1 Introduction

The immediacy of information, availability of communication channels, abundance and diversity of tasks that people routinely accomplish online, and effects of social networks shape the expectations of users when they are looking for scholarly materials. Surveys and reports published since 2005 (reviewed in Sadeh 2011) clearly demonstrate that traditional library systems are lagging in popularity behind non-library information systems. These surveys and reports describe a reality in which library users have shifted from searching and accessing content via library services to attempting to satisfy their information needs through non-library services, such as web search engines, online bookstores, blogs, online news, and e-mail. Statistical data support these trends. For example, the Association of Research Libraries (ARL), whose membership spans North America, reports that reference transactions at ARL institutions decreased by 65 % and circulation transactions decreased by 29 % between 1991 and 2011, despite an increase of about 33 % in the number of students enrolled in those institutions during that period^[2].

In 2007, Markey pointed out that libraries’ “failure to respond accordingly may permanently exile scholarly and scientific information to a netherworld where no one searches while less reliable, accurate, and objective sources of information thrive in a paradise where people prefer to search for information” (Markey 2007). Such concerns have driven librarians to think of ways to retain their users and maintain their leadership as academic information providers. However, to address users’ expectations regarding the search interface, the breadth and relevance of services, and the comprehensiveness of the body of information available through a library system, libraries have had to undergo a major conceptual shift.^[3]

The rise of “next-generation catalogs”, which were introduced in January 2007 with the Endeca-based system at North Carolina State University, marked a significant change in the way in which libraries engage with their users. Initially aiming to provide a modern, friendly user experience for library patrons, these systems, termed discovery systems or discovery-and-delivery systems, transformed the way in which users search in library catalogs and other local library collections such as institutional repositories, course materials, and even library web pages. However, only in 2009, when discovery systems shifted gear with the introduction of a central index,^[4] could these systems serve as a single point of access to the entire library collection.

Discovery services that include a central index – such as Ex Libris Primo^[5], Serial Solutions Summon^[6], EBSCO Discovery Service^[7], and OCLC WorldCat Local^[8] – have been adopted at a rapid pace, and thousands of institutions are making their local and global collections available to their users through such systems (Breeding 2007). In a more recent update, from 2013, Breeding states that “discovery services continue to represent a major component of the industry. Web-scale, or index-based, discovery services now are must-have products for libraries with large collections of electronic resources” (Breeding 2013). Nevertheless, not all librarians endorse these systems.

2 What do we mean by “discovery”, as opposed to “search”?

In this paper, traditional library information systems, such as library catalogs and databases, are referred to as search systems. Search systems offer structured search interfaces that are tailored to the specific data that they hold; the records are homogenous – they are cataloged in the same way, have the same data structure, and often relate to one topic (as in a subject-related database). Search systems typically expect users to possess medium-to-high searching literacy and enable users to accurately define their information need. Although in recent years such systems have also simplified their search interface and incorporated post-search refinement options, librarians encourage users to develop searching strategies and take advantage of the rich options of the systems’ search interface.

Library discovery systems, despite some differences between them, share several major characteristics, described in the following sections.

3 Why has the transition been so difficult?

Despite the rapid adoption of discovery systems by libraries – and the endorsement of these systems by library users – some librarians remain skeptical. At conferences where discovery systems are discussed, the implementation teams of such systems refer to the reluctance of some librarians, primarily reference librarians, to embrace the new system as an optimal tool for the majority of users searching for scholarly materials.^[13]

First, some librarians challenge the Google approach that discovery systems have adopted, which does not require users to clearly formulate their information need. With the Google approach, users employ trial-and-error methods and trust the system’s ranking algorithms to present the best results. The users assume that they will be able to identify the desired items. However, librarians attribute great importance to the mental query process, believing that the systematic articulation of an information need is a scholarly asset that must be learned and exercised. Whether or not an item is found is less important than the building of a search strategy.

Second, the information landscape indexed by a discovery system is heterogeneous, covering huge amounts of data of various types and from various sources. Consequently, the metadata is not uniform; groups of records may be subject to different cataloging rules, may adhere to different topics or subject-heading norms, and may vary in the amount of available metadata and data. The lack of a common denominator in the structure of the data and the cataloging rules and the varying quality of the data may render the user experience inconsistent. For example, when a user refines a result list by a specific topic, only those items that were assigned that topic will appear on the refined list, although it could well be that related materials express the topic through a slightly different wording or have not been assigned any topic at all. Indeed, discovery system developers continue to face the challenge of coherently and uniformly processing a wide range of content types.

