Reading Analysis Discussion Post

Human Information Interaction Fidel, Raya

Published by The MIT Press

Fidel, Raya. Human Information Interaction: An Ecological Approach to Information Behavior. The MIT Press, 2012. Project MUSE. muse.jhu.edu/book/21630. https://muse.jhu.edu/.

For additional information about this book

[ Access provided at 29 Jan 2022 01:05 GMT from The University of British Columbia Library ]

https://muse.jhu.edu/book/21630

5 Five Search Strategies

The concept search strategy has been part of the vocabulary of human information

behavior (HIB) since the earliest user studies. However, researchers only began to

investigate search strategies after the development of digital technology, when the

concept became a popular focus of study with the introduction of the World Wide

Web. Unlike information need , which is relatively stable, 1 search strategy addresses the

dynamic part of the search process itself. While an information need triggers a search

process, search strategies reflect the activities during the search. In addition, strategies

are considered to possess a great advantage as an object of study: While they are purely

cognitive in nature, they are observable because their use — that is, the activities during

a search — can be observed. 2 New research techniques that have been afforded by

digital technology made it possible to investigate the search process itself, and thus

its strategies.

Because the concept search strategy is relatively concrete and observable, its defini-

tion has not raised much discussion, but researchers have attributed to it a range of

interpretations and definitions and have often overlooked the need to provide their

construal even when search strategies were the focus of their studies. This chapter

briefly provides a few examples of some of these definitions and proposes a view on

search strategies that is relevant to the design of information systems.

5.1 What Is a Search Strategy?

Research into search strategies has been carried out since the late 1970s, but the inter-

pretation of the concept search strategies has been highly fluid, and even today the

concept is imbued with a plurality of meanings. HIB researchers have applied the term

to signify any aspect of an information search process that lacks its own name. Most

empirical researchers have also neglected to explain their understanding of the concept.

98 Chapter 5

In some cases, the investigators ’ construal can be inferred from the specific search

strategy they investigated. Only a few researchers provided explicit definitions for

search strategies , and some others borrowed these definitions for their own studies.

5.1.1 Implicit Construal of Search Strategy

Examples of search strategies that have been discovered in web searching without the

support of an explicit definition of the concept show that most address specific actions

in a search process and are highly concrete, mechanical, and concerned with observ-

able actions. Only a few implicit definitions enjoy some level of abstraction. Some

researchers were inconsistent in the level of abstraction of the search strategies they

investigated, identifying them along a range from highly concrete to the abstract.

The series of studies that Nigel Ford and his colleagues conducted is a typical

example of a concrete and actions-based interpretation of the concept. Ford began his

investigation of search strategies during the early period of bibliographical databases

(e.g., Ford, Wood, and Walsh 1994). Examining his research reports, it seems that he

understood search strategies to be the types of actions a searcher take to transform a

query. A recent article about the use of search strategies provided 18 strategies (Ford,

Eaglestone, and Madden 2009), such as page down , remove Boolean operators , include

quotation marks , reuse part of a query , and change operators only . Other researchers — such

as Martzoukou (2008) and Iivonen and White (2001) — recognized search strategies on

the same level of abstraction, identifying, for example, use Boolean operators or use

subject directory .

The concrete level of the search strategies ’ construal limits the range of their appli-

cability because they are to a large degree determined by the technology being used.

Search strategies that can be employed in best-match systems, 3 for instance, are dif-

ferent from those in systems with ranked output. 4 Moreover, some of the search strate-

gies that were identified are based on specific technical attributes of the search system,

such as the query language (e.g., include quotation marks , use subject directory ) and query

operators (e.g., use Boolean operators ). As a result, the strategies that were discovered

are pertinent to searches under the conditions in which they were discovered, but

they may not be applicable to other modes of information searching, such as browsing

the library shelves or asking a person for driving directions. The more abstract the

level of definition, the more modes of searching it represents.

A few scholars interpreted search strategies on a somewhat abstract level. An

example of such approach is the study by Ramirez et al. (2002) which examined the

role of computers in mediating human-to-human communication, that is, informa-

tion-seeking when the source of information is another human. It seems that they

Five Search Strategies 99

understood search strategies to be the relationship between the information seeker

(the communicator) and the object of the information acquired (the target). They

distinguished three main types of strategies (Ramirez et al. 2002, 219 – 221):

• Interactive strategies entail direct interaction between communicator and target

during which different tactics are enabled to elicit desired information; for example,

the communicator interrogates the target, discloses information designed to elicit

reciprocal disclosure, and attempts to relax the target in order to acquire

information.

• Active strategies involve acquiring information from other individuals but without

direct interaction with the target, as is the case, for example, with the use of third-

party information sources, such as acquiring information through email exchanges

and chats with others familiar with the target.

• Passive strategies involve acquiring information about a target through unobtrusive

observation, such as being “ carbon copied ” on messages, eavesdropping on a conversa-

tion, or lurking on a listserv.

Ramirez et al. ’ s classification demonstrates that universal, or abstract, construal of

search strategies makes them independent of the technology used, and certainly free

of association with technical attributes of an information system, whether a human

or a machine.

In summary, the unsystematic nature of the use of the concept search strategy , sup-

ported by the lack of explicit understanding of the concept, created a muddled trail

of research about search strategies in which only the term itself is common to all

investigations.

