These Things Matter » search

Search Suggestion Shortcuts

Sarah — Mon, 19 Oct 2009 06:28:23 +0000

Or, What Wolfram Alpha Does Completely Wrong

In my last post, Three Types of Searches, I proposed that there are three distinct behaviors of search, and that each behavior has its own optimal results formatting. Understanding these different search behaviors and thinking specifically about existing search tools and implementations of search suggestions has led me to this thesis:

Your site search should offer rich search suggestions if any of your users are using it for shortcut searches—and they probably are. This means not only should you make sure that you implement search suggest/assist in the first place (search suggest is the auto-complete drop-down that appears as you type in the search box), but it should also be more than just a basic text list. In fact, it should take formatting cues from the best practices of targeted search results. Does your site have entities—products, people, classes, etc.—that people might be searching for or returning to? Your search logs should confirm whether or not people are ever entering specific entity names into search. If yes, then you are doing your site a disservice if you have not implemented well thought-out search suggestions. Four points have led me to this conclusion.

First, search suggestions steer people away from the dreaded “No Results Found” page by helping users avoid typos or terms that differ from the site’s terminology. In addition, they may surface options the user didn’t even know were available. In this way, a search assist increases findability of items. (Side note: I think search assists could also be enhanced to improve re-finding. I like the way Firefox puts my previously entered search terms above the matching Google suggestions, for example. What if clicking inside the search box immediately opened the suggestions area—before I start typing—with the last five entities I viewed on the site?)

Second, if the user finds a match in the suggestions, clicking on it eliminates at least one page from the navigation process and gets the user to the optimal experience faster. Even if you’ve designed a nice shortcut search result module for every item, I would bet that your product or entity pages are a better experience. As long as your business isn’t in search advertising, skipping the search results page benefits everyone. In addition, it’s unlikely that your site has navigation pointing directly to entity pages. In order to browse to an item, a user would probably have to first choose the correct category from the nav and then find the desired item from there. Displaying entities in the search suggest gets around that problem by essentially enhancing your search box with an item-level navigation dropdown.

Hulu search suggestions

A good example of this is Hulu. Say I wanted to watch the most recent episode of The Office. If the episode isn’t linked directly on the homepage, I could try to see if I could find it under any of the main navigation items: Channels (which is really more like genres), Most Popular, Recently Added, Collections, Labs, Trailers, and Spotlight. Or, since none of those immediately screams, “Half hour NBC sitcom,” I could just search for the title. As I type ‘the office’ into the search box, I get the search suggestions shown above, which gives me the choice to jump not only to the show page, but also to immediately begin watching the latest episode. This search suggest allows me to bypass all browsing pages, the search results page, and even the show overview page if I want.

Third, rich formatting helps users distinguish between options. This is an accepted practice in targeted search results that makes sense to carry through here. In fact, I think it is a good rule of thumb to carry some of the formatting choices you’ve already made on comparison results through to the search suggest. Apple.com and Facebook both provide images and other information to make the options in their suggestions easily identifiable.

—

Apple.com and Facebook search suggestions with images

The Hulu example above also shows that multiple link destinations, when used carefully, can be very effective. Jakob Nielsen’s research findings support the use of larger dropdowns when they help users better see and understand the options. Let’s take another look at Hulu. Their current suggest drop-down is very svelte, but I’d be able to make sense of the options much more quickly if it had images.

—

Current Hulu search suggestions and a proposed enhancement (click for full size)

My proposed design includes the show images Hulu already produces, and changes the display to show the top 5 matches by popularity, rather than alphabetically. (I’m curious, did you know which show The Riches was without the image?) I am not saying every search suggestion must include an image, however. I’m simply encouraging you to consider displaying information in addition to the entity’s name (like price, for example) that would significantly help people recognize options.

And lastly, you have nothing to lose. As long as it doesn’t make the page unbearably slow, it is all just an added bonus to the search you already have. I am in NO way advocating removing any of your standard (or enhanced) search functionality. Obviously, plenty of the searches on your site won’t be for shortcuts, they’ll be open-ended searches that are more exploratory. Adding rich search suggestions won’t help those people, but it won’t hurt them either—hitting enter or clicking the search button is always an option.

And now for Wolfram Alpha. The thing about Wolfram Alpha is that it is ONLY a shortcut search. Each search results page is actually an entity page. Look at the search result for ‘health care,’ and you’ll see that there were actually three matches—health care as an industry sector, as a financial entity, and as a word—and Wolfram Alpha simply defaults you to one of the entity pages, allowing you to see and select the other options at the top of the page. Wouldn’t it be nice to see those options before landing on results page? Especially since the thing that is so cool about Wolfram Alpha is that it provides very different pages for each result type. But for some strange reason, Wolfram Alpha offers no search suggestions at all. They have a whole section of the site with sample search terms – why not populate the search assist with popular terms matching what I’ve begun typing? It would surface so many more potential opportunities to explore.

Currently there are no suggestions—just keep typing, hit enter and hope you get something good!

A proposal for search suggestions—popular matching results with clarification of result type

The suggestions don’t need to match against every possible result. If it pulled even from the top thousand popular terms, it would be a vast improvement. It would also be a clear differentiation from Google. Google was hesitant to widely implement search suggestions because they were afraid they might lose traffic volume from their search results pages. They clearly have no incentive to help you bypass them. (Google has however implemented a very nice search suggest feature within Gmail that you can enable in the Labs area.)

