A couple of weeks ago Yahoo debuted a beta version of its Search Direct technology, which was seen as a competitor to Google Instant in that it shows search options as users begin typing in a query. There were also questions raised among those in the industry about how this relates to Yahoo’s search deal with Microsoft. There was less chatter about its semantic footprint, but that’s a question worth addressing as well.
So the Semantic Web Blog took these questions to Raghu Ramakrishnan, chief scientist for Yahoo Search, whose background includes expertise in data management and mining, the cloud, and, of course, the semantic aspects of search. As Ramakrishnan sees it, Search Direct exemplifies Yahoo’s search slogan of providing answers, not links. Meaningful answers, whatever form they take, are the new direction in search, he says, and Yahoo wants compete “at the next level where technology is young and there is room for differentiation and market opportunities in being the best.”
The Microsoft deal, he thinks, hasn’t been as well-explained as it might have been – link delivery is the part that’s been outsourced to Bing. “So what is Yahoo doing over and above this? The reason for the [Microsoft] deal is that in our opinion the most exciting future in search has to do with things that don’t really center on those blue links themselves,” he says, though such links still have their value for navigational queries, for instance. But the action and the future of search “is all about understanding what the user is looking for and presenting it in the most accessible, usable, actionable form we can.”
So, even pre-Search Direct, Yahoo has gone about crawling the web to go beyond what Microsoft adds with its ten blue links, to analyze content and blend it with what it has in its own extensive content repositories from the media side of its house, and from there distill things down to a presentation it thinks the user wants to see. “So where does Search Direct take us? In order to do what I said, to provide users with answers rather than links, there are different puzzle pieces,” explains Ramakrishnan. “First, you must understand data at a semantically rich level….and that is something we’ve been working on and surfacing already; it’s the whole web of objects stream of work.” That web of objects includes ingesting feeds from site owners who submit structured data to Yahoo Search for inclusion in the Yahoo Index, which utilizes its SearchMonkey-supported vocabularies for annotating pages with metadata.
The accompanying piece to that deep and semantic understanding of content is equally crucial: Understanding the user’s intent. “We want to help you frame your intent quickly and along the way see what else is quickly interesting to you,” he says. Search Direct, he says, pushes the needle, especially when users are in a mode where they don’t quite know what they want, or perhaps they do know but are open to exploring, and the opportunity is to help guide them to related information that they could find interesting. In that respect, he likens the technology to encouraging the more exploratory “accidental tourist” phenomenon, and building a blend with search’s more traditional activity-directed focus; there, Search Direct can be the glue between the understanding of content on the back end and the user intent on the front end.
Search Direct Takes a Stab
“As you type we try to take our best stab at what you might be looking for” – type Julia, for example, and Julia Stiles, Julianne Hough, Julia Roberts, and so on appear as choices, hence the comparison to Google Instant. But Search Direct, Ramakrishnan says, is more sophisticated in that you can hover over all the different interpretations on the left side of the screen and on the right side of the screen, right within the search box, gain a further understanding of what a particular suggestion might lead to. That summary of results (leveraging crowd-sourced content from its repositories and its partners) further details what is behind each search “stab” including, as appropriate, photos, ratings, and Twitter results.
“At this point we are showing you a semantically well-packaged result. At this level you can start to compare the two in terms of coverage, depth and quality,” he says. It’s fair to say that all three search engine majors are trying to get to serving users semantically richer results, he acknowledges, but “Search Direct on canvas in real time shows you the next level of refinement for candidate searches on that list, so you see more depth. What is effectively going on is that we work very hard to promote user search activity into a more semantic level.” Click on the search term and the full panel of results follows.
Underlying types of objects, organizing content in a semantic way, having a richer understanding of user intent, and semantic matching of intent to content to provide users with answers vs. links is the goal, and “all these things illustrate why we think there’s so much exciting potential over and above what we get through our partnership with Microsoft,” he says. “All of this comes from us.” Currently Search Direct’s interpretation of answers is broad, not necessarily delivering a factual answer to a question, but Yahoo’s not excluding that, either. Nor is it ruling out how Facebook’s Like can be a social signal that can be useful for ranking its answers, something it’s playing around with now. “We see this as a long road and part of the milestones are to handle a broader range of questions including factual questions,” Ramakrishnan says.
Continuing to exploit semantics for search will be the game changer over the next few years, he says. “I think all the different pieces we talked about need to mature – considering the understanding of user intent, finding a user experience that lets them explore while refining that intent in a graceful and quick way, giving high-level answers that are actionable,” Ramakrishnan says – building further on ideas such as having a movies search result generating a search engine prompt to add the title to your Netflix queue (which Yahoo Search already enables). Equally important: Considering the shift of search activity to mobile devices where a premium gets placed on well-aggregated results and on avoiding link-clicking in bandwidth-limited settings.
Search Direct already “takes a giant step to richer answers and not simply links and we are pushing aggressively in this direction,” says Ramakrishnan. But this is no overnight project. When it comes to the future of search, “you see elements now but getting this to be broad and deep so that the bulk of searches have a satisfying high level experience takes a long time.”