Chapter 3: A Human-Web Interaction Cycle

The previous chapter presented Neisser’s (1976) perceptual cycle. This chapter outlines a human-web interaction (HuWI) cycle, which was derived from the perceptual cycle (see Figure 2).

Figure 2. The HuWI cycle is the two outermost layers placed around the perceptual cycle. The definitions of schema, exploration, and available information are now more specific to HuWI.

Overview

Figure 2 depicts the HuWI cycle. The figure shows that system knowledge (SK) directs interaction (e.g., reading, locating, clicking links, etc.), interaction samples available information in the computer system, and the sampled information then modifies SK. The outermost layer depicts the specific types of SK, interaction, and available information, all of which will be discussed in the following sections.

The reciprocal relation between SK, interaction, and information in this cycle is useful for describing how these components all interact to achieve the user’s goals. For example, if one intends to find information about the Emancipation Proclamation on an American history website, then how does one go about achieving this goal? According to the HuWI cycle, one would need to use SK, interaction, and the available information in the website. The following sections describe what happens at each stage of the cycle in more detail. As with the perceptual cycle, we begin with a discussion of goals because the user’s intentions are what set the cycle in motion.    

The Influence of Goals

As can be seen in Figure 2, there is now an explicit representation of intent or goal because of the especially important contributions that it makes within the entire HuWI cycle. The goal selects the SK to use, as depicted by the arrow connecting goal to SK in Figure 2. The SK might be, for example, a generic “how to use the web” SK when looking for information on a website that has never been used before (Dillon, McKnight, & Richardson, 1993). If the website is about American history and the user has no prior knowledge of this topic, then she would use this generic or “global” SK in order to find the desired information. More will be said about global SK later. For now, the major point is that, regardless of how much a user knows about the system, it is the goal that selects what SK will be used for interacting with that system.

Because behavior, given the selected SK, is not random, the goal must influence the ways in which users interact with the system, as depicted by an arrow pointing to the word “directs”. In addition, the goal influences what information is sampled about the system, as depicted by another arrow pointing at the word “samples.” It should be noted also that there is no direct influence of goal on schema modification. Since the only information that is sampled is a function of user interaction with the system and the user’s goal, then the influence of the user’s goal on schema modification is indirect. Therefore, Figure 2 does not depict a relation between this component of the cycle and goal. The influence of the goal on directing interaction and sampling information, and its indirect influence on schema modification, will be detailed in parallel with the respective stages in the cycle within the following sections.

System Knowledge Directs Interaction

There is little doubt that users must have some sort of conceptual representation, schema, or mental model of a system in order to interact with it successfully (Boechler, 2001; Hargittai 2003; Jenkins, Corritore, & Wiedenbeck, 2003; Kim & Hirtle, 1995; McDonald & Stevenson, 1998, 1999; Rouse & Morris, 1986). The types of SK that are important for website interaction might be classified into two types (based on Schulmeister’s [1997] classifications of navigational aids):

1. Content Knowledge: What the website is about. This includes two inseparable components:

   1. Domain understanding (e.g., what the information is, what it is used for, what it is related to, etc.)

   2. Lexical understanding (e.g., the verbal and iconic meaning of text and other media, “scent” or how hyperlink text and icons are interpreted to convey target information via category labeling [Larson & Czerwinski, 1998; Pirolli, 1997; Pirolli & Card, 1995] etc.)
       

2. Structure Knowledge: How the website is organized. This includes two inseparable components:

   1.  Architectural understanding (e.g., webpage-level structure[1]  such as where links are located on a webpage and where content is located on a webpage; website-level structure such as where one is within the website and where one can go by following hyperlinks on the website, etc.)

