Amiad Solomon, CEO, Peer39
Chris Lamb, SVP Alliances, the OpenCalais Initiative, Thomson Reuters
Nancy Harvey, Executive Director, Wolfram Research Inc.
Paul Martino , CEO and Co-founder, Aggregate Knowledge

Pereira kicked off the panel by asking panelists to give a very brief statement of the problem with search as it stands today and how it can or should be "fixed" for users and enterprises. Chris Lamb of Thomson Reuters gave an OpenCalais perspective, seeing the "bounce" effect of people completing one search and then returning to initiate another search, resulting in getting data but not necessarily a lot of information. The OpenCalais linked data cloud of links will help people to traverse content more easily than in typical search engine-driven information retrieval. Amiad Solomon of Pier39 focuses on search applications sees the post-search world focusing a lot more on monetizing content once people land on pages, providing different data layers for specific types of people engaging content. Nancy Harvey of Wolfram Research focuses on eliminating search inefficiencies by making fact retrieval more accurate and natural. A "clean" database of information has a natural language query interface on top of it that returns results "in a visually stunning way." She sees them co-existing with traditional search, providing specific answers to specific questions. Paul Martino of Aggregate Knowledge gave the analogy of seeing an ad for a television show that he watches appearing in Times Square alerting him to a new season of episodes. Being able to deliver information focused on interests without having to express those interests explicitly is key to the post-search world in his mind.

FP: Search does not have to be very good because users are very good filters, which makes search a fairly "forgiving" tool. But how do you deal with the risk of mistakes when people get the wrong results?

PM: Not a semantic person, Google AdSense gets it wrong a lot of the time, does a passive read of the page, meaning oftentimes inappropriate results. But it can be mitigated with whitelists, blacklists and so on, but ultimately the user doesn't get mad typically. The question from his perspective is how to deal with brand safety issues.

NH: Have to watch out for "artificial stupidity" in search results. For example, a search "what was the weather on Valentine's Day last year" may return different results for different locations or try to disambiguate before answering (for example, "Did you mean San Francisco in California"). They also manage this by steering away from domains that they don't understand. But the number of times that they need to do this is declining.

AS: Look at it from the advertising perspective, all major brands are saying that they don't want to be next to the wrong message, have to know how ad servers work, but takes sophistication. If there's an airline disaster, you need to be sensitive to avoid mismatches.

CL: Dealing with a more constrained group of users, optimized for business content, enables them to do a better job for that domain. But there are still domain issues, such as the term "The Fed" meaning the federal government in Australia and referring to the Federal Reserve in the U.S.

PM: Histograms of information on why an ad was displayed - nobody was clicking on them. Most people didn't even know that it was there. Sites with recommendations may be more important to them, people, content and ads are all of the same type, what's the best match across these lines, the user sees it all as potentially useful content.

CL: OpenCalais does disambiguation, offers a way to get into the linked data world, disambiguating offers a way to match content to content in meaningful ways. For example a press release may offers an Apple story and links to company information, not the fruit. This means a fundamental change to the surfing experience, over time there will be an increasing demand to match content to content.

AS: Need to understand the creative itself, look at the description (I would add, also look at the graphics for familiar patterns)

FP: With this melding of ads and content and content-to-content matching, it becomes a blurred editorial experience for users. How can this be managed?

PM: If ads are scored higher than content for a particular user, but sometimes when the best content is the ad the social contract is difficult to implement only with "sponsored content" labels.

FP: One school of thought tends to go for a statistical approach to infer relevance for content and ads. Some come from a semantic approach.

CL: They do use both natural language processing tools powered by a tech staff of about 40 people developing rules and coding.

AS: Mostly machine learning

NH: W/A is "intensely hand-crafted," engage world-class experts (comment: not scalable, not sure what business model this gives them ultimately)

PM: A semantic approach overall, used OpenCalais to build semantic maps internally to determine and test their weighting, try to treat all data sets the same and adjust weights objectively.

A good panel, was a little frustrating trying to blog this, as it covered a lot of ground, but I found it interesting. What seemed to be brushed over were the full impact of search on editorial operations as well as ad operations. I think that Paul Martino tuned in to one of the key "post-search" issues, namely the ability of search tools to build content that will have both editorial and sponsored "slots" that may appeal to different audiences on different levels. This sponsorship is thought of today mostly in terms of traditional media ads, but it is going to work its way into the enterprise space as well. How do suppliers and clients gain the attention of organizations, and how do publishers enable companies to provide factual content through their channels to enterprises on a sponsored basis? These are opportunities for search technologies to ponder. Ultimately search tools may frustrate traditional publishers, as it seems like a different language that's creating content value, but these technologies are creating opportunities for relationships between audiences and content producers that are becoming increasingly sophisticated.