Professional Documents
Culture Documents
User Modeling in Web Transactional Tasks: Visual and Behavioral Patterns
User Modeling in Web Transactional Tasks: Visual and Behavioral Patterns
ORG
Index Terms Eye tracking, user modeling, transactional task, visual attention
1 INTRODUCTION
of users can perform on the web The types to tasks thatclassification are navigational, according Broder's informational and transactional [1]. Tasks are considered navigational when the user wants to reach a particular page or element on the web site, informational when the user wants to find specific information available on the web site, and transactional when the user reaches a specific page and interaction occurs to complete a form; for example, completing a shopping transaction or filling out a form. User modeling has long been used in HumanComputer Interaction (HCI) research to investigate effective methods for optimizing user interfaces. User modeling is often used to understand how users interact with systems, such as examining critical incidents in which problems occur, and the levels of understanding of users. Services and transactions are increasingly being offered online for users, but there exists limited research on modeling users in transactional tasks when compared to navigational and informational tasks. Eye tracking has been shown to be an effective usability tool for measuring the cognitive and perceptual capabilities of users in their interaction with interfaces. Many metrics can be collected and analyzed based on the assumption that eye movement reflects the cognitive process during the users interaction. The ocular behaviors that are usually measured are: saccade, fixation, and scan path. These measures are used in many studies to
evaluate the efficiency and effectiveness of interactive systemsin supporting the users goal [2][3][4]. The focus of this study involves developing a user model that describes users behaviors in conducting transactional tasks on the web. The model is motivated by the Theory of Action [5], which quantifies the perceived relevance of web form elements to the user's goal by cognitive processing and behavioral mechanisms. The model combines behavioral and eye gaze metrics which are derived from visual exploration patterns exhibited by users. In exploratory eye tracking experiments, visual patterns of user interactions with web-based transactions were examined for verifying the models representation of user behavior and cognition.
2 RELATED WORK
Our interest in this paper is examining how users interact with forms in the context of the Web, and specifically transactional tasks. Most of the frameworks have been proposed for navigational and informational tasks. For example, navigational frameworks include the framework for Human-Web Interaction [6], and navigation in electronic worlds [7]. In terms of transactional tasks, the modal theory was proposed by Bargas-Avila et al. which depicted two modes of interacting with forms, namely the completion mode and the review mode [8]. General frameworks have been constructed based on the understanding of user navigation patterns in hyperlinked structures and are often informed by usability practice such as the User Action Framework [9]. In addition to frameworks, theories have been developed that describe the user behavior such as Normans Theory of Action [5].
Mashael Al-Saleh, MSc is with the Department of InformationTechnology, King Saud University, Riyadh, Saudi Arabia. Areej Al-Wabil, PhD is with the Department of InformationTechnology, King Saud University, Riyadh, Saudi Arabia.
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 3, MARCH 2012, ISSN 2151-9617 https://sites.google.com/site/journalofcomputing WWW.JOURNALOFCOMPUTING.ORG
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 3, MARCH 2012, ISSN 2151-9617 https://sites.google.com/site/journalofcomputing WWW.JOURNALOFCOMPUTING.ORG
In conversation layers, one way to make the questions easy to answer or fields easy to fill-out is to allow the users to visually examine their responses by showing acceptance or rejection messages. Also, in some questions flexibility is offered to users in the sense that users are not forced to answer the questions from specific options. For example, providing "others" option and allowing users to choose the date format when they enter the date of birth. The second part of the conversation layer is to write useful instructions. In Mawhiba, this was reflected by using familiar words in writing the labels and the headings of the form, starting the form with short informative title and instructions that describe the purpose of the form. The first and second part of the conversation layer is shown in Fig. 2.
The fourth part of the conversation layer is to make the form easy to follow. In our Mawhiba example, the form is not long and it is suitable to be in one page, no unexpected variation from conventions is found in the form, and the conversation finished smoothly with a "Welcome" message to indicate the completion of the registration task and end of the process as shown in Fig 3.
The last layer is the appearance layer, which is concerned with the visual design details in designing fields and labels. Moreover, it deals with the arrangement of controls and selection of colors that contribute to the visual design of the web form.
Fig. 2: Conversation layer in Mawhiba
The third part of the conversation layer is to choose the form controls. Table 2 shows a sample of controls on forms and the users' expectations of using these controls in Arabic interfaces.