5.1.2 Definitions of Search Strategy

Most explicit definitions of search strategies were universal and abstract in nature. The

most universal one was offered by Belkin and his colleagues (Belkin, Marchetti, and

Cool 1993; Belkin et al. 1995). They defined search strategies as the behaviors in which

people engage when searching for information. One might claim that this definition

is too general and actually represents the more general concept information-seeking

behavior (ISB), thus making it difficult to differentiate between the two concepts. Nev-

ertheless, using this approach, they presented four mutually exclusive dimensions (or

facets) of strategies that together create search strategies. That is, each search strategy

is a combination of elements drawn from the four facets. Each facet, in turn, includes

a continuum of elements that Belkin et al. (1993) derived from informal analysis of

empirical studies. For each dimension they listed the two extreme strategies. 5 The

100 Chapter 5

dimensions were method of interaction (from scanning to searching); goal of interac-

tion (from learning to selecting); mode of retrieval (from recognition to specification);

and resources considered (from information to metainformation). 6 These dimensions

demonstrate a very broad construal of search strategies , and raise some questions. It is

difficult to accept goal of interaction , for example, as a dimension of a strategy. A goal

may provide a reason for selecting a certain search strategy but it is not a dimension

of it. This is because strategies are usually associated with activities, whereas goals do

not represent activities and are not even directly identified by them, since various

activities may lead to the same goal and one activity may lead to the accomplishment

of more than one goal. In addition, this broad definition cannot guide researchers in

discovering other strategies, and thus limits the possible strategies to those Belkin

et al. have defined.

A definition that is universal, yet in sync with the notion of strategy in everyday

language, and the first one formulated in HIB, was offered by Marcia Bates (1981). She

explained that a search strategy is: “ An approach to or plan for a whole search. A

search strategy is used to inform or to determine specific search formulation decisions;

it operates at a level above term choice and command use ” (142). This definition is

not bounded by dimensions or technology, and places search strategies as a compo-

nent of information-seeking behavior. An example of a strategy might be: First I ’ ll try

a couple of terms, and if I don ’ t get good results, I ’ ll look for better terms either by

browsing the results or by thinking about the problem in light of what was retrieved.

Gary Marchionini (1995) construed search strategies in a similar way and also placed

the concept in an abstraction hierarchy of concepts in searching behavior, in which

each level is affected by the level above it. Marchionini ’ s hierarchy moves from the

concrete to the abstract:

• “ Moves are finely grained actions manifested as discrete behavioral actions such as

walking to a shelf, picking up a book, pressing a key, clicking a mouse, or touching

an item from a menu ” (74).

• “ Tactics are discrete intellectual choices or prompts manifested as behavioral actions

during an information-seeking session … for example, when restricting the search to

a specific field or document type in order to narrow the search results ” (74).

• “ A Strategy is the approach that an information seeker takes to a problem. Strategies

are those sets of ordered tactics that are consciously selected, applied, and monitored

to solve an information problem ” (72).

• “ Patterns are sometimes conscious but most often reflect internalized behaviors that

can be discerned over time and across different information problems and searches.

Five Search Strategies 101

Patterns may be caused by chunked strategies or tactics that people internalize though

repetition and experience ” (72). One manifestation of patterns is, for example, an

individual ’ s searching style .

Iris Xie (2007) created a similar hierarchy with an understanding of search strategies

that was more general than the previous definitions, and included the goals of a

search. She explained:

Information-seeking strategies comprise interactive intentions and retrieval tactics . Interactive inten-

tions refer to subgoals that a user has to achieve in the process of accomplishing his or her current

search goal/search task. … Retrieval tactics are represented by methods and entities with attributes.

Methods refer to the techniques users apply to interact with data/information, knowledge,

concept/term, format, item/objects/site, process/status, location, system and humans. (Xie 2007,

emphasis added)

These definitions have had an impact on other studies. Vakkari (1999), for example,

used Belkin et al. ’ s (1993) dimensions among other constructs when he analyzed how

an information problem ’ s structure (i.e., structured versus ill-structured) affects search

strategies, and Xie ’ s (2007) definitions were inspired by the approaches of Belkin

et al., Bates, and Marchionini in addition to other views. The definitions have guided

empirical studies as well. Thatcher (2006), for example, employed Marchionini ’ s hier-

archy when he investigated the search strategies that were employed by 80 study

participants. He identified 12 strategies, which he named “ cognitive search strategies, ”

including the following:

The participant went to a search engine that was known to them [ sic ]; participants used different

search engines to conduct the same search; the participant deliberately opened multiple browser

windows to conduct different searches simultaneously; the participant relied solely on hyperlinks

from the homepage to get from one webpage to another. (Thatcher 2006, 1059-1063)

Thatcher ’ s search strategies are different in nature and level of abstraction from

those identified by Marchionini, who envisioned them to be laid out on a spectrum

with opposite ends: the analytical and the browsing strategies. The analytical strategies

are “ planned, goal driven, deterministic, formal, and discrete, ” while the browsing

strategies are “ opportunistic, data driven, heuristic, informal, and continuous ”

(Marchionini 1995, 73). 7 While widely accepted (if not always correctly), the distinc-

tion between these two types of strategies is not compatible with the approach pre-

sented in this book. According to the view presented here, each search is driven by a

goal (to solve an information problem) rather than by data, regardless of the strategies

employed. In addition, every strategy is a plan. Thus, even a decision to start a search

without a specific plan (i.e., browsing) is a plan. With these conceptions, Marchionini ’ s

102 Chapter 5

definitions represent attributes of searching and surfing (see section 2.1.1.1). Since these

are two modes of acquiring information, they are dichotomous, rather than the oppo-

site ends of a spectrum.