Furthermore, at the time of this writing, there is not one discovery system that indeed covers the full spectrum of scholarly resources, even though all such systems cover the great majority of relevant resources. While most content providers have fully embraced discovery systems and willingly contribute their data for indexing in order to broaden access to their content, a few notable providers, including a number of abstracting and indexing (A&I) database providers, are not yet participating. The recommendations of the NISO Open Discovery Initiative (ODI)^[14], now in its second phase, should help remove outstanding barriers to participation.

This reality makes librarians doubt users’ ability to find the items that are likely to best satisfy the users’ information needs. Librarians claim that users are happy to settle for a “good enough” result set and may not notice, or care, if they miss better results. Such concerns arise, at least partly, from the fact that by breaking the traditional silos of information, discovery systems require that librarians invest more time and effort in understanding the exact coverage available to their users and ascertaining ways of making their users aware of the exact coverage. While the content available through a discovery system is much better defined than that available through a web search engine, the coverage is still not as clear as that of traditional databases.

There is no doubt that the introduction of library discovery systems is a significant change for libraries. However, changes in human behavior have been occurring since the beginning of civilization, helping our society move forward. Often, such changes shed common beliefs, traditions, and practices. In many cases, people embrace change easily; yet at times, they have difficulty adjusting and feel that the gain is not worth the loss. Is the reluctance to adopt discovery systems as a feasible replacement for structured, controlled, and topic-related search systems justified? Or should librarians embrace the change, understand its benefits, and help develop it further?

To better address this question, let us look at models of human information seeking and see how well discovery systems support human behavior.

4 Modeling human information-seeking behavior

In this paper, information seeking is defined as information behavior that consists of an active pursuit of information by scholars through the use of information systems. This definition acknowledges the reliance of searchers on finding information that they do not explicitly specify but that is nevertheless vital to their work tasks. Information-searching behavior is the aspect of information-seeking behavior that deals especially with active, directed searching in information systems for data that can be specified to some degree (Wilson 1999) (Figure 1).

Figure 1:

Wilson’s model: from information behavior to information searching. Modified from Wilson (1999)

According to Wilson, information-searching behavior “is the ‘micro-level’ of behavior employed by the searcher in interacting with information systems of all kinds. It consists of all the interactions with the system, whether at the level of human computer interaction (for example, use of the mouse and clicks on links) or at the intellectual level (for example, adopting a Boolean search strategy or determining the criteria for deciding which of two books selected from adjacent places on a library shelf is most useful), which will also involve mental acts, such as judging the relevance of data or information retrieved” (Wilson 2000, 49).

Several models that shed light on the way people look for information have influenced much of the research in the area of information seeking and searching. In 1989, long before information systems were readily available to the general public, Bates suggested the berrypicking model (Figure 2). This model, which applies to a range of search techniques, offers a new perspective on the nature of the query and the search process and embeds the search process in an information territory.

Bates explains:

“Users may begin with just one feature of a broader topic, or just one relevant reference, and move through a variety of sources. Each new piece of information they encounter gives them new ideas and directions to follow and, consequently, a new conception of the query. At each stage they are not just modifying the search terms used in order to get a better match for a single query. Rather the query itself (as well as the search terms used) is continually shifting, in part or whole.” (Bates 1989)

Bates also emphasizes that the search is taking place in the context of a “universe of interest,” which is within the larger context of the “universe of knowledge” (Bates 1989). The behavior of the searcher is the focus of attention in this model, and the “continuity represented by the line of the arrow is the continuity of a single human being moving through many actions toward a general goal of a satisfactory completion of research related to an information need. The changes in direction of the arrow illustrate the changes of an evolving search as the individual follows up various leads and shifts in thinking” (Bates 1989).

Figure 2:

The Bates berrypicking model, with an evolving search. Modified from Bates (1989)

Although not necessarily applicable to computerized systems, this model fits well with scholars’ use of discovery systems. With the exception of a known-item search – a search for a specific item (which is likely to appear at the top of the result list) – the initial search often leads to an interaction with the system whereby the user either rephrases or refines the search or follows leads, navigating from one item to another. Search logs indeed demonstrate such behavior (figure 3).

