Marissa Mayer, noted search guru and Google's VP of search and user experience, has a long and insightful post up on Google's search blog. She shares a brief personal search ethnography she conducted one Saturday, then explicates by breaking down some things Google may be working on to make search more ubiquitous, personal and multisensory.

Ms. Mayer goes on to describe her ideal search engine as "your best friend with instant access to all the world’s facts and a photographic memory of everything you’ve seen and know. That search engine could tailor answers to you based on your preferences, your existing knowledge and the best available information; it could ask for clarification and present the answers in whatever setting or media worked best."

Current Research is Driving Search UI

1) Search UI research method: task success measures
Task-based studies give the user a set of structured tasks to perform, while researchers track the results using cameras, eyetracking, note-taking or other methods.

Yahoo is concentrating on the context of the search, and moving the user from "to do" to "done". This involves learning where search fits into various tasks, and making the right sort of information available for each context. http://www.seroundtable.com/archives/018046.html

"Yahoo is taking a task-based approach to its search strategy, improving results to focus on the user's task stage, according to Andrew Tomkins, chief scientist for search at Yahoo" & "the next generation of search will be about understanding the task a user has in mind and changing the way search operates to get those things done." http://searchenginewatch.com/showPage.html?page=3628767

My take: Yahoo's approach ties in closely to naturalistic ideas of information availability, wayfinding and information scent. The methods we use to complete tasks, or to browse information, are still only partially understood at best. In some ways the experience of information discovery is almost like a dream where one minute we are taking a meal and the next hovering over a pond: we perceive something, and then just as we begin to comprehend it that thing may evolve into something else and take us somewhere which seems unrelated. Yet upon careful analysis we see that the first thing and the second were related in our consciousness. Perhaps we have a favorite restaurant that overlooks a pond - hence the connection between eating and a small body of water. Likewise, we still have much to learn by tracing and understanding connections among the seemingly serendipitous steps a user takes to find information online. Web search should strive for improved 'flow': present-day systems have a great utility, but they generally fail to inspire. Google's holiday logo drawings and Live search's new interactive hotspots are examples of serendipitous interfaces which can be further leveraged through personalization.

2) Search UI research method: eyetracking/heat maps
Eyetracking uses devices that track user eye movements as they attempt search tasks, usually in a lab setting. Heat maps are graphic renderings of where the users' eyes tend to focus on the page. These maps created using the data from eyetracking studies.

Assumptions about page layouts are evolving as researchers study users of pages with more varied layouts. "In the past, the definition of SERP [results page] real estate was fairly static: top and to the left. In the future, it seems it will be far more difficult to define." http://searchengineland.com/070921-070852.php

Another study found that eyetracking data supports current efforts to know more about user intent. "There are five specific patterns consumers tend to use depending on where they are in the sales/educational cycle" http://www.marketingsherpa.com/sample.cfm?contentID=3152

My take: humans have evolved to make the most of sparse information, creating a gestalt (whole) out of bits and pieces. Most researchers, when applying heatmap information to the page, assume that only the hot areas really matter. Yet, the periphery creates context and atmosphere. One possible approach is to array categorically related material in peripheral areas. This will help the user (perhaps subconsciously) to become more aware of concepts that may be related to the search. Just as ads may be effective even though they are never clicked, images and phrases placed peripherally in a reasonably clean layout may have a profound, if secondary impact, when compared the treasured "top hit". Further experimentation may uncover a better use for secondary results page areas than just a set of very similar results. Certainly the success of contextual advertising suggests there is more going on across the page than we might initially think.

3) Search UI research method: modeling expert behavior/expert systems
The expert systems approach attempts to reproduce knowledge in the form of very sophisticated heuristics (rules of thumb). Classic expert systems studies used chess as an research subject: expert players were found to have internalized 'chunks' of knowledge which gave them a superior ability to look ahead and respond in the game, when compared to the brute algorithmic strategies typically employed by a less experienced player.

One group is exploring online behavior modeling which would leverage groups of users with common profiles. "Imagine a world where your computer would generate a profile, a meme map about you based on your interactions with the web and refine your experience based on this map." http://youlicit.wordpress.com/2007/09/27/expert-systems-personalized-recommendations/

Steve Spalding envisions systems whereby "search engines would aggregate and generalize these user profiles and use them to personalize and enhance search". http://howtosplitanatom.com/news/how-to-explain-expert-systems/

My take: Search engines will begin to rank users by their effectiveness at finding what they are looking for, and the results of expert searches will be linked to their queries. Less experienced searchers will then be more likely to find what they are looking for earlier in their search path. For example, an expert searcher looking for "Las Vegas Hotels" may gravitate quickly to a reputable local guide with hotel reviews. Knowing this, the engine will subsequently rank that guide's pages higher than it otherwise would for queries related to "Las Vegas Hotels". Later users are then able to benefit from the expert user's ability to rapidly discriminate among sites.