   2.  Chronologic understanding (e.g., when a webpage was viewed, when a link or text was seen, etc.)

Content knowledge appears to be important for interacting successfully with a website. Hill and Hannafin’s (1997) verbal protocol analysis of users interacting with the web found evidence in support of this position. They observed that inadequate content SK had adverse affects on the quality of participant’s interactions. This suggests that problems in website design might be caused by website authors and users having different kinds or levels of knowledge (Calvi, 1997).  For example, the author of an American history website will almost certainly have a high level of American history content knowledge, whereas a user of the website might have little to no understanding of American history. The terminology used by the author and the way that she structures the website, therefore, will be inconsistent with the knowledge held by the naïve user. It is this incongruence between website author and website user knowledge that is the impetus for finding ways to measure user knowledge (e.g., pathfinder analysis and task analysis) and apply those results to website design (Chen, 1997; Jonassen, 1988, 1993; Kurniawan & Zaphiris, 2003; Patel, Drury, & Shalin, 1998). That is, by presenting content and structuring information in ways that are consistent with the potential user’s SK, one can enhance the usability of a website.

Some authors have argued also that structure SK is important for successful interaction. For example, Kim and Hirtle (1995) and Edwards and Hardman (1989) suggest that designers of websites should give special attention to helping users acquire structural SK (via sitemaps and “landmarks”) so that they would not become “lost” within the website. This view is shared by several other authors (Beasley & Waugh; 1995; Calvi, 1997; Hammond & Allinson, 1987; Leventhal, Teasley, Instone, Rohlman, & Farhat, 1993; McDonald & Stevenson, 1998, 1999; Schroeder & Grabowski, 1995).

The Role of System Knowledge in Directing Interaction

According to the HuWI cycle, one reason that naïve users might have problems using some websites is that SK directs interaction with the website. If users have inadequate understanding of the website’s content, then their interaction with the website will be inferior to that of users with a better understanding of the website’s content (Bhavnani, Drabenstott, & Radev, 2001 [as reported in Olson & Olson, 2003]; Carmel, Crawford, & Chen, 1992; Jenkins, et al., 2003; Korthauer & Koubek, 1994; McDonald & Stevenson, 1998; Symons & Pressley, 1993).

Content knowledge, however, is not the only kind of SK that will direct users’ interaction with the website. Structural knowledge can direct interaction also. For example, imagine that the American history website author links all historical documents together in one part of a website and all wars together in a different part of the website. If the user expects, based on structural SK acquired with previous visits to the same or similar websites, that the historical documents are linked to their respective wars or events, then she will have problems locating the information that she desires. For example, the user might look for the Civil War webpage in order to find the Emancipation Proclamation, rather than looking for a page that links all of the historical documents together.   

The Role of Goals in Directing Interaction

According to the HuWI cycle, content and structure SK, together with the user’s goal, are what direct her interactions with the system. Goals direct interaction in two ways. First, they select the SK that will be used to interact with the website. Certainly, a user with the intention of finding facts about the Emancipation Proclamation would not summon up their knowledge of refrigerators to explore an American history website.

Secondly, goals focus the interaction. For example, if the user seeks the Emancipation Proclamation and she arrives at a page with the links to the Gettysburg Address and the Emancipation Proclamation, then she is most likely to select the link that will satisfy her goals. Similarly, if the user arrives at a page that has links to the Spanish-American War and the Civil War, and she has the knowledge that the Emancipation Proclamation is associated with the Civil War, then she is more likely to select the Civil War link (assuming that she doesn’t know that the author put the Emancipation Proclamation under the “Documents” link).  

Testing that System Knowledge Directs Interaction

It is well established that users with more knowledge (either domain knowledge or knowledge of how to use the web) will be more successful at finding information on websites (Bhavnani et al., 2001 [as reported in Olson & Olson, 2003]; Carmel et al., 1992; Jenkins et al., 2003; Korthauer & Koubek, 1994; McDonald & Stevenson, 1998; Symons & Pressley, 1993). However, research has not established why this occurs. According to the HuWI cycle presented here, the reason is that SK actually directs users’ interactions with the website. Accordingly, the first experiment reported in this dissertation tested this prediction by having participants find information on a website using knowledge that was given to them. Since all participants had the same goal, it was predicted that the paths that they would take through the hyperlink structure would be determined by the SK that they had while completing the task.

