eds.), Information seeking in context: Proceedings of an international conference on research in information

needs, seeking and use in different contexts (pp. 451–464). London: Taylor Graham.

Wilson, T. D. (1999). Exploring models of information behavior: the uncertainty project. In T. Wilson, & D. K. Allen

(Eds.), Exploring the contexts of information behaviour: Proceedings of the 2nd international conference on research

in information needs, seeking and use in different contexts (pp. 55–66). London: Taylor Graham.

Xie, H. (2000). Shifts of interactive intentions and information-seeking strategies in interactive information retrieval.

Journal of the American Society for Information Science, 51(9), 841–857.

Xu, J., & Croft, W. B. (1996). Query expansion using local and global document analysis. In H.-P. Frei, D. Harman, P.

Sch€aauble, & R. Wilkinson (Eds.), SIGIR ’96, Proceedings of the 19th annual international ACM SIGIR conference

on research and development in information retrieval (pp. 4–11). New York: ACM.

Colleen CoolGraduate School of Library and Information Studies

Queens College City University of New York 65-30 Kissena BlvdFlushing, NY 11367

USAE-mail address: [email protected]

Amanda SpinkSchool of Information Sciences and Technology

The Pennsylvania State University, 004C Thomas BuildingUniversity Park, PA 16802

USAE-mail address: [email protected]

Guest editorial / Information Processing and Management 38 (2002) 605–611 611