In conclusion, definitions of search strategies are usually universal and abstract and

can guide other researchers in identifying specific strategies, whether on a conceptual

level or in empirical studies. But these definitions have had one drawback: Using them

has generated an unruly repertoire of strategies in which each researcher has employed

her own view on how to carve out strategies from an analysis of the literature or from

the data at hand. In addition, the number of search strategies is growing constantly

as new ones are discovered, usually without attempting to place them in relation to

other strategies. Most concerning is the diversity in the levels of abstraction of the

search strategies that have been generated, which ranged from the physical actions to

plans of action. 8 This inconsistency points to fundamental differences among the

interpretations of the concept. With the continually increasing number of strategies,

it is useful to find a configuration that may contain them. One promising approach

to reduce this confusion is to view a search strategy as a category of plans, general

approaches, or interactive intentions (see section 5.4).

5.2 The Conditions That Shape the Use of a Strategy

Various studies identified the conditions that shape the use of a strategy, which are

usually termed “ factors affecting the choice of search strategies. ” Some of the findings

of these studies were based on an analysis of previous studies (e.g., Vakkari 1999), and

others on empirical research (e.g., Ford, Eaglestone, and Madden 2009; Rouet 2003).

In a typical investigation the researcher selects a factor of interest and analyzes or tests

its effect. Thus, Vakkari (1999) examined the effect of the structure of the information

problem; Ford et al. (2009) looked at individual differences; and Rouet (2003) tested

the effect of task specificity and prior knowledge.

Studies of this type face various challenges. For example, the definitions that

researchers employed were unable to lead investigators to the variables that are likely

to affect the selection of search strategies. It is difficult to think about a variable that

may affect, say, the strategy “ using quotation marks ” — except for the obvious one:

whether or not a searcher is familiar with the strategy. With these definitions, research-

ers have had to use a trial-and-error approach when they select the variables to be

tested. Another challenge is the relatively large number of search strategies that were

defined by researchers. Thus, even if investigators find a variable that may affect one

strategy, the variable may leave the rest of the search strategies unaffected. Indeed,

Five Search Strategies 103

typical findings of such studies that tested an array of search strategies pointed to one

or two strategies that were affected by the tested variables but found no factors that

affected the other strategies. This way, one can state that an actor with high value on

variable X is more likely to employ category A than an actor with low values, but the

question “ Which search strategies are an actor with low value is likely to select? ”

remains unanswered. 9 Considering search strategies as a category overcomes these and

other challenges (see section 5.4.2).

5.3 Systems Designed to Support Strategies

Regardless of the definition of search strategies , most scholars agree that information

systems that support the strategies are better than those that ignore them. Yet only a

few researchers have provided systems requirements to support the strategies they

unveiled or redefined. Most systematic among these researchers were Belkin, Mar-

chetti, and Cool (1993). They methodically analyzed each strategy they had defined

to identify the problems that one may encounter when employing it. Thinking about

ways a system could alleviate the problems they identified, they generated 36 require-

ments for information systems interfaces (see section 10.3.3.1). They recommended,

for example, that a system provide a “ display of resources with explanations of link

type, ” “ direct retrieval of example information items from selected terms, ” and “ struc-

tured representation of query and search ” (Belkin et al. 1993, 330 – 331).

While Belkin et al. (1993) offered highly specific requirements, based on all the

search strategies they had identified, Bates (2007) focused on one search strategy —

browsing — and offered a much more general interface requirement. She explained that

“ [g]ood browsable interfaces would consist of rich scenes, full of potential objects of

interest, that the eye can take in at once ( massively parallel processing ), then select items

within the scene to give closer attention to. ” She also presented a model of such an

interface that was developed by Toms (2000) as an example of a good interface. 10

Both Belkin et al. (1993) and Bates (2007) offered implications for the design of

universal systems, regardless of the characteristics of the searchers. Another approach

is to focus on the searchers, identifying the strategy that would be useful to them, and

then generate design requirements based on the actors ’ information behavior. It is

unrealistic to design search systems for each individual, but it is reasonable to do so

for a particular community of actors. In this case an analyst may ask, What strategies

will play a central role in these actors ’ search for information? Once this question is

answered, implications for design could also be based on the typical characteristics of

the actors. Browsing support provided to scientists, for instance, should probably be

104 Chapter 5

different from that offered to youth looking for health information. This difference is

required not only due to the dissimilarity in the actors ’ cognitive resources and

context, but also due to the centrality of the browsing strategy for each community.

While browsing is likely to be essential to youth looking for information in an unfa-

miliar area, scientists are not likely to employ it as a central strategy. Section 5.4.3

provides a comparison between two communities ’ strategy selections and the resulting

design requirements as an example.

