A Exploratory Study of IR Interaction
for User Interface Design.

Preben Hansen
Swedish Insitute of Computer Science
preben@sics.se

Accepted for the Poster session at the SIGIR 97,
20th International ACM SIGIR Conference on Research and Development in
Information Retrieval, Philadelphia, PA, USA

July 27 -- July 31, 1997


1. Introduction

With the rapid growth of distributed network-based information systems, together with tools like World-Wide Web-browsers, there has been an increasing spread and use of information and through a more user-driven accessibility to information systems such as hypertext systems, where users can search, navigate and browse within information spaces.
The broad and diverse existens of information systems, in parallel with different user interfaces and functionalities. This situation now presents a number of challenges in the field of IR and HCI researcher. In order to understand the new challenges, we need to examine factors such as: how users interact with IR systems; how to design user interfaces for IR systems; different information seeking strategies and behaviour (Belkin 1995); the users' tasks and goals, individual differences, cognitive abilities (Chen 1996), and how to enhance users' navigation in the information space (Benyon, 1997). We used a networked document database (Dienst), containing a set of research reports within the computer science area, to examine some of these factors.

The IR Interaction and HCI problem
We are constantly involved in various interactions with the environment through different communication mechanisms or processes. Information seeking is such a process, where users in different ways interact with the information environment. Participants involved in the interactive information seeking process, are the user of the information, the information retrieval (IR) system, and the intermediary.

Recently, we have seen a growing interest and examples of interdisciplinary research within the information science area and the computer science area (Hewins, 1990, Sugar, 1995, and Koenemann and Belkin 1996). Current approaches to IR interaction research has created studies based on the assumption that the traditional IR does not support or reflect interaction (Belkin et al, 1995, Ingwersen, 1992, 1996, and Saracevic, 1996).

Studies in user behaviour and individual differences (Borgman, 1989), and user interface in information retrieval (Belkin et al 1995, Marchionini, 1995) have proved that this area is of great interest. Questions that have attracted growing interest are: how do we make a better adaptation to users' different preferences such as their tasks, goals, abilities, individual differences and to support these in the redesign of the user interface? We consider IR as an iterative and interactive communication process between the information user, the IR system, and the user interface with the purpose to transform the information need into query formulations.
This means that the user interface should have a supportive task. It is obvious that research on IR techniques solely cannot provide the understanding, support and interaction between the user and the IR system. A central focus of research, as Allen (1996) points out, is the need to establish a link between research within IR and the design of user interfaces.

A major problem inherent, and recognized, is that the methods to evaluate information retrieval systems, under a long period, have been focused on recall and precision, but not at the usability of the user interface and how well users can accomplish their goals through the system. Our present approach will be to acquire knowledge for a redesign, and we take an approach which combines the user-centered IR interaction perspective (Ingwersen 1992) and the user-centered design methods in HCI (Nielsen, 1994) on ) to evaluate information seeking interaction and tasks in a hypertext system; to identify characteristics of the user population; and finally, to make suggestions for supporting user characteristics and needs in the interface design.

We have formulated the following hypotheses: Since the current user interface to the Dienst database does not provide the users with a good support for their information seeking task. Different users show different preferences i.e. knowledge, goal, information need and expectations. Different information seeking strategies and behaviour can be observed and identified in the users' information seeking behaviour.

IR Interaction and User Interface Design
To accomplish our task, we adopt a user-centered approach, in which, based on a modified model by Bryce (1996, p.24), we identify the following framework of methods:

COMPONENT METHOD TASK
Resource Analysis Description of information system functionalityDescribe resources used to complete the tasks.
User Needs Analysis
  • Questionnaire (qualitative and quantitative data)
  • Log statistics (quantitative data)
  • Users goals, purpose, objectives, actions, individual preferences.
  • Measures like time, type of actions
  • Task analysis Hierarchical Task AnalysisUsers tasks, goals and activities that they accomplish when meeting their needs.
    (User Modeling)
    Designing for usabilityRequirement lists (qualitative data)Requirements for user interface redesign

    The design of the user interface should, among other things, take in consideration following aspects: to support user interaction with texts and other components of the system; to support evolution of goals, and change of knowledge; and to support a variety of types of interaction.

