The Semantic Web takes shape, with Twine

This one is worth waiting for: Twine. Still in private beta, at its most basic it is shared bookmarking service. It blends additional concepts from newsgroups, forums, social networking sites, online databases, and wikis. There's a lot of semantic Web theory (and technology) underneath the interface, which still needs to evolve a bit, but even in this early stage it's a compelling product.

Twine uses item metadata as well as natural-language processing to extract tags from items you enter.

Twine lets you create or participate in topic areas called "Twines." Users post Web addresses, photo and video links, files, and text comments into a Twine. Twine goes to work analyzing the items and automatically finding tags for them. For example, when I created a "Web 2.0 reviews" Twine and added Webware, it automatically tagged it "applications," "business," "silicon valley," and so forth, which are keywords that Webware.com pages has in its HTML source. The service also tries to divine meaning from the pages and add its own tags, and it attempts to figure out the people and the locations associated with an item, and put that information in as well.

From all these tags, including tags that you manually attach to stories, Twine will then look up items that it thinks you will be interested in, as well as other Twine members that have similarly-tagged items in their portfolios. It also uses the strength of your connections to various other members to weight item recommendations.

The upshot of this is that its recommendations should be pretty good. Immediately after I entered in my Webware link, Twine recommended to me a video of Eric Schmidt (Google's CEO), a review of the Web 2.0 Summit Launchpad, and other items I thought I should definitely check out.

It's easy to get items into Twine. There's a bookmarklet that grabs URLs and images, and, if the item comes from a source such as Flickr, YouTube, or Amazon, additional metadata as well. You can also e-mail items into Twine, either to your own (or other peoples') Twine address or directly into a topic Twine.

Each item also gets a discussion thread, or you can make a discussion an item itself.

In a nutshell, Twine builds a semantic web (small w) from all the items, people, collections, and tags that are contributed to it. I think it does a killer job of weaving everything together. However, it's a rough cloth. The user interface appears straightforward at first, but it takes some study to understand what's going on and how to exploit it. As other writers have said, even at this early stage, with only 30,000 users, it's easy to see how Twine could contribute to personal information overload.

The database that Twine builds is as open as the company can make it. All pages can alternatively be viewed in machine-readable RDF format, and a two-way API is in the works. That's pretty cool, although Twine is neither a general-purpose social site such as Twitter nor a database such as Freebase, so I'm not sure who's going to bother creating applications for Twine. Though if the social aggregators (such as FriendFeed and Plaxo) want to do so, Twine's open strategy should make it easy.

A "Twine" in the system is where users can add items and discuss them.

See also: JetEye, Plum, Squidoo, and of course Delicious.

No word yet on when the site opens up to public beta. We had some invitations to the private beta, but they're gone. We'll update this post if we get more.