2.3 Eyetracking in Usability and Web form design In eye tracking experiments, the users are asked to perform a specific task while their ocular behavior is recorded. The eye movements can be played back and participants can be asked to describe their behavior while they examine their own interaction with the stimuli. This pro-
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 3, MARCH 2012, ISSN 2151-9617 https://sites.google.com/site/journalofcomputing WWW.JOURNALOFCOMPUTING.ORG
cedure is called Retrospective Think Aloud (RTA). Many metrics can be measured and analyzed based on the assumption that eye movements reflect the cognitive processes of users during the user interaction. One of these measures is the fixations when the eye is stable and aimed to specific visual display elements. Some of the metrics related to the fixation are: Number of fixations Fixation duration Sequence of fixations across a visual display which is called a scan path. Recently, there has been a proliferation of studies using eye tracking in web usability for understanding the effect of some design issues on users and gaining insights into user behavior in performing specific web tasks. The most recent research concerned with optimizing the design of web interfaces using eye tracking methods has focused on navigational and informational tasks. While there are few studies on transactional tasks, their focus was how the changes of the design layout and elements affect user behavior. Nielsen and Pernice, in their Eye Tracking Web Usability book, have compiled a set of guidelines related to designing web forms from eye tracking studies on web forms [14]. They discussed positive and negative aspects of User Interface (UI) elements and presented advice on how the design can reduce unnecessary complexity measured by increased gaze fixations of users on web forms. Another set of guidelines was proposed from an eye tracking study conducted by C. Tan et al. [15].They focused their study on how easy the user can complete the form and user satisfaction. They chose four sign-in pages: Yahoo mail, Hotmail, Google mail, and eBay; where these pages showed different design combinations in layout design, grouping, and ways in indicating optional fields. Another study conducted by Penzo et al. was based on recommendations from the 'Web Form Design' book [16]. They conducted eye tracking tests measuring eye movement activity and duration for different label designs, formatting and types of form fields. They designed four forms for the test where each of them contains four input fields. Like the other studies, they concluded the study with design guidelines regarding label position, alignment of the labels, bold labels, the use of drop-down list boxes and placement of labels. A recent study was conducted by Terai et al. to investigate the effect of task types in information-seeking behavior [17]. Participants were asked to perform two types
of web search, an informational task and a transactional task, and their behavior was examined. Moreover a think aloud protocol was followed in that study to record users thoughts and opinions as they interacted with the system during the sessions. The participants behaviors have been classified into eight categories and the measures have been calculated for each category. Their results showed that the participants visit more pages and complete the tasks in shorter periods in transactional tasks when compared to informational tasks in interactive systems.
A model for human interaction in transactional tasks is depicted in Fig 4. The model describes how the user conducts a transactional task through filling out a web form.
The model considers that each field in the form contains one or more of the following parts: label, introduction/ instruction, and controls (textbox, list) as shown in Fig 5.
The model contains six major stages: form a goal, determine strategy, act, perceive system state, evaluate field, and evaluate form. Each main stage may contain sub stages. The user begins with the first two stages to choose
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 3, MARCH 2012, ISSN 2151-9617 https://sites.google.com/site/journalofcomputing WWW.JOURNALOFCOMPUTING.ORG
the form. The following four steps are concerned with filling out individual fields. Finally, users end the transaction by evaluating the form prior to submission or moving on to a different activity in interaction.
3.1 Form a goal The users start the transaction with a goal or service they want to obtain. The assumption is that this goal is clear in the mind of the user and reflected by the target link that the users choose to reach the transactional interface. Moreover, the selection of the forms link is affected by the users knowledge. When the users have experience about transactions or some familiarity with the interactive system, they can choose the transaction form or locate it within the information structure easily and quickly. 3.2 Intention After entering the form to do the transaction the users check whether the form matches their needs or not. The intention can be established by verifying the action prior to proceeding with examining content by reading the forms introduction, instruction, or scanning the form. A helpful introduction, which is often the instructions in the beginning of the form with a heading, is often a paragraph that describes the purpose of the form, privacy statement, and list of anything the users might need to complete the form [13]. When the users perceive that this is not a form that matches their needs, they start searching for an alternate link reflecting their goal of locating the transactional form. 3.3 Determine strategy
After deciding that the form is the correct form, the users proceed to filling out the form fields by first determining the fields strategy. The field has a predefined strategy that the user should understand to manipulate the controls in the correct manner (e.g. click the arrow for a drop-down list). The information about the field strategy can be acquired from field instructions, label and user knowledge (e.g. prior experience in completing similar fields such as drop-down lists and calendar popup). The users perform this stage by scanning the field and/or label and moving to the next step when they understand the strategy.
3.7 Evaluate Form After filling out all the fields, users evaluate the form by reading the form fields and examining whether there is any system feedback before ending the transaction.
4 MODEL VERIFICATION
An eye tracking exploratory experiment was conducted to verify this proposed model by examining visual and behavioral patterns that are exhibited in the different phases of the model.
