AllThingsD » contextual search

Qwhisper Is Looking to Solve Social Search With a Dose of Uber-Geek

Drake Martinet — Fri, 16 Sep 2011 18:01:15 +0000

Sometimes a start-up’s product is pretty, sometimes it’s from famous founders and occasionally it’s dead simple.

Qwhisper is none of those things — in fact, it’s barely even a product at this point. But its team of founders are attacking a devilishly hard problem.

The company and Web app of the same name attempt to search and categorize social media updates with an accuracy that even the sector’s giants have been unable to deliver thus far.

“Search for social is really tough. When someone mentions Mars, you don’t know if they mean Mars the planet, the god, Bruno Mars, the rover, or the candy bar,” said Qwhisper co-founder Eldar Sadikov. “With Web pages, there are all kinds of context clues to help you figure things out, like links and other data. Social content is just so much shorter — you have to be very sophisticated to [make sense of it].”

What that means for us avid Twitterers is that, as of now, searching for a category of tweets is not a useful endeavor — and forget about searching for tweets about a simple but amorphous topic such as “popular music.”

But Sadikov’s Qwhisper, which is in private beta, makes use of some new search algorithms to reorganize a user’s social streams.

Its founders claim the search and sort technology of Qwhisper can reliably deliver tweets to the user based on a topic, category and search term.

So, how does Qwhisper do it?

Sadikov made an attempt at outlining just how complex it is for a computer to make sense of a stream of single tweets:

“You need much more sophisticated natural language processing technology [for social] than what is needed for Web pages. [The system must] understand words like “lol,” “cuz,” “gonna,” “gotta” — because there is so much colloquial language in social content, compared to Web sites.”

Only after dealing with those problems, which are in themselves complex enough for several research papers, can Qwhisper layer in the really complex processing to answer such contextual questions as: What does this person do normally? And, what does that person normally talk about?

But every start-up with a search component boasts custom algorithms, so why should users be confident that Qwhisper’s are superior?

Qwhisper is touting the company’s intellectual pedigree.

Sadikov and some of the other co-founders left their PhD programs at Stanford’s InfoLab to start Qwhisper — the same InfoLab where Larry Page and Sergey Brin developed some of the early parts of Google.

Sadikov also spent time at Google, where he worked on building an algorithm for organizing small sets of words together in contextually relevant groups.

Not too long after, he gathered a group together to launch Qwhisper using some of the same concepts.

If Qwhisper or the engine that powers it proves successful, the consequences could be far reaching.

Delivering tweets and other social content in contextual channels could mean a whole new class of applications — and advertising — all built around social content.

But complex graph-modeling and multivariate algorithms aside, the litmus test for Qwhisper will be simple user interaction.

“Ultimately, if I post something like ‘saw inception last weekend – amazing,’ the system needs to recognize what that is about … even though it says nothing about movies or genre,” said Sadikov.

I caught him and one of his co-founders, Montse Medina, at the recent Stanford StartX incubator demo day to talk more about Qwhisper:

[ See post to watch video ]

May Search Metrics: Google Losing Share or Gaining It? All Depends on How You Look at the Data

John Paczkowski — Fri, 11 Jun 2010 12:30:39 +0000

Getting an accurate read on the tides of search market share these days is no easy feat given the interface changes being rolled out by the major players.

Consider comScore’s May search market report (see tables above; click to enlarge), which shows Microsoft’s Bing and Yahoo gaining share, ostensibly at Google’s expense. For the month, Yahoo gained 60 basis points for a share of 18.3 percent, and Bing 30 points for a share of 12.1 percent. Meanwhile, Google’s (GOOG) share declined 70 basis points to 63.7 percent.

While those numbers seem straightforward enough, they really aren’t, given some of the navigational ploys Yahoo (YHOO) and Microsoft (MSFT) have been using to boost their numbers. Slideshows and contextual search links have helped raise the share of both companies, but they need to be backed out of the data to get an accurate view of the sector, as J.P. Morgan (JPM) analyst Imran Khan explains.

“User interface changes continue to cloud the picture,” Khan said in a note to clients. “Google, Yahoo! and [Bing] all made notable changes in April and May, according to comScore. As such, numbers may not be directly comparable to past months. On a reported basis, Google lost ~70 bps of search share in May vs. April, while Yahoo! was the biggest gainer, up ~50 bps. Excluding the impact of all adjustments, Google gained ~30 bps of share, Yahoo! was down ~25 bps and Microsoft was flat.”

In other words, the numbers here do lie and will continue to do so until measurement outfits like comScore (SCOR) account for context-driven searches and slideshows. ComScore plans to do just that.

“[The] continued evolution of search and emerging innovations in how it is used to enhance user experience, calls for a thoughtful review of how we classify various types of searches, count them and report them,” Cameron Meierhoefer, comScore’s executive VP of analytics, wrote in a blog post released along with the company’s latest search stats.

“We want to ensure that we provide comprehensive and flexible measurement that meets the needs of the various constituencies in the digital marketplace,” Meierhoefer added. “As our thinking evolves, we will include relevant stakeholders in the discussion and clearly communicate our thinking and rationale to the marketplace.”

The timeline? “While we will maintain the current method through the end of the second quarter to avoid reporting disruptions, we will aim to implement proposed revisions in the third quarter, ideally starting with the release of July data in the first half of August.”

Social Networking More Popular Than Email, More Profitable Than…Er…Um

John Paczkowski — Mon, 09 Mar 2009 16:55:19 +0000

Hard to believe, but social networking has eclipsed email in popularity. The latest Nielsen survey (PDF) found that 66.8 percent of the global online population spends time at “Member Communities,” a category that includes both blogs and social networks. That makes social networking about two percent more popular than email, with one in every 11 minutes online globally spent on Facebook, MySpace, etc.

That’s a lot of Superpokin’….

Apparently our self-righteous outrage over Facebook’s tweaks of its Terms of Service or claims that social networks are haunted by sexual predators has done little to dampen our enthusiasm for them. “Social networking has become a fundamental part of the global online experience,” said Nielsen Online CEO John Burbank in one of those “Fire: Hot, Bread: Good” announcements. “While two-thirds of the global online population already accesses member community sites, their vigorous adoption and the migration of time show no signs of slowing. Social networking will continue to alter not just the global online landscape, but the consumer experience at large.”

Unfortunately for the companies playing in the space, social networking hasn’t yet done the same for the advertising experience. Indeed, as Nielsen notes, “The current level of advertising activity on social networks isn’t consummate with the size–and highly engaged levels–of the audience.” Nor will it be, unless the sites figure out a way to exploit the personal data of their users without making them feel like their privacy has been invaded. Nielsen’s big idea for achieving that goal: trial and error. “New approaches to the ad model are required for this challenging and complex arena,” the company explains. “It will take time to work out the magic formula for successfully advertising in social networks. The diversity and personalised nature of the environment means standard ad models–such as contextual search and standard unit sizes–won’t cut it. Different approaches across ad units and ad inventory will have to be tried, involving a trial and error mindset.”

Trial and error, eh? That’s been Facebook’s strategy hasn’t it? And as I recall, it hasn’t exactly been working out too well…