Figure 3:

Two snippets of a Primo search log from May 30, 2013, showing the interaction of two searchers with the discovery system. The numbers in square brackets indicate which result is at the top of the screen (the first result in most of the examples) and the number of results on the page (10 in these examples). Note that in one example, the eleventh result is at the top of the page – the second page of results.

The interactive nature of the information-seeking process has also inspired Belkin et al. (1995), who describe this process as multiple interactions with an information system. They argue that “people’s conceptions of their information problems change through their interactions with the IR [information retrieval] system” (Belkin et al. 1995) and that each type of information need requires a different kind of interaction with the information system.

Belkin et al. propose an information-seeking behavior model that is based on “a multidimensional space of information-seeking strategies (ISSs)” (Belkin et al. 1995). Each such ISS is derived from a certain context, which consists of the person’s information-seeking goals and the knowledge that the person possesses before starting the process (e.g., the specification of the information needed). The ISSs establish interactions between the person and an information system, and several such interactions form an “episode” (Belkin et al. 1995). Through the information-seeking episode, the searcher’s knowledge and goals develop, changing the specific values of the ISSs. Four modes, or dimensions, of interaction are proposed: method of interaction, goal of interaction, mode of retrieval, and resource considered (Figure 4).

Figure 4:

Belkin et al.’s modes of interaction. Modified from Belkin et al. (1995)

“According to our conceptualization”, note Belkin et al., “information-seeking behavior is characterized by movement from one strategy to another within the course of a single information-seeking episode, as a person’s problematic situation changes” (Belkin et al. 1995).

A discovery system clearly supports most of the user interactions described through this model, and every type of interaction can be mapped along the four dimensions. Figure 5 and Figure 6 illustrate two of the most common interactions with a discovery system.

Figure 5 shows a search for a known item: the mode of interaction is searching, because the user is interested in a specific item and is not likely to scan long lists of results; hence, the mode of retrieval is specification rather than post-search recognition. The user is interested in obtaining an item, not just knowing that it exists, so the goal of interaction is selecting rather than learning, and the resource considered is the item itself, be it physical or electronic, rather than the metadata describing it.

Another example is a search for literature about a specific topic (Figure 6). In this case, the user is likely to start with a loose description of an information need and is therefore likely to receive a large result set. In the range between scanning and searching, the mode of interaction is around the middle. The goal of interaction is primarily learning, although the user may decide to select items. The mode of retrieval is mostly recognition, because the user is not referring to a specific item; and the resource considered is likely to be only the metadata (assuming it includes the abstract).

In 2002, Bates introduced another model of information seeking, presented as a matrix that embeds information searching in the broader context of information seeking (Figure 7).

Figure 5:

Search for a known item, mapped to Belkin et al.’s modes of interaction (1995)

Figure 6:

A literature search, mapped to Belkin et al.’s modes of interaction (1995)

Figure 7:

Modes of information seeking proposed by Bates. Modified from Bates (2002)

In this model, the first row (“directed” modes) refers to modes in which a person can specify, to some extent, her information needs, while the second row (“undirected” modes) refers to modes in which the person encounters information in a random manner. In one column, a person is actively looking for information, whereas in the other column, a person is just absorbing information in a passive way. According to this model, an academic searcher demonstrates all four combinations of modes: searching in an information system would be considered directed active behavior; browsing through enlisted newsletters, announcements, or personalized feeds of new submissions would be undirected active behavior; attending a conference or a workshop in one’s area of research would be directed passive behavior; and, finally, learning about new publications through haphazard conversations would be considered undirected passive information-seeking behavior.

A search system accommodates only the directed active mode. A discovery system opens up new options; for example, by suggesting articles related to one that a searcher has selected, the system brings to the attention of the searcher articles that do not necessarily share textual cues with the selected article. In this way, the system supports the undirected active mode of information seeking. Alerts defined by the user also fall into this realm. Passive information seeking, on the other hand, remains out of the scope of a discovery system.

The last model presented for this discussion represents the information-searching behavior of academic users (Figure 8) (Sadeh 2010).