5.4 Search Strategy as a Category

A search strategy is cognitive in nature — because plans, general approaches, or interac-

tive intentions are all hatched in the human mind — regardless of the contextual situ-

ation that shapes it. In my work I have applied the conceptual framework cognitive

work analysis (CWA) to HIB (see chapters 11 and 12). CWA views strategies in associa-

tion with decision-making processes (see section 11.1). Vicente (1999) — based on

Rasmussen (1981) — defined a strategy as “ a category of cognitive task procedures that

transform an initial state of knowledge into a final state of knowledge ” (220).

Rasmussen, Pejtersen, and Goodstein (1994) explained that cognitive processes

within the same category — that is, the same strategy — “ share important characteristics,

such as a particular kind of mental model, a certain mode of interpretation of the

observed evidence, and a coherent set of tactical planning rules ” (70). 11 Vicente (1999)

further explained that each strategy is “ based on a different set of performance criteria,

and requires a different kind of information support ” (219).

Strategies can serve various decision processes, such as diagnosis, evaluation, or

planning (Rasmussen et al. 1994).

5.4.1 Five Search Strategies

In the area of information science, field studies in information retrieval (IR) that were

guided by CWA have defined strategies that are employed in the information search

process. 12 More specifically, Pejtersen (1984) uncovered five distinct search strategies

(Pejtersen 1979) in her study of fiction retrieval in public libraries. Later studies have

observed the use of these strategies and found no additional ones. 13 Browsing and

analytical strategies are included in this set, but their definitions are different from

Marchionini ’ s (1995). The strategies are presented in table 5.1

Although each search strategy is derived from a certain mental model, actors may

switch strategy in the middle of a search. 14 One may use a library catalog employing

the analytical strategy, for instance, to find the location of a book on a particular topic,

Five Search Strategies 105

but browse the shelf for additional sources once that book has been located. Similarly,

an actor may enter a complex search query but continue browsing through links when

the results are not satisfactory. When conducting a study of searching behavior, it is

sometimes difficult to detect a strategy shift. This difficulty is particularly the case

when the analysis is based only on observation or on transaction logs. In fact, it is

very difficult to identify search strategies without access to the cognitive processes

involved in the specific search. A transaction log of a web search may show, for

example, two terms in the search box followed by many clicks on links. Without

understanding the mental model the actor had, it is impossible to determine if he

employed the browsing or the analytical strategy. An awareness of the cognitive pro-

cesses is required for the definition of search strategies because they reflect a mental

model rather than specific procedures. Observation and analyses of transaction logs

by themselves can identify only procedures and cannot provide insight to the mental

model that is employed in a search.

5.4.1.1 The Browsing Strategy

The browsing strategy (intuitive scanning following leads by association without much

planning ahead ) had been identified long before computers began to be used for infor-

mation retrieval. Although its most commonly recognized manifestation has been

browsing bookshelves, the introduction of hypertext made browsing a highly viable

strategy when searching digital information systems. A person who decides to browse

in order to find information for making a decision might think: “ Let me start here

and see where it takes me. ” When searching the web, one might follow this decision

by clicking on links or using a directory.

Table 5.1 Search strategies and their definitions

Search strategy Definition

Browsing Intuitive scanning following leads by association without much planning ahead

Analytical Explicit consideration of attributes of the information problem and of the search system

Empirical Based on previous experience, using rules and tactics that were successful in the past

Known site Going directly to the place where the information is located

Similarity Finding information based on a previous example that is similar to the current need

106 Chapter 5

This view of browsing is different from Marchionini ’ s (1995, 73) not only in meaning

but also in type (he argued that browsing strategies are “ opportunistic, data driven,

heuristic, informal, and continuous ” ). His interpretation of the strategy is based on

the category “ elements that drive a search ” (opportunistic, data driven) and on the

category “ manner in which the search progresses ” (heuristic, informal, continuous).

That is, while all these elements that define browsing are cognitive, they belong

to different categories. In fact, according to the CWA definition, a browsing strategy

can fit in Marchionini ’ s analytical one because it can be goal driven, deterministic,

and formal.

The browsing strategy has attracted more research interest than any other strategy,

and has had the widest range of interpretations (see reviews of these in Chang

and Rice 1993 and in Rice, McCreadie, and Chang 2001). One example of a

thorough conceptual investigation into the concept is Bates ’ s (2007) question:

“ What is browsing — really? ” She placed the concept in human development and

found that “ most animals have a propensity toward exploratory behaviour. ” Viewing

browsing in the context of this behavior led her to conclude that “ browsing is a cogni-

tive and behavioural expression of this exploratory behaviour, ” and that in humans,

curiosity is “ the in-built motivation for this exploratory behaviour. ” Thus, her defini-

tion is:

Browsing is the activity of engaging in a series of glimpses, each of which exposes the browser

to objects of potential interest; depending on interest, the browser may or may not examine

more closely one or more of the (physical or represented) objects; this examination, depending

on interest, may or may not lead the browser to (physically or conceptually) acquire the object.

(Bates 2007) 15

On the empirical research front, Shan-Ju L. Chang (2005) carried out the most

comprehensive series of studies on browsing. Besides identifying the dimensions that

can support a description of browsing, 16 she created a multidimensional framework

for understanding the influences on the process as well as the consequences of

browsing.

