Browse Books

Reviewer: Dagobert Soergel

Information seeking and retrieval (IS&R) is a very complex phenomenon; any one component cannot be understood without considering the other components it interacts with. It is suboptimal to build and test systems without understanding users' tasks and resulting information needs, as well as the cognitive structures held by users and the structure of documents and how they match. This is the important message addressed in this book. The authors are perhaps the first to take an integrated view of information seeking and information retrieval in an overall framework. Their message needs to be heard, but, unfortunately at times, it gets buried in wordy prose. The message is most important for those who may be least likely to read such a book-systems-oriented computer science information retrieval (IR) researchers and developers of IR systems, who often work in a vacuum with respect to real-life users; this audience needs a short, crisp version. This is a programmatic book, but not a textbook, even though parts of it will be useful as course readings. The book's major contribution is a conceptual framework for IS&R research (which might have made a better subtitle). It discusses concepts and tools for implementing an integrative approach to IS&R research and development. The conceptual framework presented consists of nine dimensions in five groups: organizational task, actor, document, algorithms, and access and interaction. Each dimension has many variables; for example, "complexity of a work task," "user knowledge of domain and search system," "retrieval system designer knowledge of user tasks," "nature and degree of structure of documents," "use of domain ontology in a retrieval system," "visualization used in the interface," and "how a user interacts with an interface." These variables are elaborated throughout the book, and their possible interrelations are shown in many diagrams. The book would be a good starting point for a systematic repository of such variables that researchers could refer to when planning a study. For many readers, it may be best to start with chapter 8, "Towards a Research Program," which lays out several good study designs using varying combinations of control, independent, and dependent variables. For example, it suggests the development of test collections geared toward specific types of work tasks so one can study the suitability of retrieval systems to support different kinds of work tasks, thus incorporating a context that is now sorely missing in retrieval testing programs like TREC. Knowing the final destination may make it easier to understand each step in the journey. The book is organized into three parts. Part 1 consists of chapter 1, "Introduction," and chapter 2, "The Cognitive Framework for Information." Part 2, chapters 3 through 5, presents a selective review of work in information-seeking research, system-oriented information retrieval, and cognitive and user-oriented information retrieval. Part 3 develops the authors' integrated IS&R framework (chapter 6), and elaborates many variables (chapter 7) and sample research designs (chapter 8). Chapter 9 provides a very short concluding high-level summary of principles. Chapter 1 contains a useful discussion of the role of conceptual models in research. Chapter 2 discusses many different approaches to defining information and models of cognition. Much of this is rather dense, often embellishing the obvious and not necessarily leaving the reader any wiser. For example, the five statements summarizing the central tenets (in the book, misleadingly called dimensions) of the cognitive view are all obvious. An important reference is missing [1], which links information to decision making under uncertainty. The three chapters of Part 2 provide selective overviews of work in each of the three areas, covering a cross section of typical papers. Each of the chapters is somewhat uneven. There are good informative reports on some important topics covered in the literature, mixed with merely indicative abstracts; the range of the topics discussed is so wide that it is not possible to do justice to all or even most of them. All chapters have a section on research methods with very brief descriptions that together give a somewhat eclectic overview of the wide range of methods used in IS&R research, with pointers to studies using each method. One would wish for more in-depth analysis in these chapters. There is more direct summarization of individual studies than in-depth mining of the relationships and trends emerging across the studies. These chapters lack a good framework to organize all of the literature they discuss (this is left to Part 3). In addition, the relationships between individual studies are not analyzed, and findings are not integrated. As a result, the major research questions, problems, and trends across different studies sometimes remain unclear, leaving Part 2 somewhat disjointed. Many of the claims made and trends and research issues identified do not seem convincing; they seem imposed rather than naturally derived from the studies discussed. Surprisingly, the discussion of the cognitive perspective is quite limited. Most of the discussion centers on combining different cognitive representations (polyrepresentation) in information seeking and retrieval, but does not go beyond that. The authors try to impose a cognitive perspective on interpreting the user-centered studies in the field, but the connections they make seem shallow. What the real value or the real contribution of the cognitive perspective is or how it really differs from the user-centered view is not as well elaborated as it could be. Part 3 is intended to give the authors' grand integration. It has a number of useful diagrams and tables showing components of the larger IS&R system, their relationships, and resulting variables (for example, Figure 6.8 on page 274). Chapter 6, "The Integrated IS&R Research Framework," discusses the different actors, their functions in the IS&R system, and, more importantly, the contexts they work in, the work tasks they are engaged in, and the resulting search tasks they must pursue. In doing so, the chapter identifies variables and issues. While each process or component relationship is discussed as it is situated in the overall system (as emphasized by using the same basic IS&R system diagram with different parts highlighted), the overall impression is still one of a series of specific discussions, rather than an integrated whole. Chapter 7, "Implications of the Cognitive Framework for IS&R," elaborates on this framework, considering design and evaluation, and associated research, identifying many variables for IS&R system components and their relationships. This book suffers from an issue that is all too common in this field: giving credit for "firsts," when the real "first" was actually much earlier. For example, Wilson [2] (in the bibliography) and Cuadra and Katter [3] (not in the bibliography) are not cited where they should be (pages 1 and 29). McCarn and Stein's study [4] (not in the bibliography) was one of the very few that looked at the contributions of one information system to the informedness of users who use many other systems as well; this idea is not "recent" (page 1). That personal information needs are dynamic rather than static has been articulated since the 1960s, and is not new to the 1990s (page 3). The first to introduce logical deduction to IR was not Van Rijsbergen [5], but (as acknowledged by Van Rijsbergen) Cooper [6] (in the bibliography, but not cited on page 118). The idea of task-based retrieval, perhaps not under that name and perhaps not studied with the same explicitness, was around in the field long before Järvelin [7] (page 282); the textbook by Soergel [8] builds on earlier work, including Mooers and Brenner [9], in putting a major focus on supporting tasks as the purpose of retrieval. The discussion of "the dimensionality of relevance" misses Cuadra and Katter [3], which well predates the 1990 article by Schamber and others [10] (page 235). In the discussion on the indexer "investigating how the 'utility community' understands and uses objects," the seminal paper by Mooers and Brenner [9] should be cited before Hutchins [11]. Other important references are missing; for example, Saracevic and Rees [12] and Saracevic [13], which describe information retrieval evaluation procedures (with a large list of variables) far closer to the reasonable approaches the book suggests than the Cranfield-TREC approach that the book rightly criticizes. The book focuses on information seeking and the retrieval of documents and document-like objects by individual actors searching single systems. This is the problem of information retrieval: We now must think much more broadly about systems that integrate support for all user task (work task) functions, from information seeking to information processing and application, to information production (vertical integration, which the book faintly hints at), with support for sharing and collaboration in computer-supported cooperative work environments (horizontal integration) to provide a seamless environment in which users-individuals and teams-can accomplish all information seeking, information processing, and communication tasks. The ultimate goal is a system that supports all functions of individual users and teams and all functions and workflows in an organization. This includes accessing multiple information systems for the same information need (the problem of interoperability, especially semantic interoperability) and the chaining or combination of information from different sources. It also includes a host of information processing functions, including document (text, images, and video) summarization, question answering, and information extraction. The methods and conceptual frameworks presented in the book should be extended to this broader realm. Online Computing Reviews Service