4.1 Method
The model was tested against a detailed set of eye tracking data collected from 10 participants as they engaged in three transactional tasks using Arabic interfaces of web-based forms. The testing environment was controlled for consist lighting, temperature, and noise.
4.2 Participants
The number of participants in the experiment was 10. Participants ranged in age between 17-34 years. The average was 23.4 years, and standard deviation 4.54 years. Their computer experience ranged between 8-14 years, with mean 10.7 and standard deviation of 2.37 years. Their internet usage was between 7-14 years, with average 9.7 and standard deviation of 2.45 years. In order to record and analyze eye movements, aTobii X120 eye tracker was used. Eye movements were monitored using Tobii studio gaze analysis software version 2.2.1. The stimuli were displayed in Internet Explorer version 7 using
3.4 Act
After understanding the field strategy, the users start filling out the field. The field actions are categorized under two types: select and fill. The users start one of
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 3, MARCH 2012, ISSN 2151-9617 https://sites.google.com/site/journalofcomputing WWW.JOURNALOFCOMPUTING.ORG
an HP 22 inch monitor. The resolution was set to 1024 X 768 in the external monitor and the laptop monitor.
4.3 Procedure
Participants were tested individually. The experiment started with a 5 minute introduction about this study followed by requesting participants to fill out a demographic questionnaire. Then the experiment began with calibration, followed by asking the participants to conduct 3 web transactional tasks. After that, a Retrospective Think Aloud protocol (RTA) was conducted by playing back the participant's eye movement in selected tasks. The aim of the RTA was to ask the participants to describe their reasoning for observed behavior when they were conducting the transactional task.
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 3, MARCH 2012, ISSN 2151-9617 https://sites.google.com/site/journalofcomputing WWW.JOURNALOFCOMPUTING.ORG
system state, and evaluating the field. The verification of this part was studied in the task of registration in educational portal (i.e. Mawhiba stimulus). For the first step [Act], it was observed that users found this intuitive and directly completed this step without any difficulties or hesitation in responding. An example of this step is shown in Fig 10, in which the users attempt to understand the strategy by scanning the label and text box and then filling out the form. The eye gaze and mouse click shows the user understanding of what is required to complete this step of the transaction. Common controls are often dealt with by users intuitively because they have been previously exposed to them in other web forms and they are familiar with how they work.
The rest of participants did not understand the calendar strategy directly. They tried to understand the calendar strategy by exhibiting a focused visual examination of the area, as this was evident by long fixations on the calendar button as shown in Fig 8, or by exhibiting random fixations around the calendar as shown in Fig 9.
Fig. 8. Scan path of a participant who did not understand the calender strategy directly-1
To study the users perception of the system state and the Evaluate field segments, we examined cases in which users entered incorrectly-formatted data and the system responded with feedback. This change was shown as the change of field value, and system feedback to indicate the acceptance or rejection of the entered value. It was observed that the participants deal with system state as follows: 50% of the participants scanned their entered value and whether there was an error message provided by the system as shown in Fig 11.
Fig. 9. Scan path of a participant who did not understand calender strategy directly-2
By comparing the behavior of the two participants' groups, they were markedly different in the duration of visual examination and the number of fixations. The participants who understood the strategy directly exhibited a mean fixation duration of .81s and a total of 12 fixations in this segment. In contrast, the participants who didn't understand the strategy directly exhibited a mean fixation duration of 1.7s and a total of 33 fixations in the same segment.
Fig. 11. The participant reviews the data that was entered incorrectly
40% of the participants scanned the error messages immediately after filling out the field. Moreover, 30% of the participants scanned the acceptance message immediately after filling out the fields as shown in Fig 12.
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 3, MARCH 2012, ISSN 2151-9617 https://sites.google.com/site/journalofcomputing WWW.JOURNALOFCOMPUTING.ORG
Fig. 12. Visual pattern of a participant visually scanning the acceptance message
Fig. 15. Visual pattern of participant who did not understand the system feedback and repeated the same stages [ Act- precive system state- evalute field]-2
After perceiving the system state, the user is expected to examine and understand the system feedback. This understanding affects the user behavior after the feedback. For example, when the system provide error message the participants after understanding the error message, corrected the error by repeating the same steps (determinestrategy>act>perceivesystemstate>evaluation). The movement of the users in conducting these four steps can be easy if the form conforms to good design principles in the conversation layer in writing instructions, choosing the form controls, and validation and showing messages [13]. Finally, when the result was accepted by the users, they move on to the next field as shown in Fig 13.
Fig. 13: The user accepted the system state and moved to the next field
When the system feedback is not acceptable or not understandable for the user, the steps are repeated more than one time. This was reflected on their behavior and it was evident as long fixations, and long scans on the field and label in attempts to try to understand this issue and determine the source of the error as shown in Figures 1415.