In addition to being the most explored strategy, browsing is also the most perva-

sively used strategy in information searching. While it is a strategy on its own, it can

also occur as a sequence when other strategies are employed. Retrieving a desired book

from the library shelves, for example, requires some browsing on the shelf before the

specific book can be located. Similarly, when one finds a web site, using any search

strategy, that provides the needed information, one might click on additional links

for further exploration. Despite its prevalence, no formal training about how to browse

Five Search Strategies 107

exists (to my knowledge), 17 and search engines provide no support for the strategy, 18

as evidenced by the common lost-in-cyberspace situation.

5.4.1.2 The Analytical Strategy

Using the analytical strategy, one explores the information need on the one hand

and systems capabilities on the other. 19 The next step is to match the need and

the system ’ s attributes — or, translate the need into a query in the system ’ s “ language ” —

evaluate the options for search actions, and select the most promising one. This is

the rational, decision-making and problem-solving strategy. In fact, it is considered

almost the normative strategy, and training in searching for information usually

involves instructions about how to employ the analytical strategy. 20 The use of

advanced search options or of syntactical operators is typical of an analytical search

on the web, as in any search for which a query is constructed with the aim of arriving

directly at the desired information. Research about this strategy has been primarily

focused on practical advice and suggestions on ways to apply it for efficient and effec-

tive results.

5.4.1.3 The Empirical Strategy

The empirical strategy can be used by actors who categorize previous experiences and

generate rules and tactics. Then, when they face a new situation, they may intuitively

recognize that it fits one of the previously defined categories, and apply the rules or

tactics generated for it. An individual employing this strategy may think: “ In a situa-

tion like this, this is what I usually do. ” An experienced web searcher may have a rule:

“ When I want to buy something, I just put it as a URL. For example, if I need to buy

a flag, I ’ ll enter www.flags.com. ” 21 Or: “ When I need a homepage of a person or an

organization, I just enter as many identifying terms as I can in the Google search box.

Most of the times the homepage will appear at the top of the list of results. ” The

empirical strategy is a shortcut to the analytical one and can be employed only if an

actor has developed rules or tactics.

5.4.1.4 The Known-Site Strategy

In certain situations, an actor knows exactly where the desired information is located.

The trivial case of this strategy is when an actor is led to the site, as is the case when

one is given the URL of a site that has the information, or when one knows exactly

where a book is on the library shelves. To employ this strategy in a nontrivial mode,

however, one has to remember that certain sites exist as well as the information

108 Chapter 5

included in them. To help them to locate known sites, people have created files and

lists that include either the sites themselves — such as a file of physical documents — or

an easy point of access to them — such as bookmarks or a list of favorites.

5.4.1.5 The Similarity Strategy

The similarity strategy is useful when an actor can best express an information need

by presenting a specific item as a basis for retrieving another information item that is

similar to the specific one. In the study of fiction retrieval (Pejtersen 1979), for

example, readers sometimes could best articulate the kind of book they were after by

presenting a book they had already read as an example of the type they wanted.

On the surface, it may appear that the similarity strategy is identical to the retrieval

tactic query-by-example, in which the actor provides an exemplar of what is to be

retrieved. Query-by-example, however, is a procedure, rather than a strategy. As such,

it might be used when the similarity strategy is employed, but it can also support

other strategies. Consider the case of an actor who wants to retrieve the musical scores

of a song from which he knows only one tune. This situation requires the browsing

strategy because entering the tune as an exemplar will retrieve all scores to songs that

have a tune similar to the entered one. To find the relevant score, the actor will browse

through the retrieved ones. Query-by-example can also be used when employing the

analytical strategy, as would be the case if the actor knew the song and purposely

considered which tune would be most effective in retrieving the song ’ s scores. In other

words, the query-by-example procedure may be employed when the actor searches for

a specific document, while actors employ the similarity strategy when they do not

know what documents would satisfy their need and they cannot articulate the docu-

ments ’ attributes.

To design a machine that can retrieve a similar item requires overcoming various

hurdles. The main challenge in defining similarity is identifying appropriate metrics

for its measurement. If an actor likes a certain picture of a tulip, say, and she wants a

similar picture of a rose, how can the system figure out what attributes are most

important to her? This strategy could be most effectively supported, therefore, when

the attributes that actors employ to determine similarity are known and can be rep-

resented unambiguously, or with an interactive and iterative process that can effec-

tively lead to the desired information without explicit similarity criteria. Pejtersen ’ s

(1989, 1992) BookHouse system is an example of an explicit consideration of similarity

attributes. Based on her analysis of conversations between users and librarians, and

on other data, Pejtersen (1992) designed this retrieval system for fiction in which the

retrieval of similar books is based on a network of book attributes.

Five Search Strategies 109

Although many search engines make it possible to search for similar items, most

similarity searches facilitated by the engines are based on the topic or the author of

the retrieved document and ignore other attributes that might play an important

role in an actor ’ s tacit criteria for similarity — such as the quality of graphs and

charts, the intensity of an image ’ s colors, or the technical level. 22 Measurements of

similarity are also used in recommender systems (Resnick and Varian 1997), such as

the familiar recommendation of electronic bookstores: “ Others who bought this

book also bought the following books. ” Recommender systems have not been

used thus far to support the similarity search strategy in information retrieval. Thus,

while the similarity strategy is a viable search strategy, most search systems today do

not afford it. 23

5.4.1.6 Five Search Strategies: Conclusions

The five strategies presented here are general enough to apply to any search regardless

of the technology used. An example may illustrate this point.