4) Search UI research method: ethnographic field studies
Ethnographic study relies on rich qualitative data collected in the field, often in a naturalistic manner. Data may take the form of interviews, journals, film, etc.

Yahoo's Life Series explored relationships between life events and search habits. "people experiencing such events share certain common characteristics including a desire for trusted information sources, heavy reliance on research prior to making decisions and increased time spent on research during their life event." http://docs.yahoo.com/docs/pr/release1267.html

Opinion leader findings appear to drive much of Yahoo's current efforts (including UI-related initiatives such as Search Monkey). "Brand advocates are adventurous opinion leaders who are socially well-connected, express their opinions and viewpoints and continually discover new content online." http://createwithcontext.com/media/yahoo-summit.pdf

My take: Yahoo is using rich qualitative data to uncover the dynamics behind how people search. This highlights the present difference between Google and Yahoo. Google is focused on having the best algorithm, and being able to accomplish the most using the largest data set while using the least reasonable amount of processing. Google has more or less perfected the mechanistic model of search, and it shows. Yahoo, on the other hand, is chasing game-changing innovation through a dual effort: allowing radical collaboration (SearchMonkey, BOSS) and gaining better understanding of the human side of the search equation. Yahoo's approach is human-centered, and resembles that of the successful inventor: discover a market need that has not been addressed, and then find a way to fill that need.

permalink

New engine Cuil launched recently to an enormous flurry of press, and mostly negative critical notices. A great deal has already been written about problems with Cuil's results (an excess of "nothing found" pages, mismatched thumbnails, etc), but if they can straighten those out I think people will begin to see that they are onto some good ideas.

Cuil seems to be a meaningful step toward a search engine that does a better job of understanding user intent, and offers a UI and categorized content meant to serve that intent.

A concrete example of this would be when a user searches for a very general query like "Olympics". A quality-ranking search engine will produce the 10 most 'popular' or 'relevant' pages, even if all 10 deal with one subtopic such as the Tibet controversy. But a category-driven engine like Cuil will offer pages and refinement suggestions for a variety of related categories: official Olympic sites, travel to the games, athlete profiles, etc. This should make the results more browseable, and satisfy more users with the first page of results.

Cuil's Our Philosophy page explains their approach in more depth.

Naturally, Cuil's UI is heavy on features that show off their categorization focus. Tabs across the top and a dynamic menu on the right give suggestions for refinement and exploration. For each query, the content is different in the tabs and the menu, but it's not clear to me what guides the choices.

Another notable UI feature is a columnar results layout. The results look rather newspaper-like. I find it easy on they eyes, but might worry about missing something while scanning the results. Giving users the option to select either 2 or 3 columns is a nice touch.

Tag Clouds

However, tag clouds are now being used in a rather elegant way by a couple of search engines. Collarity and Quintura use a small cloud of terms to present possibilities for topic refinement. A search for "Las Vegas" on Collarity yields a dozen suggested refinements ranging from caesars to guide. Quintura offers a dynamic tag cloud, with terms flying in and out as the user selects or deletes refinements.

Offering suggestions after the first query is nothing new, of course, but the tag cloud seems well placed here. The user can see at a glance the relative popularity of the suggestions, and a cloud feels more inviting than an ordinary list. Given that users have been slow to adopt query refinement mechanisms, this may be a small but meaningful improvement in search UI.

permalink

Results Descriptions and Powerset

Microsoft plans to use the Powerset technology to make sure results descriptions match the pages they reference. One shortcut that might be immediately helpful would be to compare the relevancy of the site's home page description with the description of the deep page and use the site's description if it is a better match or generally much more coherent.

I personally like Google's deep descriptions which show the chunks of text my keywords appear in. But it would be interesting to learn how many users find that helpful, and what trade-offs might make sense for certain types of queries and users. Perhaps the intentions of some searchers would be better served by using the webmaster's page description, or the first paragraph of the page. Power users might enjoy having the ability to change the type of description they get by default. With a dynamic technology such as AJAX, any user could easily toggle between descriptions such as "where my keywords appear" and "first paragraph".

permalink

Browse

Of course, what we usually mean by browse is to traverse a directory. Yahoo is sitting on a unique and nearly priceless resource -- their famed directory -- yet despite being a category leader they treat the directory like a poor relation of search.

I believe browse has a bright future. For one thing, most of the popular social applications emphasize browse navigation of one sort or another - and social media is a fast-rising source for content discovery: http://searchblog.tamar.com/2008/06/social-networks.html

Aside from this site, I maintain a niche web forum. Not long after I developed the forum software, I added a topical refinement mechanism for the master thread index, calling it "browse" because that is a term people recognize for exploring links by topic. I was surprised by the immediate popularity of that interface, which has always remained strong. I also have a search mechanism available, but the browse vastly outpulls search.

