About two years ago, I wrote an article called Why Search Sucks & You Won’t Fix It The Way You Think. In it, I explained various ways people have tried to make search “visual” and why those have largely failed. That’s mainly because “list view” or “10 blue links” still works for lots of search activities. But visual search has picked up some attention recently with new players coming in. Is visual search the future, where we’ll be flying through our results Minority Report-style? Maybe in years, but for now, I still see a lot of eye candy and no real breakthroughs.

I looked at Searchme, Viewzi and PicLens, all of which have been reviewed recently in various places. For each, I purposely went into them without reading any of the help information. As a result, I might be missing out on some cool features or capabilities they have. But then again, so too are the typical people these services hope to attract. No one is reading how to search at Google. Even fewer than no one will read how to use these places.

Searchme

At Searchme, when you enter a search query, immediately below the search box, categories appear. The exact categories change depending on what you search for. For example for dvd player, these options show up:

• luggage & bags
• software
• movies
• business news
• photography

Selecting one of the category icons, rather than doing a general “All” search, narrows your results to pages deemed to be relevant to that category. This is both cool and weird. Luggage as a “dvd player” category? It turns out that the results coming back by doing this, showing DVD player cases, were indeed interesting. But DVD players narrowed by the Martial Arts category? Not so compelling!

More important, most people probably won’t use these options. There’s a long history of users ignoring links like these around the search box, which tends to be like a black hole that sucks users in.

How about the main attraction, the “visual” search results:

Sure, they look cool. But as is often the case, “cool” doesn’t mean useful. The screenshots of each page are fairly hard to read — in some cases, impossible to do so. That means you’re having to judge whether a site matches what you want almost entirely based on what it looks like. Consider how “useful” this is for Sears:

OK, you can get a description for each page if you know to hover your mouse over it. One will appear at the bottom of the image. But that leads to another flaw. Which is easier — to quickly scan 10 textual descriptions or to painfully click-click-click your way through a screenshot at a time?

The “stacks” feature is pretty neat. You can drag any screenshot to the top left of the page, where you can make a collection of search results. Of course, Microsoft offers this for image search. And they even have it on their own “visual” site Tafiti. Plus, over the years, we’ve had other search engines offer a drag-and-collect feature for web search. It still hasn’t taken off. I do think it’s a great idea, and it would be nice to see it come to places like Google. But Searchme Stacks is hardly a killer feature.

At Searchme, you can also narrow searches to just video or images. Here, the visual display is more compelling. But still, I think it’s easier for the searcher to review a lot of pages at once using the “old school” listing model.

Searchme is also slow. It takes noticeably longer to get back results than on a major search engine, and there’s no great payoff for that wait. The Flash-based service also crashed my Firefox 3 setup once. OK, Firefox 3 is new, and I’m not using the final release candidate. Nevertheless, none of the major search engines including Google have crashed it once.

Viewzi

Like Searchme, Viewzi is slower than a regular search engine. Worse, after doing my DVD player search and waiting for a response, when it came, I then had to decide what “view” I wanted to see:

The screenshot above only shows a slice of the many “views” that are out there. Working through them was somewhat overwhelming for me, and I write about search all the time. A typical person hitting this page is being asked to decide if they want:

• Video: Blinkx, Veoh, YouTube
    
• Site Information: Alexa, Delicious, Google, Summize
   
• 3D Photo Cloud: Flickr
   
• 4 Source: Ask, Google, Yahoo & Microsoft MSN Live Search
   
• and more…

Assuming you figure out the choice to make, it gets worse. Consider if I just want to see the results the four major search engines at once:

You don’t know much of anything about what’s showing up. After studying the page (something a typical user isn’t going to do), I realize that the first result in the top left corner is showing up because it is listed on all four major search engines (thus the “stacked” look to icon, a page for each of the search engines it is on). But the purpose of also showing the lower row? Plus, I kept having a problem where if you click, sometimes I’d get a larger preview of a page but other times I’d be jumped out to it. This is a user interface nightmare.

Maybe if I were to dig further into some of the “views,” I’d find an example of a compelling reason for this display. Something like Album View (check it out for The Weepies) is intriguing. Viewzi might have more success if it focused on a few particular views where visual display is really useful. But right now, it’s just a search engine relying on visual gimmick that people don’t need.

PicLens

Quick history lesson. How many search engines have made it as a success by requiring people to download software before they can search? Zero. That’s bad news for PicLens, since it assumes it can swim against the tide.

Let’s assume you do download PicLens. The next mystery is figuring out how it works. I fired it up on Firefox, then after Firefox restarted, I was clueless what to do next. Remember — I’m being a typical user who doesn’t read the help files.

Eventually I noticed a new icon next to the Firefox search box in the top right-hand corner. Clicking on that caused an entire new window to appear, sort of freaking me out.

I ended up having to go to the demo page to better understand how the tool worked — which to me is a failure on the search usability side.