Figure 8:

A model of information-seeking behavior of an academic user (reprinted from Sadeh 2010)

Information seeking as an active, directed interaction starts with a perceived information need. The searcher has an idea of the material required and converts the perceived information need to an articulated information need, or query, in order to start the interaction with the information system. A query typically represents the searcher’s hypothesis about the specific keywords that are found in the metadata or the full text of the required documents. The distinction between a perceived information need and an articulated information need is crucial to the information-searching process. A failure of the system to return the results that a searcher is looking for can be attributed to the nature of the need itself (if, for example, the information to satisfy the need does not exist), the user’s failure to describe the need in a suitable way, or other factors.

The model addresses three kinds of query modes: Explore, Search, and Ask For. The mode depends on the information need and the way it is expressed. Queries vary from being very precise – when searchers describe a specific item – to very vague – typically when a searcher is not an expert in the field and thus cannot define the information need clearly. For example, when a searcher is looking for a specific article, she is likely to enter information such as the article’s title or the names of the authors. In this case, the user’s mode of searching is Ask For, as opposed to Search or Explore, even though the technical process is similar for the three modes. In the Ask For mode of searching, if the document is well defined by the user, the result of the search is likely to be short, and a good relevance-ranking algorithm would position the article as the first item on the list.

The Explore mode can be exemplified by a novice user’s search for information on a specific topic or a more experienced user’s search for information in a field outside her area of expertise.^[15] In these cases, the searcher is likely to be unfamiliar with the most appropriate search terms or with the kind of materials that the query results might offer. Therefore, the initial result list may not satisfy the user’s information need.

Most queries, however, are somewhere between these two extremes: searchers are familiar enough with their field of interest to clearly define their information need yet do not know of a specific item that would provide the desired information. The variations of the possible queries in such cases relate to the amount of information that the searchers know about the topic or decide to provide when they articulate their perceived information need. The more information they know and provide, the closer the query comes to asking for a specific item. This mode of searching, which corresponds to what we intuitively perceive as “searching”, can be described as Search mode. The balance between formulating a query with too little of the hypothesized information – which can lead to too many results – and including too much of the hypothesized information – which can eliminate relevant results – is of great concern to most searchers and triggers a trial-and-error mode of articulating the information need.

Once a query is submitted, the system displays a result list to the searcher. Even before scanning the results, the searcher obtains valuable information that the system has provided: the number of items in the list and the system’s suggestions that relate to the query (such as Did You Mean ...?) or relate to the result list. The latter might include, for example, post-search groupings and suggestions for new searches. All this information enables the searcher to understand right away whether the result list is worth exploring in more depth. For example, a lack of results may signify one of several possibilities: that the information does not exist, that the query contains wrong terms or even an error, or that the searcher has to rethink the articulated information need. Too many results may signify that the query was too broad. A system’s Did You Mean ...? suggestion may draw the searcher’s attention to a misspelled name or a variation in a term. Topic lists and other information (date ranges, author names, journal titles, and more) that serve as post-search groupings provide a brief summary of the result list; by looking at the terms displayed in these groups, the searcher can see the major characteristics that the items on the list have in common.

Searchers typically scan the first items in a result list before taking an action. Even a brief look at these items usually provides enough clues as to whether the searcher is on the right track, especially when the list is sorted by relevance. Often the first result, or one of the first results, is the requested item – especially when the query mode is Ask For. If none of the first items seems relevant, searchers typically reevaluate their query.

After analyzing the first screen, a searcher chooses one of the following options:

1. Focus: If the result list is satisfactory and there are results that seem relevant, at least at first glance, the searcher may focus on a specific item.

2. Narrow down: If there are too many results, the searcher may choose to narrow down the list so that it shows only the items that are more relevant.

3. Reformulate: If there are no results or the results do not seem relevant, the searcher may decide to reformulate the query.

Narrowing down can be carried out in two ways: either the searcher takes advantage of the system’s options – perhaps by clicking a facet (such as a specific date range, topic, or journal name) or by choosing to see only materials available online – or the searcher decides to modify the articulated information need by providing more information.

However, if the searcher decides that the query was not adequately phrased, she needs to reformulate it. Reformulation can be minor or can be suggested by the system – when a name or term has been misspelled, for instance. However, in some cases, the searcher might need to consider the information need again, modifying the perceived information need and restarting the process.

