Web Site Searching and the User Experience
BayCHI, January 1999
Avi Rappoport, Search Tools Consulting

Presentation Summary by Howard Tamler

Web Site Searching and the User Experience by Avi Rappoport, Search Tools Consulting Avi introduced her talk by emphasizing that search is a central part of users' experience on the web, where the information space is so vast. Search is crucially important because it gives users a measure of control, enabling them to access information that's not obvious, and to bypass navigational hierarchies as they go directly to the information they need. And needless to say, search is one of the major advantages of electonic media as opposed to paper-based information. Because of its centrality, the search interface should be on the Home page of any web site, and preferably on all pages.

Turning to the subject of how the search interface should be designed, Avi recommended that search forms should conform to the look and feel of the rest of the site, and should include at least the AND and OR Boolean operators, since most users are used to them by now, and they look for these tools. Moreover, search tools should be tailored to the particular audience needs of a given site, whether it's for kids, librarians, or used-car buyers. She provided several examples of such tailored search interfaces:

* KidZone Search, with its visual appeal to kids (color, simplicity, large text, etc.)

* the American Verse Project site, which provides access to poetry. Accordingly, the search tool is tailored to enable proximity searching (e.g., find "dreary" in close proximity to "weary") and searching for specific phrases.

* date-range searching for time-sensitive information, which is very useful for serious researchers

Other important features of search tool interfaces include specifying the domain or range of information over which the search executes, the kind of data to be included (e.g., press releases), and metadata such as author, title, etc.

Avi recommended that search instructions be on the same page as the search tool, so that users don't have to search for help on how to search. Likewise, it's a good idea to offer simple versus advanced search tools, with the most popular options included in the simple tool. While each site needs to strike a useful balance between simplicity and functionality, in many cases less is more. In any case, usability testing is recommended to fine-tune the design.

Avi's next subject was the design of the search results page, which is just as important as the design of the search tool itself. The results page should have a look and feel which is consistent with the rest of the site, so that users are reassured that they're still in the same site; if not, users may become confused or think they got lost. Moreover, the widgets for navigating the results should conform to UI standards and be in a standard location, such as the upper-left.

Various kinds of meta-information about the search results are very important, such as:

* how to navigate the results

* options for sorting results by various criteria, such as relevance, recency, author's name, etc.

* the number of items found, and the number of pages of results

* what page of the result set the user is currently on

* the key word count for each item retrieved

Related to the above features, Avi argued that follow-up, incremental searching for a subset of the current results is very useful, and may work better than trying to specify the ultimate target on the first try. Moreover, without incremental search, the user has to return to the original search tool to refine the search.

In cases where the search fails to return any matches, with users being frustrated as a consequence, she advocated providing extra support. This support includes navigational help, an explanation of why there were no matches, which words were not matched, and the option to search again on the same page, so that the user doesn't have to find the search tool again. Since many searches fail because of mispelled words, an option to spellcheck the search terms can be very helpful. In this regard, Avi showed a variety of examples of how failed searches are handled, including the American Psychological Association's site, which says "0 items returned" but not much else, and is therefore a good example of inadequate feedback. She pointed out that any instructions at this point must rely on the user's faith that the action will prove worthwhile, so it should be explained as thoroughly as possible in order to reassure the user.

An obvious aspect of the search experience that needs good design is the actual list of search results. Avi's rule of thumb here is that the more information, the better -- she's a proponent of full disclosure. The result set should include the page title (linked to the page), a meta-description or page summary, the date of publication (rather than the date last modified), the URL, which keywords were matched, how many keywords were found in various parts of the item, and so on. If a large file needs to be accessed, the file size should be included to alert users about the time required to download. Ideally, all of these features should be standard for results lists.

In addition, there are other kinds of results that can benefit from added features. For example, specially formatted text such as poetry, spreadsheets, and train schedules should be properly formatted. As another example, the PBS site shows a thumbnail picture of the contents of each matching file, so that the user can preview it before retrieving the complete file. In any case, the order of the results is very important, and result sets should preferably be ranked by relevance, though this is very hard to define and measure.

Regarding the overall quality of search results, information retrieval theory identifies two contrasting search strategies:

* recall -- getting all the information on a topic, which may include some irrelevant information (false positives), and

* precision -- getting only the relevant information, with the possibility of missing something useful (false negatives)

Since these two criteria have an inverse relationship, there is always a trade-off that has to be made when deciding which strategy to pursue in a search. More generally, the ultimate quality of search results depends not only on the search strategy and the quality of the search engine, but also on the content and organization of the information space being searched, since the search tool interacts intimately with the data being searched. Hence, it's arbitrary to blame poor results on just the search tool alone. Incidentally, human searches are not always so accurate either; people can make big mistakes by misinterpreting someone else's search criteria.

A significant problem with searching is the existence of extended character sets (Mac versus Windows), non-Roman character sets, and multiple languages; and there are no easy solutions to these problems. A related problem is "stopwords" such as "a," "an," and "the," which are ignored in searches and removed from indexes. However, these words can be crucial to queries if you're looking for a particular song title, for example. Another potential problem derives from "stemming," which means indexing and searching on root forms of words rather than exactly what the user entered (e.g., searching on "paid" retrieves "pay"); this can help or hinder, depending on the user's goal.

Finally, Avi sketched a few principles of site and information architecture, since the site provides the wider context for the search mechanism, and a coherent, consistent site structure makes users more comfortable with the search process.

Table of Contents | Previous Page |