Interaction Samples Available Information in the Website

It goes without saying that users learn information as they use a website. What is unclear, however, is what kinds of information are picked up or sampled from the website as a user interacts with it. Accordingly, “information” as it is used here must be classified. The potential kinds of information available for pick up are the same as the kinds of SK that are useful for interacting with a website:

• Content Information: What the website is about. This includes the two inseparable components of domain data and lexical data

• Structure Information: How the website is organized. This includes the two inseparable components of architectural data and chronologic data

Both content and structure information are potential sources of information available for pick up. For example, when interacting with an American history website, one could sample content information and/or structure information. If a user learns that the Emancipation Proclamation is associated with the Civil War, then she sampled this piece of content information. If, on the other hand, she learns that the website author placed the Emancipation Proclamation in a section containing all historical documents in the website, then she sampled structural information. It is possible, however, that both types of information are sampled, for example learning that the website author put the Emancipation Proclamation in a section with only historical documents and then reading the document to learn that it is associated with the Civil War.

The Role of Interaction in Sampling Information

Interaction determines what information is made available for pick up because interaction generates the information. Simply stated, that which is not interacted with cannot be sampled. For example, if a user interacts with the “Documents” section of an American history website, and never interacts with the “Wars” section of the website, then the only information made available for pick up is that which was in the documents section. It is impossible for them to pick up information about something that they did not interact with. They could make inferences about information in parts of the website that were not explored (i.e., rely on SK to create expectations of what information might exist there), but there is no way that this information could have been sampled. 

The Role of Goals in Sampling Information

According to the HuWI cycle, users’ interaction with a website samples some kind of information (i.e., content and/or structure information). More specifically, their interactions sample goal-relevant information. Of the information made available for pick up, the primary type of information sampled will be relevant to the goal. For example, Eveland and Dunwoody’s (2000) verbal protocol analysis of users’ information processing while interacting with the web suggests that not all information is processed during interaction. Eveland and Dunwoody state that “Most of the thoughts generated by the think aloud procedure referred to the content of the sites… instead of their structure.” (p. 233). If one assumes that participants were attempting to achieve their goals, and their goals related to content rather than system architecture, then there was no need for them to sample information that was relevant to the website’s structure.

The idea that interaction samples goal-relevant information may account for results obtained by Farris, et al. (2002). They demonstrated that users of a website did not remember one form of information (the structure) but did remember another (the content). Although participants’ goals were not defined clearly in the study, one might assume that the task that they completed was relevant to content and not structure. Specifically, participants were instructed to seek out information (i.e., images) that they would be interested in using to design a website, which appears to be a content-related goal. Their participants were not given a goal that related to structure, which, for example, could have been to examine the way the website was organized in order to make recommendations about enhancing its usability. The research reported in this dissertation attempts to clarify this issue. However, it is conceivable, with the available evidence, that if one knows how the user interacted with the website and what the user’s goals were, then one can predict the information sampled from the website.

This position is counter to the argument that the hypermedia/hypertext^[2] linking organization (i.e., concepts presented on webpages linked with hyperlinks) enables users to not only acquire the content knowledge embedded in this structure but also the structure itself. Cunningham, Duffy, and Knuth (1993) state that “Users will browse the hyperspace, serendipitously acquiring knowledge and the structure of the database as represented in the links. In other words, by traversing the links within the hypertext, a user will acquire the content and the form of the database” (p.38). This assumption is based on what is known as the “plausibility hypothesis,” which states essentially that the network-like representation of a subject matter in hypertext, and the nature of the links between information units, support associative browsing which corresponds to the structure of human knowledge and the basic principles of the functioning of the human mind (Churcher 1989; Cunningham et al., 1993; Jonassen, 1988; Jonassen & Wang, 1993; Vrasidas, 2002). Hypertext, therefore, should improve learning because it focuses attention on the relationships between ideas rather than isolated facts (Kearsly, 1988).

However, when tested empirically, this plausibility hypothesis was not supported (Tergan, 1997). Research suggests that the non-linear structuring of subject matter in hypertext format does not improve comprehension and retention of subject matter compared to linear text (Landauer, 1995). Moreover, the results of several studies like this “indicate that although hypertext-based learning may have advantages in the amount of facts reproduced in a recall test, [linear] text-based learning often resulted in better comprehension and reproduction of central concepts” (Tergan, 1997, p. 263).