The flow described by this model is iterative and may well span a long period of time. Sometimes users, particularly undergraduates, have a distinct, focused information need, and once the system satisfies the need, the process is terminated. However, in many other cases, especially with researchers, the interaction with the discovery system is ongoing.

This model of information searching covers only part of a user’s information-seeking behavior. Other elements, such as undirected information seeking, are beyond the scope of this paper. However, navigation from one item to another, another aspect of information seeking, deserves special attention. Navigation is a very common practice of information seeking, regardless of the type of information that a user is looking for. In the scholarly domain, researchers carry out such navigation by following a citation trail: by moving backward and forward through references, a scholar can explore the origins of an idea and trace its impact on the academic discourse. As part of an information-seeking process, references may expand the knowledge of a searcher into adjacent areas that are only briefly discussed in the document that the scholar has found. Often, it is enough for a scholar to search only once and continue the information-seeking process through such navigation.

This theoretical model is well supported by discovery systems. Although conceptually the various modes of searching – Explore, Search, and Ask For – differ from one another, a working system supports them all through the same user-interface functions. The type of search determines the kind of options that the system makes available and that are useful in a specific context. For example, the display of facets is relevant only when the searcher is looking for materials on a certain topic and there are too many results for the searcher to scan.

5 What else can discovery systems offer?

Like every new technology, library discovery systems started by increasing the efficiency of existing practices. Users were able to satisfy their information need before a discovery system was available; they had to use multiple information resources, were often challenged with complex workflows when trying to find relevant materials, but would usually succeed in obtaining the required items. From the outset, discovery systems enabled searchers to use only one information resource through a user-friendly interface and streamlined the searchers’ workflows from the initial search to the obtaining of the item. However, once this basic framework was set in place, the developers of discovery systems started introducing new capabilities that were not previously feasible.

6 Personalization

Another feature that is increasingly expected in library discovery systems (University of Minnesota Libraries, 2010), and is now starting to appear, is their ability to tailor the search results to a specific searcher’s needs. Although this feature is common in nonacademic information systems, academic systems have responded to growing concerns about privacy issues by keeping away from customizing results on the basis of a searcher’s attributes. However, with the ever-increasing number of publications that are available to users and the dismantling of subject-specific silos, a discovery system that takes into account a user’s discipline and academic level is able to address the user’s information need more accurately, especially when the user conducts an exploratory search.

7 Linking, clustering, and navigation

The availability of a very large corpus of data in a single system is one of the greatest advantages of a discovery system. The implementation of smart linking between items leverages this data and helps the system provide the searcher with usable information rather than lists of items. For example, the clustering of various aspects of a scientific work, including the research data, project reports, and published articles, is likely to provide searchers with a rich user experience and help them grasp the information in a more complete way. Furthermore, because institutional repositories are also harvested by discovery systems, the linking of various versions of a work, such as the preprint and post-print versions, becomes feasible. Another example relates to other entities, such as authors and institutions, which can be rendered into hubs of information, enabling searchers to discover other individuals and teams conducting research in their area and other departments that are dealing with similar projects.

8 Conclusions

The rapid pace of adoption of discovery systems demonstrates the appeal of such systems to institutions and searchers. Discovery systems have transformed the academic search process, making it more intuitive and interactive and integrating it with other user activities. To support the broad, heterogeneous information landscape that is available today, the current discovery process puts more emphasis on post-search activities and better supports serendipity, leading users to a richer and more diverse learning experience.

Discovery has continued to evolve with the introduction of new services that could not have been implemented in the prediscovery era. In the future, we are likely to see other innovations, such as visual representations of results and of clusters of results, the expansion of semantic search capabilities, and the creation of tools that address additional aspects of the scholarly publishing and dissemination process. Recent statistics demonstrating an increase in the usage of scholarly materials following the adoption of discovery systems (Levine-Clark et al., 2014) may ease the minds of those who are resistant to discovery systems. Capabilities that are now being added, such as new forms of navigation are also likely to contribute to a realization on the part of librarians that discovery systems serve today’s searchers in ways that were not technologically possible in the past and provide stimulating new paths for finding scholarly materials.