Fig. 14. Visual pattern of participant who did not understand the system feedback and repeated the same stages [ Act- precive system state- evalute field]-1
The other three conducted a bottom-up evaluation, starting from the last field and moving up while scanning the interface as shown in Figures 17-18. During the evaluation, the users scan correctly entered data, check wrong fields or in some cases both.
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 3, MARCH 2012, ISSN 2151-9617 https://sites.google.com/site/journalofcomputing WWW.JOURNALOFCOMPUTING.ORG
line with cognitive engineering approaches in HCI research, a reliable model needs to be established in order to utilize this information in automated evaluations of interactive systems.
ACKNOWLEDGMENT
This work was supported in part by grant #A-S-110565 from King Abdulaziz City for Science and Technology (KACST), Riyadh, Saudi Arabia.
REFERENCES
[1] A Broder, "A taxonomy of web search," SIGIR Forum, vol. 36, [2] L. Ball and A. Pool, "Eye Tracking in Human-Computer Interaction and Usability Research: Current Status and Future Prospects," in Interaction, Encyclopedia of Human Computer, Ghaou [3] L. Cowen, "An eye movement analysis of web-page usability," M.S by research in the Desgin and Evalualtion of Advanced Interactive Systems 2001. [4] C. Ehmke and S. Wlison, "Identify web usability issues from eye tracking data.," in paper persented at 21st Group Annual Conference,University Lancaster,UK, vol. 1, UK, 2007, pp. 119-128. [5] D. Norman, "Cognitive Engineering," in User Centered System Desgin, LEA, Ed. Hillsdale, 1986, ch. 3, pp. 31-62. [6] C. Pilgrim, G. Lindgaard, and Y. Leung, "A Framework for Human-Web Interaction," in CHISIG Annual Conference on HumanComputer Interaction (OZCHI ), Ergonomics Society of Australia, Wollongong., 2004. [7] S. Jul and G. Furnas, "Navigation in Electronic Worlds," CHI 97 Workshop Report 1997. [8] J.A. Bargas-Avila, G. Oberholzer, P. Schmutz, M. de Vito, and K. Opwis. Usable error message presentation in the World Wide Web: Do not show errors right away. Interacting with Computers, Volume 19, Issues 3 , pp. 330-341, May. 2007. [9] T. Andre, H. Hartson, S. Belz, and F. Mccreary, "The user action framework: a reliable foundation for usability engineering support tools," Human-Computer Studies, vol. 54, pp. 107-136, 2001. [10] C. Choo, B. Detlo, and B. Turnbull, "A Behavioral Model of Information Seeking on the Web-- Preliminary Results of a Study of How Managers and IT Specialists Use the Web," ASIS Annual Meeting Contributed Paper 1998. [11] S. Farris, "The Human-Web Interaction Cycle: Aproposed and tested framework of perception, cognition, and action on the Web," Kansas State Uninversity, Manhattan, Kansas, PhD Thesis 2003. [12] The World Wide Web Consortium (W3C). W3C recommendation[Online]. http://www.w3.org/TR/html4/interact/forms.html#h-17.1 [13] C. Jarrett and G. Gaffney, Forms that Work. Burlington, USA: Elsevier, 2009. [14] J. Nilsen and K. Pernince, Eyetracking Web Usability. New York, USA: New Rider, 2010. [15] C. Tan. (2009, April) Web form design guidelines: an eyetracking study. [Online]. www.cxpartners.co.uk/thoughts/web_forms_design_guidelines _an_eyetracking_study.htm [16] M. Penzo. (2006, July) Label Placement in Forms. [Online]. www.uxmatters.com/mt/archives/2006/07/label-placement-informs.php [17] H. Terai, H.Saito, Y. Egusa, M. Takaku, M. Miwa, and N. Kando.
6 CONCLUSION
This work aimed to develop a model of user behavior in transactional web tasks. This was accomplished by constructing a user model based on HCI cognitive and behavioral models. Moreover, the model was verified by examining visual attention of users and investigating the degree to which visual patterns relate to the model in an exploratory eye tracking study. The model provided a good match to participants' interactions in transactional tasks, particularly completing text and selection fields. In
JOURNAL OF COMPUTING, VOLUME 4, ISSUE 3, MARCH 2012, ISSN 2151-9617 https://sites.google.com/site/journalofcomputing WWW.JOURNALOFCOMPUTING.ORG
10
2008. Differences between informational and transactional tasks in information seeking on the web. In Proceedings of the second international symposium on Information interaction in context (IIiX '08), Pia Borlund, Jesper W. Schneider, Mounia Lalmas, Anastasios Tombros, John Feather, Diane Kelly, and Arjen P. de Vries (Eds.). ACM, New York, NY, USA, 152-159.