Erik Nilsen

This workshop is coming at an ideal time for me. I am currently engaged in research and teaching projects involving the WWW. I am also developing plans for future research which will benefit greatly from discussing central issues in Web development from an HCI perspective.

This semester I am teaching (for the second time) an undergraduate course in HCI with an emphasis on searching the WWW and HTML programming. The centerpiece of the course is a collaborative project in which a student team works with a faculty "client" to search for Internet resources and develop a home page relevant to the faculty members research. A description of the first time the class was offered entitled "The Gnosis of Knowledge Navigation on the Net" has just been accepted for publication by the Behavior, Research Methods, Instrumentation and Computers journal. Along with the development of the Web page as a deliverable product, students keep a detailed log which gives insights into the development of strategies in the search process. The rapid pace of change in available resources and searching tools contributes to a highly collaborative class setting where students take an active role in the learning process.

The students quickly gave up opportunistic browsing in favor of search engines and indices. They had very little patience with sites that were inaccessable and seldom returned for a second look. Many of the groups had the similar experience of initially locating a large number of sites when exploring a general area, only to be disappointed at the dearth of material when they focussed their search on a more specific topic. One analogy which suggests itself is of an information sieve. At the aggregate level, the sieve holds a great deal. However, when the clumps of information are broken apart, the smaller particles fall through the holes of the sieve, leaving an empty container. Whether this problem is due to poor search mechanisms or the early stage of development of the WWW is unclear.

One current research project developed out of the classroom experience is a comparative evaluation of several popular search engines. A student collaborator and I are conducting a Signal Detection Analysis of Alta Vista, Excite, InfoSeek Guide, and Lycos. Faculty "subjects" generate a list of keywords for a topic of current interest. After running the keywords though the search engines, the subjects will go through the list of hits (including the first 20-30 found by each search engine) and rate their quality. The outcome of the analysis will be 2 scores for each search engine. One score (d') measures the sensitivity of the search engine in finding useful information. The other score (beta) measures how conservative or liberal the search engine is in determining which sites to include in its output.

My research plan for the next year involves extending my previous research on performance models and method choice in navigating in spreadsheets to navigating on the Web (Method engineering: From data to model to practice; Nilsen, Jong, Olson, & Polson; CHI proceedings, 1992).

We have shown that experienced spreadsheet users only use a fraction of the available navigation methods provided by the software designers. Formal models (e.g. GOMS) can generate predictions about which navigation method is "ideal" in a given circumstance. This analysis shows that some methods are never optimal and this calls into question why they are provided as options. Providing multiple methods has real costs in terms of decision time (Olson, J.R., Nilsen, E. Analysis of the cognition involved in spreadsheet software interaction; HCI, 1988). While users choose among methods in a consistent manner, there is more going on than a rule-based calculation of efficiency. Perceptual landmarks and procedural context plays an important role in method choice.

Jakob Nielsen has identified 15 different hypertext navigation methods, 9 of which are used in current WWW browsers. I have pilot data which suggests that only 3 of the methods are utilized over 80% of the time. Furthermore, there are differences in method choice between novice and experienced users. Can GOMS or other formal models provide insight into which of these navigation methods are needed and provide a real benefit to the user? What are the psychological mechanisms driving the choice of navigation methods in this huge information space. Take for example the two most commonly used methods of new users, clicking on hotlinks and using the "Back" button. The hotlinks capitalizes on the attentional focus and the current goals of the user. This method should be highly susceptible to the "Garden Path" error. Links which seem to match the users current goal will cause them to prematurely jump to a new page before fully reading the current page. The "Back" button, located in a fixed position minimizes search time and serves a perceptually salient landmark. The likelihood of using it should be high during the first minute of looking at a page since the contents of the previous page will still be available in short term memory.

I have recently acquired an eye tracking computer which will strengthen my ability to examine the perceptual saliency of web page features. The frames feature incorporated in the latest Netscape browser is one area ripe for research. HCI researchers should provide some guidelines for when, where, how many and what to use frames for.

I look forward to discussing HCI research issues concerning the WWW as well as broader issues of the structure and content of HCI related material that should be made available on the Web. As a field we must put our best interface forward in our Web sites in order to remain credible in our recommendations to other disciplines using this powerful new communication medium.