The browsing strategy Michael, a resident of a retirement home, won $1,000 toward

buying a computer in a talent competition, and he wants to buy a laptop. He decides

to search the web and find the best computer for him. He is not familiar with the web

and knows very little about laptops and selects to enter “ best laptop computer ” in the

search box and see where it leads him.

The analytical strategy Susan is in the same situation, but she has some experience in

web searching and knows a thing or two about laptops. Moreover, she needs a com-

puter that can stand harsh conditions because it will accompany her on her field trips.

She formulates a query with Boolean operators that include several synonyms for

harsh , such as rugged and rough .

The empirical strategy Steve has much experience with web shopping, and in particu-

lar with buying electronic devices. He has developed a tactic that worked well in the

past. First he displays a list of all the electronic stores, deletes those that do not sell

computers, and selects the web pages of those with which he had good experience in

the past.

The known-site strategy Dave has it best: He knows which store has the top computers

and lowest prices, and he knows what computer he would like to get. He goes directly

to the store ’ s web site.

The similarity strategy Rachel knows which computer she wants — the one her sister

has — but it is too expensive for her. She likes working on it because it has a pleasing

shape, the keyboard is sturdy, and the display is clear. She wishes she could enter the

110 Chapter 5

specifications of her sister ’ s computer and ask for similar ones, so she could scan them

and hopefully find one that fit her requirements.

Search strategies can also be used when looking for information from other people.

The browsing strategy Allyson wishes to get a laptop. She decides to ask advice from

her acquaintances, but she is not familiar with their computer expertise. After her

yoga class she usually walks home with her classmate Ben. Asking him for help, she

finds out he knows very little about computers and cannot help her. Walking her dog

a week later, she meets her next-door neighbor who has never used a computer but

knows that a neighbor down the street is an expert.

The analytical strategy Natasha has many friends who are very familiar with comput-

ers, and so before she buys one, she wishes to consult with some of them. Nora

might not be the best consultant because she doesn ’ t listen well. Asking Kevin is not

simple because he will insist on taking part in the buying process; to be independent,

she will have to pretend that she is helping a friend who needs a laptop. Jesse might

be the best person to ask because she listens well and does not interfere with others ’

lives, but she has not bought a new computer for a long time. Which one should she

choose?

The empirical strategy Haejin likes to talk to people and has many friends. When she

has to make an important decision, such as what laptop to buy, she talks first with

her closest friends and asks for their opinion. Next she turns to her classmates and

her professor. She also calls her family in Korea. Armed with the advice and informa-

tion she collected, she finally makes the decision.

The known-site strategy Amir asks his wife, who is a computer expert, for

recommendations.

The similarity strategy Karen explains to the sales person that she would like to have

a laptop that is similar to the desktop computer she has in her office.

These examples illustrate that under certain cognitive and contextual conditions a

particular strategy, or a certain combination of strategies, is more promising than

others. Moreover, defining search strategy as a category makes it possible to understand

the conditions that shape the use of a strategy, and therefore supports informed

design.

5.4.2 The Conditions That Shape the Use of a Strategy

The view of search strategies as categories of cognitive task procedures that are

induced by a mental model makes it possible to infer which cognitive and contextual

Five Search Strategies 111

conditions motivate the use of a strategy. To illustrate this possibility, I present

an examination of the relations between each strategy and a sample of resources an

actor has:

• Time — how much time the actor has for a search

• Prior knowledge required for the employment of a strategy

• Cognitive processes — how much thinking is required when using a strategy

• Search in memory — how much human remembering is required for a strategy ’ s use. 24

This analysis does not rely on rigorous measurements, such as the number of

minutes an actor has to perform a search. Instead, it assesses the level required from

each condition to proceed with a strategy in comparison to the other strategies. For

example, browsing requires much time, while the known-site strategy requires little

time.

The browsing strategy Browsing is the easiest strategy to apply because it does not

require previous knowledge, much thinking, or remembering. It is, however, the least

efficient strategy to resolve an information problem because it rarely leads to the

desired answer directly, and therefore requires more time than the other search strate-

gies. In other words, a person who has little knowledge on the information problem,

the subject domain, or web searching (or a combination of those), but has no strict

time limits, would find the browsing strategy most useful.

The analytical strategy To be productive, this strategy requires some knowledge of the

search system and its functionality and some knowledge of the subject domain. Search

time is shorter than when using the browsing strategy, and some search in memory

is required. On the other hand, to proceed with the strategy entails much thinking

and analysis. It is most useful when an actor is searching for something new — such

as a new topic or a variation on a topic — or is searching on a new system, and the

actor has neither rules that can direct the search nor a site he knows will provide the

needed information.

The empirical strategy This strategy requires analysis as well, but instead of comparing

the information need with the system ’ s capabilities, an actor compares the need

with rules and tactics she has developed through previous searches. Even though

the empirical strategy is, in a way, a shortcut to the analytical strategy, it has

different requirements. While it saves time, it requires an actor to search her

memory to a greater extent than the analytical strategy. In addition, only actors

who have enough knowledge and experience and who have developed rules and

tactics can employ it. The empirical strategy can be beneficial to actors who have

112 Chapter 5

much experience in searching but have limited time and are looking for something

new.