So to re-state: if anyone is using your site’s search as a shortcut mechanism, and they probably are, you should offer rich search suggestions. Of course, all of this assumes that you understand what people are looking for and that their search terms are already well-matched to your data. So spend some time with your search logs and definitely do a little bit of testing on your search suggestions design. I am looking forward to seeing and using a lot more rich search assists in the future!

Three Types of Searches

Sarah — Mon, 05 Oct 2009 04:51:24 +0000

I seem to have had the same discussion over and over lately here at Yahoo about site search. Yahoo has (or had, anyway) a high-profile web search technology and design team. So when we discuss these site-specific searches, talk always seems to circle back to the question, “How much should this site’s search resemble—or maybe even be—the Yahoo web search?”

On one hand, consistency and efficiency across the company are good, and the web search team has spent a lot of money and time optimizing their experience. On the other hand, sites such as Y! Music and Y! Sports have the luxury of knowing more context behind the user’s search requests than Yahoo.com does, and so more content-specific results seem like an easy win for users.

These conversations got me thinking quite a bit about search. Although search has only one user action—typing terms into a search box and submitting it to get results—the motivations behind that action can differ. As such, I think it’s wrong to try to find a single pattern for all search results. I propose that there are three distinct categories of search, and that each deserves its own set of patterns.

1 – Web/Exploratory Search
The first and most familiar type of search is the web or exploratory search. The user goals for this kind of search are “I’m looking for something/anything about [search term]” or “I’m looking for a specific [search term] and think [search engine] can find it.” The available data set (usually the entire known web) is vast and largely undefined. The user’s query is general, usually because s/he didn’t know what exactly to look for or didn’t think the search engine needed any more detail. The large number of results that are returned are formatted as web pages, mostly agnostic of the content those pages contain.

Google, Yahoo and others have spent a LOT of money and time optimizing the pattern for this type of search result. This is the reason why these companies’ search results all look virtually identical—compare searches for ‘UX’ on Yahoo!, Google, Bing, and AOL.

In the Google search results for ‘health care’ you can find pages containing articles, videos, local care centers, books, a web utility, the term definition, and more. Google changes the standard document result display, however, when they understand the content contained within the pages, such as local results, news, videos, books, etc., which leads me to the next category of search.

2 – Targeted Search
The second type of search is the Targeted search. The user goals for this type of search would be, “I’m looking for a specific subset of information on [search term]” or “I’m trying to find the optimal one of [search term].” This kind of search is possible when there is a single overall structure to the data because the user has explicitly selected a data type, e.g. books, flights, images, etc. There is usually a large set of results that are formatted to help users scan and compare the content on the resulting pages without pogo-sticking to and from detail pages. Good examples of this are Kayak, Amazon, New Egg, and sub-domain searches of the main search engines such as images.yahoo.com and shopping.google.com.

Obviously these results experiences should differ depending on the content they are searching. Much has been studied and written about optimizing these types of results, but it appears that even for the most popular data types, definitive patterns haven’t yet emerged (see video on Google vs. video on Bing). However, one widely adopted best practice is to surface important data facets (price, size, departure time, etc., often to the left of the results) and provide easy filtering of results based on those facets.

Google and Bing Video Results

3 – Shortcut search
The third type of search is the Shortcut search. The user goal in this instance is “I’m trying to find [search term] on this site.” Examples of shortcut searches are searching for a specific movie title on IMDb or a team name on ESPN. The data has a defined structure and the results set is very small—hopefully one exact match and fewer than 10 partial matches.

Interestingly, there is also overlap with the user goal I mentioned above for web searches, “I’m looking for a specific [search term] and think [search engine] can find it.” The user is searching for a single known or remembered entity, and is essentially counting on the search engine to produce it faster than browsing or even than using bookmarks.

Currently, most search experiences—both top-tier search engines and single-site searches—try to meet this search need by adding a custom result module at or near the top of the page. Notice the large Lakers module at the top of the ESPN results, for example. Or if you search for ‘Prius’ on the Toyota website you get a big module above the results with a nice picture and some price information. Similarly, the first Google result on a search for ‘Golden Gate Bridge Photos’ is a module that features and links to results from Google’s image search. In all of these cases the full results page contains a mix of either shortcut or targeted results and more general page results.

Shortcut results above the standard results on ESPN and Toyota

Additionally, many sites provide auto-complete functionality that is especially helpful for shortcut searches. This drop-down that appears as you type, containing matching popular or available terms, is called “search assist” by Yahoo and “search suggestions” by Google. Search assists help people avoid typos, discover possible terms, and sometimes allow them to jump directly to an item while completely bypassing the search results page, which makes them a true shortcut.

Where This Leads Me…
Understanding these three distinct types of search—web/exploratory search, targeted search, and shortcut search—has clarified my understanding of how results in different contexts should be formatted. It has also given me a framework for clarifying and making progress on our internal site-search discussions.

Additionally, I think that there is an opportunity to extend this differentiation into the design and implementation of search assists. In my next post, I will propose a new best practice for search assists, based on this understanding of the different types of searches. In the mean time, what are your thoughts on this? Is there a whole category of search that I’ve forgotten? I look forward to your comments and seeing you back here for part 2!