So browse remains useful, and used, but where does it go from here? I believe the user needs to have more of a sense of place within the directory. The top page is probably not the optimum starting point for most users. A personalized directory home page that has integration with social media is probably the way to go.

The greatest improvements yet to make in web browse may be improvements to the directory hierarchy -- also known as the ontology. Perhaps the killer app of tomorrow is a directory ontology which has been refined by user data (both user-supplied and drawn from click path analysis). But instead of hosting all the link data (the traditional Yahoo directory model), links and content are drawn from myriad sources (Google API, Wikipedia, Flickr, YouTube, RSS feeds). Or maybe the ontology is simply made available for mashups.

permalink

Viewzi is a recently released engine featuring a query parser that suggests "views" for the user's results. These views feature domain specific results and unique interfaces, often emphasizing visual features such as scrolling thumbnails. Some views have a widget feel (MP3s, for instance, are playable right in the results). Viewzi now has 16 of these views, for such things as celebrity photos, weather and video. They plan to open their API up to developers who want to create more views.

This is a bit like Yahoo's SearchMonkey initiative, but each view alters the engine's entire results interface, rather than just a slice of the results.

Viewzi seems like a nicely thought out combination of query parsing, domain-specific search and an open API. I will be curious to see how they manage with, say, 100 views. Even selecting from the 16 views they have now can be a bit tedious; the logic that presents the views in order of utility based on your search seems somewhat hit or miss, but should benefit greatly from query log analysis.

Interrogative Search: beyond "did you mean"

To some extent, this happens already, and the user is growing accustomed to it. Google's "did you mean" interrogative has yielded some strange results at times, but it is certainly widely appreciated.

According to researcher Amanda Spink, "stimulating users to talk with someone or thing (agent) about their information problem helps generate terms and look at the results for additional terms". Long Tails and Short Queries

Interrogative interfaces might prove helpful for disambiguation and also for query tuning. Search logs are full of examples of common queries which could be improved by tuning (phrases which should be placed in quotes, phrases which would benefit by a proximity operator, etc). An interrogative system could make suggestions which are likely to be of immediate benefit, and would have the side benefit of teaching the user query construction tips which might otherwise remain buried in help files.

Combining an interrogative query tuning mechanism with live search might also produce some useful results. Live search display allows the user to "try before they buy", and while interacting with the engine a user could examine the potential of query refinements without necessarily losing their progress.

An extreme example of an interrogative interface is the 20 questions game at http://www.20q.net/ which is described as follows: "users log onto the website and play against an artificial intelligence (A.I.) foe. Players think of an animal, vegetable, mineral, or other object and 20Q guesses what the player is thinking in twenty questions or less. And, the more people play, the more the game 'learns.' 20Q.net ... played its 44,000,000th game in September 2006". But is this sufficiently searchlike to be of interest?

What if the same principles were applied to information retrieval? We wouldn't want to answer 20 questions for each query, of course, but if we could get great gain from three or four questions then we would probably learn to love it. And if a system showed continuous improvement, it could be trained by the early adopters and that expertise could be leveraged by novice users in later versions. For instance, a disambiguation system would develop a large library of popular disambiguations to present.

In web search we have seen a series of eras in which one search engine has dominated, and then declined: a Webcrawler era, an AltaVista era, a Yahoo era and a Google era. In each case, one engine has emerged as clearly superior and then gradually lost some (or most) of its luster due to lack of scale, webmaster spamming, or some other creeping issue. How refreshing would it be to have at least one engine for which continual improvement was practically guaranteed?

Note: at one time I felt that DirectHit was that engine that would continue to improve forever. While it did not use an interrogative method, it was designed to leverage user data. Sadly it did not survive as an independent entity.

permalink

Wikia has rolled out several social features, including the ability to share related keywords and make edits to search result fields such as title and summary. Users can also help build out the richness of results by adding deep links and images.

Alt Search Engines has a good summary: http://altsearchengines.com/2008/06/03/wikia-announces-wiked-new-search-features/

Wikia have opened up aspects of their UI for discussion, and maintain a mailing list for that purpose. http://lists.wikia.com/mailman/listinfo/search-ui

Thank you for visiting Search UI, a site dedicated to trends in user interface for search. In coming weeks, I plan to add regularly updated news content, commentary, interviews, reader discussion capabilities, an extensive search ui glossary, and other resources.

The next several years should bring extensive new choices in search interfaces, and may even redefine what we think of as "search". Search UI will be there and I hope you will be too.

Sincerely,

Chris Mitchell