The known-site strategy On the surface it seems that the known site is an ideal strategy:

It is the most efficient, and it does not require much knowledge and analysis. Yet it

can be employed only under very strict conditions. Not only must the actor remember

a site that is likely to have the desired information, she must remember particular

attributes of the site: those that can provide her with direct access to the information.

Thus, while highly efficient, using the known-site strategy creates high memory load

and requires an extensive search in memory.

The similarity strategy This strategy is most useful when one cannot articulate the

information problem but can present to the system information that is similar to what

is needed. In fact, in many cases this might be the only strategy an actor can use

because he cannot come up with terms that might lead him into a productive search.

All that is required to apply this strategy is an item that can serve as an example.

While the strategy requires little effort on the actor ’ s part, it can be materialized only

if the search system — whether human or machine — can carry out similarity searches

successfully. Moreover, its efficiency depends on the ability of the search system to

produce relevant results. The better the system, the less scanning is required.

Comparing the strategies The relationships between conditions and strategies are sum-

marized in table 5.2 . The values little , medium , and much represent a relative amount

in comparison to the other strategies.

What is the relevance of these relationships? Once we know what conditions are

typical to a community of actors, we can anticipate what strategies they are likely to

use, and therefore which strategies require most support. These predictions could

guide the design of an information system that supports search strategies.

Table 5.2 A comparison of the resources required for each strategy

Strategy Prior knowledge Cognitive processing Search in memory

Time spent on searching

Browsing Little Little Little Much

Analytical Medium Much Little Medium

Empirical Much Little Much Little

Known site Much Little Much Little

Similarity Medium Little Little Little

Five Search Strategies 113

5.4.3 Systems Designed to Support Strategies

While the few researchers who analyzed strategy use and offered design requirements

have focused on general-context search systems, a more extensive application of

search strategy definitions and analysis can take place when an information system is

designed for a particular community of actors. When focusing on a community,

researchers are familiar with the actors ’ information behavior and can integrate this

knowledge into design requirements. This approach is based on the assumption that

an understanding of the searching behavior of a community ’ s members is relevant to

design that supports search strategies. A comparison between two communities of

actors — high school students and engineers, both using the web to find information

for their work — may serve as an example to demonstrate this approach.

Table 5.2 displays resources required for the application of each of the five strategies

defined in section 5.4.1. With regard to these resources, an empirical study of high

school students (Fidel et al. 1999) revealed that the participants had low levels in all

resources but one: time. They were searching in a subject area of which they were

completely uninformed, and thus had no prior knowledge. Their priority was to find

the easiest way to complete the assignments, and thus they avoided heavy thinking

efforts. Because they lacked web-searching experience as well as subject knowledge,

they did not have to search in their memory. At the same time, they had plenty of

time to complete the assignments. Examining table 5.2 shows that browsing is the

most promising strategy for them. Indeed, this is the only search strategy we observed

them to employ.

The designer of a search system for students with similar resources — and for other

actors under the same conditions — would make a special effort to provide support for

the browsing strategy. The specific support to be provided would depend on additional

constraints that might be specific to each community of actors. In our study we dis-

covered that the main challenge the students confronted was finding their place on

the web space; that is, they “ got lost in cyberspace ” very easily. Therefore, providing

means to quickly find where actors are and providing mechanisms for an easy access

to pages visited earlier would have improved their interaction with the search system

and its outcome.

Such support mechanisms are best when their design is based on the conditions

under which the interaction takes place. The students we observed selected landmark

pages during the search to help them find their way. At some point in their browsing

they selected a certain web page that they used as a “ home base, ” or a “ starting point, ”

from which to venture into the web space, expecting to come back to it if they needed

114 Chapter 5

to continue their explorations in a different direction. 25 Their greatest challenge was

to find their home base when they wanted to get back to it. An interface to support

this technique in browsing would offer an easy and fast way to return to a landmark,

that is, to the home base. 26

Search support may also address strategy shifts. The students who participated in

our study became frustrated at times with the poor results they received through

browsing. Their searches might have been more effective if they could have employed

the analytical strategy. Their lack of subject knowledge and of experience in web

searching, however, made it difficult for them to employ the strategy. A search system

that at any time during browsing guides searchers in formulating their queries for the

analytical strategy would have been highly useful to them.

The conditions under which engineers work, and the resources available to them,

are unlike those of the high school students. Therefore, engineers require different

types of support when searching the web. My studies of engineers searching the web

(e.g., Fidel and Green 2004) revealed that they almost always have some level of prior

knowledge of the subject matter, and they are not reluctant to think or to search their

memory, which has been nourished by their experience in web searching. In most

workplaces, however, time is in short supply and engineers are frequently under time

pressure. With these levels of resources, the best strategies for engineers are the known-

site and empirical strategies. The analytical strategy is also viable — particularly in

searches in which the other two strategies cannot be applied — but requires additional

time.