    Evaluation
    Generally, one of the main task for evaluating IR systems is to obtain information concerning the satisfaction of the userŐs task in a specific work environment. This will include subtasks like the userŐs information need, their knowledge, the systems functionalities etc. In many of the subtasks users are involved. How each task is performed, and how the different tasks is connected will be influenced by the user interface.

    Within the HCI research, there has been extensive work within the usability evaluation area. This includes highly controlled user tests in a laboratory environment, where subjects are performing specific tasks that are observed using different techniques like GOMS.
    We have applied HCI evaluation techniques to our of IR evaluation to make a connection between the traditional IR evaluation and HCI evaluation. The formar usually measures precision and recall, while the latter involves task analysis, domain description, and sometimes user evaluation and analysis. This is already performed within the IR community, that is within the library and information science research area.

    2. Research Design

    The data used in our study were collected during August-November 1996, as follows:

    Data collection methods Types of data collected Data analysis methods
    Internet-based evaluation
    questionnaires before
    and after information seeking
    task.
  • Quantitative data:
    5-point Likert scale from questionnaire.
  • Qualitative data:
    Written data to the 5-point Likert scale.
  • Statistical analysis of quantitative data
  • Content analysis of qualitative data
  • Comparison of statistical data.
  • Task analysis of qualitative data
  • Download of search log history.
  • Questionnaire (qualitative and quantitative data)
  • Log statistics (quantitative data)
  • Quantitative data analysis.


    The study set-up
    This study is based on an previous project, initiated by the ERCIM , in which we participated. A Dienst (Lagoze 1995) server was installed and SICS research reports were stored, indexed and made accessible through the WWW interface. We used an experimental online evaluation methodology setup, combining questionnaires and log statistics of database search logs to examine users actions, time estimates concerning the whole search and individual actions made by the user.

    To accomplish our task, we used the following methods to collect data: Questionnaires: two questionnaires were used, one before and one after using the system. The pre-search questionnaire explored the user's preferences, intentions and goals and contained five questions. The post-search questionnaire contained questions like satisfaction with search result, navigation support issues and design issues, and contained nine questions.
    The questionnaires were set up in a non-controlled situation, i.e. the subjects were asked to participate but not "forced" to answer every question or comment. Each questions is followed by a Likert scale of 1 to 5, to be checked by the user. Verbal (comments) data: In addition, to every question, there is a "comment"-field, where the subject could clarify or verify his/her statement in the 5-point Likert scale. Database search log: Data collection for usage investigation will be facilitated by logging each online user' server request. Each usage record will contain a users identification, the total search time, search actions and errors encountered using the system.

    Data collection and analysis methods and procedures
    The goal of the data analysis was to capture information about the user's information seeking behaviour and strategies to extract lists of values and requirements that form the framework for design decisions and feed the information to the user through the user interface. The data collection and analysis were conducted as follows: We contacted potential participants in either three ways: personal contacts, electronic mailing lists and through recommendations and the selection was based on the potential interest in the subject domain (computer science). When the participant had conducted the task, the answers from the questionnaires and the log statistics of a user were linked together, creating an individual log-file of a user. After that we created a coding scheme and ran our data through that scheme.
    We examined the collected data at three levels: a general level (all users), at a group level (all users in that group) and at an individual level (a single verbal statement).

    We used both descriptive and statistical analysis methods:

    1. quantitative analysis of the score (mean) for all the users;
    2. quantitative analysis of the score (mean) on a group level; and
    3. qualitative (verbal analysis) and quantitative analysis on the individual level.
    When analyzing the statistical data, the scores on the Likert scale were grouped as follows: point 1 and 2 ("low score"), and 4 and 5 ("high score") were grouped together. Scores on point 3, were treated separately.

