Learning Semantics

The "Semantic Web" is one of those fuzzy Web 3.0 concepts that is occupying think tanks, VCs, and even a few marketers of late. Also known as "The Data Web" or "The Implied Web," the idea is that various kinds of data across multiple sites can be combined in powerful ways that make the Web more like a giant computer. It can understand, even anticipate, our queries and even perform complex tasks. Alex Iskold, founder and CEO of AdaptiveBlue, has a rudimentary example of this concept that leverages past- and real-time user behavior to link information in more intelligent ways across sites. His BlueOrganizer plugs into Firefox, reads where you are and where you have been online to link information on the page to activities you are most likely to need.

Behavioral Insider: Explain how your BlueOrganizer takes a different approach to linking data sets across sites than some of the futuristic proposals for a Semantic Web.

Alex Iskold: The main difference is that the traditional semantic Web is about annotating pages so that computers can attain a near-natural language understanding of information and solve complex tasks. So, referring back to John Markoff's article in the New York Times, that would translate into typing in a query saying 'find me a perfect vacation on a budget of $3,000, and by the way I have two three-year-olds.' That is kind of the traditional semantic Web approach. We do it upside down. Instead of worrying about natural language understanding, which we think is a really hard problem, we ask, what if we just taught computers very basic things first? What if we encoded the notion of a book, a movie, travel destinations, wine, etc.? So we are dealing with everyday things that really dominate our online life. Our answer is, if computers knew what you are looking at, then they should help you get to more relevant information faster.

Behavioral Insider: Explain how the technology works on the page. I see it loaded in my Firefox browser toolbar and the drop-down menu seems to recognize keywords on the page offers links. How is the technology working?

BlueOrganizer is really about things, shortcuts to things and personalized shortcuts to things. It is leveraging a lot of different methods and techniques that are available today to present a uniform experience to the end user. So a Web site like Amazon has an API or a Web service that allows you to retrieve information about the products Amazon carries. So specifically if you go to a music album by Norah Jones, BlueOrganizer actually queries Amazon's Web servers and fetches information like title, the year of release, the genres and the artists. So we can leverage any Web site that has a Web service. That is how it should be and will be someday.

A lot of sites don't expose their data via an API. So in those circumstance we actually do analysis of the page you are on to retrieve the same attributes. We actually analyze paragraphs surrounding your selection much like humans do: When reading a piece of text, you have a context. If you are reading about a recipe on some blog, you can just highlight chicken and right-click. and say get me more chicken recipes on other sites.

Behavioral Insider: How do you incorporate personal data and history?

Arguably, an even more valuable piece is personalization. It scans your browsing history and automatically configures the menus to reflect the sites that you visit often. It leverages your browsing habits and aims to be automatically useful out of the box. If you go to a DVD page the smart menu gives a movies menu with shortcuts relevant to movies, and we have 20 different categories: books, apparel, technology. If you have a browsing history, the menus would automatically be configured in each category to show the sites you would use.

Behavioral Insider: How will marketing monetize this?

There are likely going to be contextual ads in SmartLinks. When users are looking at a movie they might see an ad for a new movie coming out. There will also be an opportunity to do sponsored links. When looking at a movie, there might be a sponsored link -- 'See clips of this movie on AOL.'

Behavioral Insider: Is this a better approach to the problem of building the Semantic Web?

When we started this company I personally thought this was just a step to a smarter Web, a transition point. I am more convinced that a combination of different things, including our technology, has a better chance than the traditional semantic Web approach. Vertical search engines are effectively semantic engines. searches. They do very similar things where it focuses on the verticals on the simple thing and then builds semantics around it.

Behavioral Insider: What does it mean to 'build semantics around it'?

Take Spock, a vertical search engine for people. It has an exclusive notion of relationships. It understands that everybody is related to other people in different ways. Look at Bill Clinton to Hillary, and the relationship is marital; to George Bush. the relationship would be successor. This is a very basic kind of assumption the vertical search engine makes. It understands that each person has a birth year, place, occupations. It looks at the web through the prism of people. Whatever query you ask, the only thing you get out of it is lists of people.

Behavioral Insider: If you narrow the focus down to a tight enough set of assumptions about what the data can offer you, then you can take a simple text page and turn it into a set of data points.

That is right.

Behavioral Insider: And that moves from a text Web to a semantic Web made up of data points that can be grouped and compared?

Computers don't get that chance now, and who knows when they will. To teach a computer a vertical is so much easier, because what all you are doing is hard-wiring semantics. Leveraging specific information is a far more promising approach in my mind. Instead of teaching computers semantics, let's ask what problem we are trying to solve -- sifting through large volumes of information while having shorter attention spans. We can make computers just smart enough to help us find relevant information faster.



Next story loading loading..