A search system to support engineers ’ web searching therefore would help them in

applying these three strategies. Memory is the resource most in demand to carry out

the known-site and empirical strategies; therefore, search systems for engineers could

be designed to support them when they search their memory. For the known-site

strategy, engineers need to remember addresses of web sites, and for the empirical one,

they have to store their searching rules and tactics. The “ bookmarks ” (or “ favorites ” )

facility is already providing support for the known-site strategy and it could be a focus

for improvements. 27 Facilities to store and organize search rules and tactics have not

been developed yet (to my knowledge). The analytical strategy could receive support

as well. The engineers who participated in our study almost always planned their

search when they employed the analytical strategy. Most of them expressed their desire

to have a facility that would make it possible for them to predict the search results

during planning.

Thus, an analysis of the constraints under which actors look for information can

reveal the search strategies that are likely to be most useful. An understanding of these

Five Search Strategies 115

constraints and the searching behavior of actors of a certain community can lead to

the creation of system requirements that would support the use of these search

strategies.

5.5 Five Search Strategies: Conclusions

The concept search strategy is associated with the search process and has become a

regular object of research since the introduction of digital technology increased both

the number of possibilities to employ search strategies and their visibility to research-

ers. Search strategies are considered essential to the analysis of a search process,

and the many definitions of the concept were inspired by the views their creators have

on the process. Most of the implicit understandings of the concept — that is, the use

of the term search strategy without defining it — have been derived from observable

search procedures, and their level of specificity has varied. Thus, strategies with very

different levels of abstraction — such as include quotation marks , rely solely on hyperlinks

from the homepage to get from one webpage to another , and employ rules and tactics that

were successful in the past — were identified, mostly through empirical research. At the

same time, most explicit definitions of the concept search strategy have been abstract.

The plurality of ways to construe the concept and the continuous identification of

new search strategies make cumulative research unattainable. 28 Therefore, it is neces-

sary for the scholarly community to arrive at some common understanding of the

concept if it wishes to reduce the fragmentation in search strategy research.

Most empirical research on search strategy has been dedicated to the study of the

factors that might affect the selection of a strategy. Only a few researchers have

addressed the implication of this research for design, that is, the design of information

systems that support the use of search strategies. One of the challenges to such design

stems from the reality that requirements based on the technology in use at the time

of their generation are short lived and become obsolete when a new technology is

introduced. Therefore, requirements that are likely to be stable over time should be

independent of the technology in use. Yet both of the examples of design recommen-

dations presented above (i.e., Belkin et al. 1993 and Bates 2007) are technology-

dependent. Bates ’ s (2007) suggestion that the interface for browsing “ would consist

of rich scenes, full of potential objects of interest, that the eye can take in at once ”

addresses browsing with the aid of digital technology. This requirement cannot be

applied to browsing with other types of technology, such as books or file cabinets,

and might bring serious design difficulties for systems with small screens such as cell

phones.

116 Chapter 5

Belkin et al. ’ s (1993) requirements were highly concrete and on the level of actions.

As such, they were based on the search systems that were in place at the time of their

research, and some of them are immaterial for web-based search systems. To generate

these requirements, the researchers identified problems that might occur when carry-

ing out several information-retrieval tasks and then offered a design requirement that

would help to alleviate each problem. For example, one of the problems they identi-

fied that related to “ search strategy formulation ” (i.e., employing the analytical strat-

egy) was the difficulty users had in understanding the use of the search logic. To

alleviate this problem, the researchers suggested that the search interface should “ mask

logic from user. ” A contrary requirement may alleviate the problem to the same degree,

however: Search interfaces could represent the logic, that of Boolean algebra for

example, and lead users to enter Boolean statements by providing a structure in which

query formulations are entered. Such a structure could be in the form of a table in

which each term that is connected to another term with the “ AND ” operator is entered

in a different column, and those associated with the “ OR ” operator are in the same

column.

These examples show that defining search strategies on the actions level is unlikely

to lead to the generation of useful design requirements. An individual search action

might be supported by more than one possible design feature. Which one is the

designer to choose? In addition, requirements based on a list of actions may at times

contradict one another since an overall view, a direction, or an approach is missing.

Moreover, actions are shaped by the technology in use, and some cannot be material-

ized with new technologies. Technology also limits the modes of information seeking

in which a strategy can be employed. An action-level strategy that is identified in web

searching, for instance, would not be applicable for searching in physical spaces, such

as libraries and other people.

Defining strategies as categories of cognitive task procedures — rather than specific

actions — resolves several challenges to producing stable and useful design require-

ments and thus makes the concept search strategy relevant for design. First, focusing

on cognitive procedures — rather than on physical actions — results in requirements

that are independent of technology. Having this focus is not foreign to search strategy

scholars. Despite the large diversity in the construal of search strategy , most scholars

would probably agree that it is cognitive in nature. Therefore, it is reasonable to focus

on cognitive processes instead of recording the mechanics of actor-system interaction

with no insight into the cognitive processes that generated it. Second, defining the

different categories of search strategies enables a relatively comprehensive identifica-

tion of possible search strategies. A design to support all possible actions-based strate-

Five Search Strategies 117

gies requires the identification of all possible actions-based strategies, which is an

impossible task. On the other hand, the number of categories that cover possible

procedures or actions is relatively small and stable, and does not require the identifica-

tion of all possible actions. Third, viewing cognitive categories as a focal point opens

a window to understanding the reasons for strategy selection that can guide the design

of systems that cater to specific communities of actors, because the actual implementa-

tion of category-based search strategies is informed by the contextual constraints that

shape them. That is, this approach is context-centered and beneficial to ecological

design.