    Log statistics
    The mean time for a information seeking task was 11,5 minutes, and the mean numbers of actions was 9,6 actions, which gives 1,1 minute per action. The users spent a total of 438 minutes in the system and conducted 366 actions. 19 subjects (50%) out of 38 only browsed or searched. 17 subjects used the search functionalities, and 2 subjects used browsing. This is evidence for the poor support for browsing within the Dienst system. The analysis of the user log indicates that subjects used both browse and search to accomplish their task.

    The Users
    The subject entered this particular system for the first time. About 150 subjects were asked to voluntarily participate in this study. We wanted to conduct this study in a real environment and with as real users and situation as possible. The subjects were asked to perform some information seeking task based on a real information need. 38 subjects (16 female , 21 male and 1 anonymous) completed the two questionnaires. The users were separated into the following groups: Computer Science researchers (CS), Industry (Industry) and Information Specialists and Librarians (ISL). 37% of the participants were computer science researchers, 24% worked within the industry, and 39% were information specialists and librarians.

    3. Preliminary findings

    NOTE that this document only shows the preliminary result. A comprehensive list of result will be presented later. This will include correlation measures as well.

    Our basic assumptions was that different users show different preferences i.e. knowledge, goal, information need and expectations. We also believe that different information seeking strategies and behaviour can be observed and identified. Based on our findings, we will make suggestions for a redesign of the user interface.
    To study this, we identified the following main categories: User background, knowledge and preferences; User satisfaction; User behaviour; and User requirements elicitation. The categories were measured through a set of variables described in each category.

    Single Variables examined We collected information about the users' background knowledge and preferences.
    We also made a simple structure of the users into 3 user groups. When analyzing the written comments, we also created a list of requirements of the users' expectations before and after using the system. The data also resulted in a general table of user tasks and goals.
    In chapter 5.2, we have examined the category of user background, knowledge and preferences (selected single variables) and results from statistical analysis is presented. We will noe present some conciderations that will influence the user interface redesign. Based on these, we will in chapter 7, make suggestion for a user interface redesign.

    Relations between variables
    Some preliminary results from the study of the combined variables. We found a strong relation between users with both a high level of previous experience and a high level of satisfaction with search result. The level of user satisfaction with search result depends on several factors. When comparing the variable "high user satisfaction witth search result", with other variables, we found that there was a close relation between the variables a high satisfaction with content, satisfaction with system effectiveness and navigation support and the variable high satisfaction with search result. We could also see a relation between high satisfaction with navigation support to complete a information seeking task and high IR knowledge. We also found a strong relation between satisfaction with navigation support to complete a information seeking task and domain knowledge understanding (terminology used in the system). It seems that most of the subjects had a good knowledge about the subject domain, and that this knowledge affected the satisfaction with navigation support. This means that we need to design the interface according to the knowledge resource domain.
    Also, log analysis of mean time and mean action resulted in that there is a stronger relation between time spent in the system and satisfaction with navigation support to complete a task, rather than actions made satisfaction with navigation support to complete a task. Subjects that spent more time (+30 sek) in the system had a higher level of satisfaction with the navigation support to complete a information seeking task.
    ISL had the highest score for IR knowledge. The score for low knowledge on how to formulate queries for the CS group were 21%, and for the ISL group 0%. This shows that a distinction between the ISL and CS group.

    Browse- or search-oriented strategy.
    The subjects who had high level of previous experience wanted to use the browsing strategy (36%), and the subjects who had a high IR knowledge also wanted to browse (36%). Of all user with high level of navigation support, 59% wanted to use browsing strategies and 24% wanted to use search (analytical) strategies. We also wanted to see if it was true what the users said that they wanted to do and what they really did. To do this we examined the answers given in the questionnaires, and then compared them with the users logfiles. The result shows that the subject who said that they browsed, used the search options, and the subject who said that they searched, also did so. The reason for this could be that the user interface have a poor support for browsing and that the users were "forced" to use the search functions within the system. Furthermore we analyzed the log statistics concerning type of action and categorized them as browse (35%) or (analytical) search actions (65%). We found that both the CS and ISL group made equal numbers of search actions categorized. Another observation is that the CS group made 17 actions to inspect the retrieved item and the ISL group only one (1). The reason for this could be the obvious: that the ISL group, by profession does not conduct the evaluation of document, rather the customer.

