Brittany
08-03-2004, 02:53 PM
Personalized search is viewed as one of the directions search may head in the future, as Search Engine Watch’s Danny Sullivan pointed out at the Search Engine Strategies San Jose conference.
This has been a much-discussed topic (http://www.webpronews.com/news/ebusinessnews/wpn-45-20040803YahooTalksPersonalizedSearch.html) in the past, and today’s conference offered an entire session on personalization. During that session, Eurekster CEO Grant Ryan revealed Eurekster’s newly released idea of Information Nations.
Moving away from the “primitive” ways of the past, Eurekster asked: who should decide relevance? Through Information Nations, Eurekster allows people to build search engines around areas they are passionate about and determine what’s relevant to those areas.
“The really powerful thing,” Grant said, “is that we learn from user behavior about what’s relevant.” In narrow, focused areas, Eurekster discovered it is possible to learn from user behavior because when users click links, they are essentially telling you what sites are important to those niche areas. “We’re adapting policies based on what the end user thinks is relevant,” Grant said.
As people begin to set up Information Nations, they’ll be able to decide whether their individual nations have an autonomous approach (where the creator has “ultimate control” over what’s considered relevant), a democratic approach (for example, where a group of users determines what’s relevant); or an anarchistic approach. Over time, he believes it will be interesting to see which of these methods works best.
ChoiceStream CTO Michael Strickman pointed out that personalized search means different things to different people. By his definition personalization is any method that uses understanding of the user to provide better results.
He’s found that consumers do have quite an interest in personalized search but they need more information before they’ll be willing to participate.
A good example of personalized search, he said, is A9, which keeps track of searches and is able to retrieve the information to allow people to repeat those same searches at a later time.
Google is working on its own subject-based form of personalized search, where the user sets up a profile and search results can be filtered to the user’s area of interest. However, this poses a challenge. Search is temporal. Just because people search for something doesn’t mean they are truly interested in it. For example, if you’re sick and search for information on a particular medicine, that doesn’t mean you are always going to be interested in finding information on that medicine. That’s one obstacle that must be overcome for this type of personalization to work, he said.
Another problem is ambiguous terms. The popular example is: if I search for the term “jaguar” do I mean the feline or the car? Search engines will need to learn to differentiate and determine which term the user means.
One form of personalization, attribute-based personalized search, uses an understanding of page content to improve the search results the user sees. ChoiceStream analyzes and scores pages by their attributes, looking for categories, the type of page (ex: product review sites, blogs, etc.), and the style.
Personalization is a relatively new area and there are many setbacks, including the fact that users don’t like to answer a lot of questions. You must be careful with what you are asking. Watching user behavior is also difficult because it often has a lot of static. For example, what if multiple users share a computer? Or what if a user performs a search for a friend?
Another problem is that many people don’t want to reveal too much information. I ran into that problem just last night when I was discussing personalized search with a friend. “You mean I have to give the search engines information about myself in order for this to work?” he asked. “Forget that! I’m not telling them anything about me!” He clearly didn’t trust the engines or what they might do with his information, and I’ve run into many people who share similar views. They seem to be paranoid that the search engines will use this information to “spy” on them or share their information with other sources. This seems to be a huge barrier to the success of personalized search.
There are also risks to personalized search, Michael said, including the risk of actually making the search results worse, rather than more relevant to the user. To be successful, the users’ interests should also be the top priority and should always remain in control of what’s considered relevant. Attributes must be chosen correctly, should reflect user interest, and also cut across many search categories. Questions should be quick and easy to answer, gaining a basic handle of the user. It’s always possible to “drill deeper” later to gather more information.
This has been a much-discussed topic (http://www.webpronews.com/news/ebusinessnews/wpn-45-20040803YahooTalksPersonalizedSearch.html) in the past, and today’s conference offered an entire session on personalization. During that session, Eurekster CEO Grant Ryan revealed Eurekster’s newly released idea of Information Nations.
Moving away from the “primitive” ways of the past, Eurekster asked: who should decide relevance? Through Information Nations, Eurekster allows people to build search engines around areas they are passionate about and determine what’s relevant to those areas.
“The really powerful thing,” Grant said, “is that we learn from user behavior about what’s relevant.” In narrow, focused areas, Eurekster discovered it is possible to learn from user behavior because when users click links, they are essentially telling you what sites are important to those niche areas. “We’re adapting policies based on what the end user thinks is relevant,” Grant said.
As people begin to set up Information Nations, they’ll be able to decide whether their individual nations have an autonomous approach (where the creator has “ultimate control” over what’s considered relevant), a democratic approach (for example, where a group of users determines what’s relevant); or an anarchistic approach. Over time, he believes it will be interesting to see which of these methods works best.
ChoiceStream CTO Michael Strickman pointed out that personalized search means different things to different people. By his definition personalization is any method that uses understanding of the user to provide better results.
He’s found that consumers do have quite an interest in personalized search but they need more information before they’ll be willing to participate.
A good example of personalized search, he said, is A9, which keeps track of searches and is able to retrieve the information to allow people to repeat those same searches at a later time.
Google is working on its own subject-based form of personalized search, where the user sets up a profile and search results can be filtered to the user’s area of interest. However, this poses a challenge. Search is temporal. Just because people search for something doesn’t mean they are truly interested in it. For example, if you’re sick and search for information on a particular medicine, that doesn’t mean you are always going to be interested in finding information on that medicine. That’s one obstacle that must be overcome for this type of personalization to work, he said.
Another problem is ambiguous terms. The popular example is: if I search for the term “jaguar” do I mean the feline or the car? Search engines will need to learn to differentiate and determine which term the user means.
One form of personalization, attribute-based personalized search, uses an understanding of page content to improve the search results the user sees. ChoiceStream analyzes and scores pages by their attributes, looking for categories, the type of page (ex: product review sites, blogs, etc.), and the style.
Personalization is a relatively new area and there are many setbacks, including the fact that users don’t like to answer a lot of questions. You must be careful with what you are asking. Watching user behavior is also difficult because it often has a lot of static. For example, what if multiple users share a computer? Or what if a user performs a search for a friend?
Another problem is that many people don’t want to reveal too much information. I ran into that problem just last night when I was discussing personalized search with a friend. “You mean I have to give the search engines information about myself in order for this to work?” he asked. “Forget that! I’m not telling them anything about me!” He clearly didn’t trust the engines or what they might do with his information, and I’ve run into many people who share similar views. They seem to be paranoid that the search engines will use this information to “spy” on them or share their information with other sources. This seems to be a huge barrier to the success of personalized search.
There are also risks to personalized search, Michael said, including the risk of actually making the search results worse, rather than more relevant to the user. To be successful, the users’ interests should also be the top priority and should always remain in control of what’s considered relevant. Attributes must be chosen correctly, should reflect user interest, and also cut across many search categories. Questions should be quick and easy to answer, gaining a basic handle of the user. It’s always possible to “drill deeper” later to gather more information.