Communications of the ACM

Search engines have become a part of the global culture, reaching a vast and diverse audience.

The challenges we face today include: How much better a search engine can do by knowing more about the desired task? What is the best UI to get this extra information? Given the benefits of extra information, it seems likely that we are entering an era of broad experimentation in UI to facilitate the search experience.

Pitkow et al. looks at the context as well as the personalization of search. Context includes simple things such as the country or language of the user, but also more complex clues such as the location of the search box (the page it is on) and the list of recent searches. Personalization takes this capability further by examining the historical search tasks of the user, or even the user's vocabulary. For example, the word "java" means different things to different people. Deducing the primary meaning for a given user will improve future search results. As with all forms of personalization, there must be a balance between improved productivity and privacy. For example, which searches would you like the engine not to remember?

Mowshowitz and Kawaguchi argue that like other media, search engines have a bias in what they present. As engines become a key source of information, this bias must be taken seriously. It should at least be understood, if not reduced. Although this area is young, there are strong reasons to encourage discussion about bias. For example, there are at least three kinds of admitted bias in search engines today: advertisements, paid placement, and paid inclusion. Ads are the most visible and the most similar to other media; there is always an implicit issue about how the ad dollars affect the integrity of the publication (or engine). Paid placement is the practice of selling top placement in search engine results. The leader in this field—Overture—shows how much the buyer paid for the placement alongside the result. Although biased, the bias is made clear to the user. Paid inclusion is when content owners pay search engines for better coverage, but not better placement. For example, a large site may pay a fee to ensure all of its content occurs correctly as represented in the engine, rather than leaving it to chance.

To date, these forms of bias seem to have minor and sometimes even positive effects on the search quality. But it may not remain this way; therefore understanding bias will only help. The authors argue the best approach is simply to maintain a diversity of engines, much like reading a diversity of newspapers tends to provide a less biased view as a group.

The time is here to rely on search engines as the mainstream tool they have truly become. We hope the information this section provides helps to promote that shift.