    User Interface Design
    The design of effective human-computer interfaces becomes ever more critical to overall system performance. One task for the design of user interfaces would be to reflect the users' tasks and information problem, and help users to solve tasks by being adaptive and supportive to a user's knowledge and goals. We have identified several levels of work that must be understood in order to understand information seeking activities: the users' specific preferences, tasks and goals; the users' information seeking behaviour; and the use of an information system. Suggestions for a user interface design will be presented later, based on the knowledge acquired from studies above. Our study has also produced a simple model for knowledge acquisition for user interface design (and redesign) of information retrieval systems.


    References
    Allen, B. (1996). From Research to Design: A User-Centered Approach. In: Ingwersen, P. and Pors, N. O. (eds.). Second International Conference on Conceptions of Library and Information Science: Integration in Perspective (CoLIS 2), Copenhagen, Denmark. October 13-16, 1996. pp. 45-59.

    Belkin, N. J., C. Cool, A. Stein and U. Thiel. (1995). Cases, Scripts, and Information-Seeking Strategies: On the Design of Interactive Information Retrieval Systems. Expert Systems With Applications, 1995, Vol. 9 (3), pp. 379-396.

    Benyon, D. and K. Höök. (1997). Navigation in information Space: supporting the individual. INTERACT'97. Australia. (Forthcoming)

    Borgman, C. (1989). All users of information retrieval systems are not created equal: an exploration into individual differences. Information Processing & Management, 25 (3), p. 237-252

    Bryce, A. (1996). Information Tasks. Towards a User-Centered Approach to Information Systems. San Diego : Academic Press.

    Chen, C. and R. Rada. (1996). Interacting With Hypertext: A Meta-Analysis of Experimental Studies. Human-Computer Interaction, Vol. 11 (2). Pages???

    Hewins, E. T. (1990). Information Need and Use Studies. In : Williams, Martha E. (ed.). Annual Review of Information Science and Technology (ARIST), Volume 25, 1990. pp. 145-172. Amsterdam, The Netherlands, Elsevier : ASIS.

    Ingwersen, Peter. (1992). Information Retrieval Interaction. London : Taylor Graham.

    Koenemann, J. and N. Belkin (1996). A case for interaction: A study of interactive information retrieval behaviour and effectivness. In : Proceedings of CHI'96 Human Factors in Computing Systems. New York : ACM Press. pp. 205-212

    Lagoze, C. and Davis, J. (1995). Dienst: An Architecture for Distributed Document Libraries. Communications of the ACM, 1995, 38 (4), p. 47.

    Marchionini, Gary (1995). Information Seeking in Electronic Environments. Cambridge Series on Human Computer Interaction 9. Cambridge : Cambridge University Press.

    Nielsen, J. and R. Mack (eds.) (1994) Usability Inspection Methods. New York : John Wiley & Sons.

    Saracevic, T. (1995). Evaluation of evaluation in information retrieval. In: Fox, E. A., P. Ingwersen, and F. Fidel (eds.). Proceedings of the 18th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR'95). Seattle, Washington, USA. July 9-13, 1995. pp. 138-146. New York: ACM.

    Su, L. T. (1992). Evaluation measures for interactive information retrieval. Information Processing & Management Vol. 28 (4), pp. 503-516.

    Sugar, W. (1995). User-Centered Perspective of Information Retrieval Research and Analysis Methods. In : Williams, Martha E (ed.). Annual Review of Information Science and Technology (ARIST), Volume 30, 1995. pp. 77-109. Medford, N. J. : ASIS.


    Preben Hansen | preben@sics.se

    1997-04-12