In addition, an implicit assumption of the plausibility hypothesis is that users will “serendipitously” sample the goal-irrelevant information. Farris et al. (2002), however, demonstrated that participants were unable to reconstruct the structure of a browsed website when asked to diagram it, but did recall the basic concepts of the domain. Although their participants’ goal when browsing the website was unclear, it seemed that participants were more focused on content information than on structure information. Either way, the serendipity assumption of the plausibility hypothesis was unsupported because participants could not recall the link-webpage relationships that would be expected if users acquired the structure of the website serendipitously. Others (e.g., Eveland & Dunwoody, 2000) have found similar results using different methodologies.

Similarly, in a series of studies, Jonassen and Wang’s (1993) participants were unable to learn the semantic relationships between concepts (i.e., relationships between different pieces of content knowledge) in a hypertext. In fact, the only way that they could get participants to pick up this information was to construct a task that required participants to attend to the semantic relationships. There appeared to be no significant serendipitous pick up of goal-irrelevant information. They concluded that “It is the assigned processing task and goals for learning while interacting with a hypertext that appears to most significantly determine the effects of its use on learner’s knowledge structures” (p. 1).

Testing that Interaction Samples Available Information in the Website

The aforementioned studies appear to be in accord with the HuWI cycle, which states that the information sampled is goal-relevant. Although there is evidence to suggest that this is indeed the case, there has been no research aimed specifically at resolving this issue. Accordingly, the second experiment that is reported in this dissertation addressed this by assigning different tasks or goals to participants using the same website and then testing for what type of information was sampled (content or structure). Based on the HuWI cycle, it was predicted that they would sample goal-relevant information primarily. The plausibility hypothesis, on the other hand, predicted that they would sample both kinds of information regardless of the goal.

Available Information in the Website Modifies System Knowledge

It has been demonstrated empirically that information, via experience with a website, modifies users’ SK (e.g., Elgin, Jones, Anders, & Farris, 2001; Leventhal, et al., 1993; Rouet, 2003; Shapiro, 1999). However, if all of the arguments presented heretofore are true, then not all information on a website will modify the user’s SK. Specifically, if interaction with the website samples goal-relevant information, then that is the only information that will modify the user’s SK. Consequently, the role of the goal in modifying system knowledge is indirect. 

According to the HuWI cycle, goal-relevant information that was sampled from the website modifies SK. In other words, if the user’s SK prior to using a website is known and the user’s experience with the website is known, then the information that will be made available to modify SK can be predicted. For example, in order to learn about the relationship between the Emancipation Proclamation and the Civil War, a user must have a goal that will lead to seeking that information (i.e., goal-relevant information) in order to make it available and one must interact with the website in order to make the information available. Without both of these, SK cannot be modified because the information would not be sampled. Simply stated, one cannot know information that was never made available by one’s interactions and goals. Therefore, by knowing what the users’ goals are and what they interacted with on the website, one can predict what information was made available to modify their SK.

The Role of Information in Modifying System Knowledge

According to the HuWI cycle, the available information in the website modified SK. This process is best described as a sort of cognitive adaptation, which is an adjustment in cognitive organization that results from the demands of interaction with the website. This is the same as Piaget’s (1952) concept of accommodation. In accord with this concept, Dillon et al. (1993) argue that users have an initial schema, or “global schema,” that is a “...general knowledge of the world that aids humans in navigation tasks” (p. 172). This schema is later modified (i.e., accommodated), or becomes an “instantiated schema,” via interaction with the system.

