This topic was the hottest thing in search before the SEMPO scandal (just kidding), and I look forward to hearing what some of the panelists have to say, specifically Eurekster. Danny starts asking who hear feels that personalized search will be the next big wave in search. Hands went up in the audience. Danny says that this might be one avenue towards this area.
Grant Ryan from Eurekster was up first, he has an European accent (Irish maybe?). He began with the evolution of search, and he said he believes that the next step search is going is to tell you what you want as an individual. He then went into this concept of "Information Nation", which he describes as a global view of the internet. To extend this global view into a focused information nations that segment information from different perspectives. With Eurekster, the end users define the "information nations".
Eurekster allows you to create a nation and then define where to pull the information from (currently you have a choice between Yahoo and Feedster). Then you add more information for Eurekster to help create this nation for you. Then you have a search box that pulls from that new information nation.
How does it work? Eurekster learns from user behavior, so if they type something in and then click around, it picks that up and changes the next search based on that behavior. When people create these nations, they can look at what 'policies' were set up when creating this nation.
In the future, they feel that there will be 10,000+ information nations with lots of different ways of viewing the internet. I guess people will navigate to these nations via links from community sites. Advertisers can buy into keywords AND into keywords targeted to specific information nations. At the end, he even mentioned his accent :).
Michael Strickman from ChoiceStream was next. They work on narrowing choices to the few that matter (with search and other methods). They look at content, user tastes and more. What is personalized search? Any method that applies an understanding of a users preferences or behavior to improve their search results. Do people want personalizes search? 81% say they want personalized content and 64% were willing to provide personal data to get it. Types of personalized Web search are; (1) search history, (2) subject-based personalization (see Google beta), (3) collaborative filtering (correlated affinities), (4) disambiguation (the jaguar example - if I know about the user, I can show you the car). Newer focuses, social networking (Eurekster) and attribute based personalization.
Attribute Based Personalization automatically looks at categories, type of site, and style of the person (hip, upscale, etc.). Then it it looks at users' preferences, either they ask or watch the user to get this information. Then the algorithm takes into account individual relevance ranking, including general relevance and user preferences. And finally they continue to learn from user specific actions and at group level actions. Attribute based search works well with broad searches and is not so good with "navigational" search.
Jan Pedersen (chief scientist) from Yahoo is our last speaker on this topic. Customization is brought by explicit feedback from the user, and that implies your storing this information. An other form of personalization is contextualization, task specific meaning and inferred from the environment. Yahoo has "My Yahoo" is the "grand-daddy of personalization". You can customize the look, colors, add rss feeds, mail, news, weather and more. He pulled up an RSS feed from My Yahoo, Search Engine Watch was selected and so was Andy's blog, mine was visible but the last one in the 5 or so listed. He then shows personalization in Yahoo! Shopping where he shows a "save my products" link, this is the storage component. Moving towards "contextualization" he showed "Smartview" by Yahoo! maps. So you type in a location and then you can select food, then narrow it down and it shows you on the map, local restaurants (you get the idea).
Possible Search Future, but he is not talking about Yahoo's plans but concepts that are possibilities. Within customization you can store results (information management), an other form is by making an explicit statement of an interest and then results follow based on that statement (type in Eagles and then narrow it down to classic rock). Within the realm of contextualization, you can apply a location bias based on the location you are searching from.
Q & A:
There is a huge barrier to enter to personalized search, how would a company get beyond that? With Eurekster, one person creates the "information nation" the others just use the search box at the community Web site. Michael adds that as long as the barriers are very small, the users will give up some information for the better quality of results.
Why will personalized search succeed? Amazon uses personalized search well but will a Google.com or Yahoo Search succeed if they used the personalized approach? They all say yes but it will be hard to get there.
Is Eurekster worried about creating thousands of results full of spam? No, they won't be used because they won't be trust. All Eurekster is doing is allowing them to create these information nations. In the long run, they are looking to do a revenue share with large partners with Eurekster information nations.
When is personalized search going to hit mainstream? Yahoo said its very hard to predict these types of things. Yahoo said, he thinks soon. Danny Sullivan said 6 months to a year you will see a bigger push towards personalization (like we saw with local search).
Why do you think that a click equals relevance? Eurekster said that if they click and come back, they don't count it. But these things are based on inferences. Both Michael and Yahoo add to the click question that you can't look at them in an individual basis but rather an aggregate basis.
I was thinking about bringing up a statement made at a SES conference in the past by Google. Where Google said its more confusing to the user to change the results based on which computer you are using. This was a reply by Google based on a question about personalization. It would be an unfair question to bring up because (1) Google wasn't on the panel and (2) these companies are personalization engines. However, I do believe Google would have a different response today then it did a year or so ago.
Discuss this at Search Engine Watch Forums.