PicLens probably produces the “prettiest” visual results of any of the tools I’ve covered. But again, I couldn’t see that doing a “visual” search for “DVD player” on Amazon using PicLens increased the search experience. Rather, it just slowed things down.

Save Me From The Future

Clearly, I wasn’t impressed with the visual offerings. It’s not that I don’t like cool things. It’s just that there should be a reason to display things visually. It shouldn’t just be an excuse to look different. Moreover, text IS visual — and the textual display metaphor continues to be used largely because it does work. But for particular types of searches, a more graphical display can make sense. Tying local search to maps is a classic example of this.

I am looking forward to a more visual search future — but don’t make me fly through results unless it helps me to do so!

When you start typing a query into the search box at Yahoo, you’ll see a dropdown appear under the search box with some suggestions predicting queries that you may want to see Web search results even before you finish typing.

But presently you only see those suggestions for Web search results. I wrote about those Yahoo search suggestions in Predictive Queries versus Unique Searches.

It would be interesting to see suggestions from some of Yahoo’s other databases appearing, such as image search or local search.

A couple of recent patent applications from Yahoo, related to the “predictive queries” patent filing, explore showing how the context of a search and historic search patterns may cause suggestions from other search databases.

One of them also describes how a number of these vertical results could be presented together on a search portal page in response to a completed search.

Does this patent application describe what is going on at Yahoo’s alpha search? It just might.

The Search Equalizer

The first Yahoo patent application is the Search equalizer (20080016034), which determines a relevance score for possible results in different search contexts, such as Web, images, video, local, and shopping results, and shows them as suggestions in a dropdown.

This search equalizer might suggest predictive queries from different vertical searches that might contain the most relevant results for a specific query, while a searcher is typing the letters of their query into the search box.

It might also provide suggestions that are “related” to predicted queries. For example, a searcher starts typing in “interna” into the search box, and it shows a prediction for “international trade” and also for topics that might be related such as “GATT, WTO, UN, US trade policies.”

Tabs, buttons, or links might also be shown, which a searcher could select to see suggested queries from for query sets from other search verticals, such as “Web,” “Images,” “Video,” and “Shopping.”

Right now in Yahoo, when you choose a type of search other than a Web search, you don’t receive suggestions for those other search types.

While this process would show predictions from different databases, depending upon which button you choose, it might also provide predictive queries from the different databases in the dropdown regardless of which button you choose, based upon which databases it might think contains the most relevant results for a possible query. The dropdown predictive suggestion would include a tag showing which vertical that suggestion was from.

Thus, relevance scores may be displayed for Web based searches and for searches over one or more particular verticals, allowing users to compare and choose between them. This has the benefit of promoting effective and efficient selection of search contexts for queries that have a higher probability of returning relevant results.

The Query Categorizer

    The second Yahoo patent application, the Query categorizer (20080016046) describes in more detail how the different verticals are selected, based upon “historical search result selection data for similar sets of query terms.”

    That historical selection data for a set of query terms includes the number of times that someone might have “selected a search result from that vertical after conducting a search based on that set of query terms.”

Yahoo Universal Search?

While this patent application tells us that it is related to the “predictive query” patent, it also describes how it works in the context of showing a full page of search results after a completed query has been typed in and submitted by a searcher”

According to techniques described herein, in response to a user’s submission of a set of query terms through an Internet search engine’s user interface, the Internet search engine automatically ranks multiple verticals based on the estimated probabilities that those verticals will contain content that would be of interest to the user.

After the verticals have been ranked, a set of the highest-ranked verticals is automatically selected. Search results from each of the verticals in the selected set are presented to the user.

Thus, in one embodiment of the invention, the user is able to obtain a sample from several different verticals–and more specifically, the verticals that are the most likely to contain content in which the user is interested.

The frequency of searches from within the different verticals using specific sets of queries may determine whether search results from those verticals appear within a set of search results conducted within the default main search box at Yahoo.

The searches through the different verticals might include searches for “related query term sets,” so that for instance, the Yahoo document tells us that a search where the main query is “Britney Spears,” might also include a secondary query for “Baby One More Time” or “Kevin Federline.” Those choices might depend upon the time period of the search, and which of those related queries might be searched for more frequently.

Presenting Search Results from Selected Verticals

A number of different ways for presenting search results from different vertical searches are described in this patent application, including:

1. Segregating results from different verticals into different areas on the search portal page.

2. Possibly presenting those segregated results in an order reflecting which of the vertical results ranked highest.

3. Displaying some results in a nonverbal manner, such as thumbnails for images, or a thumbnail frame image for a video.

4. Search results from multiple verticals could also be shown in applications other than a browser, such as an email reader or news reading program, or a word processing application. These results might be shown in a “pop-up” window generated by one of these applications. The results might also be generated from a set of implicit search queries selected by the application.

5. The results from these different verticals could also be presented on a smaller screen, like a Web enabled phone.