Some have argued that instantiated schemata have certain “spatial” properties (e.g., Beasley & Waugh; 1995; Calvi, 1997; Edwards & Hardman, 1989; Hammond & Allinson, 1987; Kim & Hirtle 1995; Leventhal, et al., 1993; McDonald & Stevenson, 1998, 1999; Schroeder & Grabowski, 1995) while others have argued against this position (e.g., Boechler, 2001; Dias & Souza, 1997; Farris et al., 2001, 2002; Jones & Dumais, 1986; Mayes, Kibby, & Anderson, 1990; Shum 1990; Stanton & Baber, 1994). This “spatial hypothesis” assumes that users of a website form a cognitive map of the web, just as navigators in a city form a cognitive map of the city. Boechler (2001) and Farris et al. (2001, 2002) point out that this assumption is impossible because websites lack spatial properties (e.g., distance, shape, size, optical flow, etc.) that are necessary to form a cognitive map.^[3] Indeed, the only identifiable relation between a cognitive map of a website and a cognitive map of a city is the idea that landmarks (or webpages) are connected by paths (or hyperlinks). Even still, this similarity seems more metaphorical than practical (Boechler 2001). In addition, Farris et al. (2002) reported that participants did not remember the link-node relationships (i.e., system structure) on a website.  From this they argued that the spatial assumption was not supported, since remembering these relationships is a necessary condition to establish that users form cognitive maps.

However, Farris et al. (2002) did not manipulate their participants’ goals. As mentioned, the experimental tasks that were completed apparently related to the website’s content and not its structure. Therefore, one might assume that participants were sampling and modifying their SK with content information and not structure information.

Perhaps this debate can be resolved by asking a more informative question, i.e., “what kind of mental representation do users form?” According to the HuWI cycle, the SK has content and structure knowledge. Therefore, counter to Farris et al.’s (2002) conclusions, users can acquire structure knowledge. This is predicated on the assumption that the information that they sample, which can be content and structure, is structure primarily. For that to occur, the user must interact with the system with goals that relate to the system’s structure. If, on the other hand, the user’s goals do not relate to system structure, but rather to the content, then the information that they pick up is content primarily and there is little modification of the user’s structure knowledge.

Testing that Available Information in the Website Modifies System Knowledge

One purpose of the first experiment reported in this dissertation was to test whether or not SK is modified as information is sampled. This was done by having participants repeat an information-finding task several times. It was predicted that if the available information modifies SK, and SK directs interaction, then participants should find the target information more efficiently each time that they complete the task. In addition, comparisons of pretest and posttest measures of SK were used to test for improved SK after interacting with the website.

In addition, one purpose of the second experiment reported in this dissertation was to examine the issue of what kind of information modifies SK (i.e., only goal-relevant information or all information, as in the spatial hypothesis and the plausibility hypothesis, respectively). In order to accomplish this, participants completed a task on a website that was either related to system structure or to system content. It was predicted that if users sample goal-relevant information primarily, then those assigned to the task relating to system structure should know more about the hyperlink relationships (i.e., what pages are hyperlinked) in the website than participants assigned to a system content task. In addition, it was predicted that participants assigned to the system content task should know more about the content (e.g., what concepts are related to each other) in the website than participants assigned to the task relating to system structure. The following section summarizes the experimental hypotheses.

Hypotheses

H₁: If SK directs interaction on a website, then the paths taken through the hyperlink structure will be predicted by the SK (Experiment 1).

H₂: If the information sampled is goal-relevant, then the type of information sampled (content or structure) while interacting with a website, and subsequently recalled, will be predicted by goals (Experiment 2).

H₃: If the goal-relevant information sampled from a website modifies SK, then information will be found more efficiently as experience using a website is gained (Experiments 1) and more goal-relevant than goal-irrelevant information will be in SK (Experiment 2).

[1] Research has demonstrated that users have consistent expectations, or schemas, for where web objects (i.e., menus, text, advertisement banners, etc) are located on a webpage (e.g., Bernard, 2001). Although these schemas are definitely important and should be investigated, we will limit our discussion to website-level structure (i.e., structure of the webpages within the larger website) since this is an area that is less understood and fraught with more controversy in the literature.

[2] Hypertext is the text typically displayed on the web, including content and links. Hypermedia includes hypertext, pictures, sound, and all other types of multimedia displayed on webpages. Although they actually mean different things, hypertext and hypermedia are typically used interchangeably to mean "everything we interact with on the web" and will be used interchangeably here.

[3] It is for this reason that a distinction should be made between structure SK and a cognitive map. Structure SK, as previously defined, includes the knowledge of how webpages are linked together within a website. It is not assumed that a cognitive map-like representation is formed.

Document Last Updated December 31 1969 19:00:00.