What is Tagyu?

Though it sounds like some sort of Japanese animé character, Tagyu is actually the latest offering from the smart Adam Kalsey. If his name is familiar, one reason might well be because he cofounded Pheedo a while back...

Tagyu is what Adam calls a "social tagging engine" and it's supposed to help you tag your own content by figuring out what you're talking about in a given blog entry or article, identifying others who are talking about the same topic, then suggesting specific tags based on how others are tagging their material.

Does that make sense?

I think it's darn interesting as a meta-keyword analysis system, though I'm skeptical of tagging as an organizational concept in the first place, as I explain in my article Technorati Tags: Good idea, terrible implementation.

Tagyu also has a fascinating -- and troubling -- problem identifying tags if you're not part of the crowd, if you're writing about unusual, non-mainstream topics or are covering topics before lots of other people write about them. As Adam says: "New concepts and ideas don't always make it into the indexes before people start querying against them. But the index is updated in near real-time so it shouldn't take long for Tagyu to become aware of new ideas and start sending targeted tags."

For more on this topic, Adam explained to a group of us why the Tagyu system had a hard time with Dylan Greene's blog entry Inspiration from Mahatma Gandhi, suggesting "audio poetry mp3 free":

"To provide a little insight into Tagyu, here's an overly-simplified description of how things work...

"Tagyu uses the "wisdom of crowds" to determine your tags. By figuring out what documents in its index are related to your content and how those documents were tagged by others, it can determine what tags are likely to be relevant for you.

"So step one is to figure out what text to use as the source of your query. When you give Tagyu a URL, it can't really figure out what on the page is the post and what is navigation, comments, and other such things. So all those extra bits are used in your query. For that reason, giving Tagyu the actual text of your entry gives you much better results than giving it a URL.

"The next step is to figure out which documents are related to yours. At this moment, Tagyu thinks these (among others) are relevant to your text (in no particular order):

http://en.wikipedia.org/wiki/Mahatma_Gandhi
http://www.forusa.org/programs/decade/const-program.html
http://www.web-developer-india.com/web/index.html
http://www.foreignpolicy.com/

The next step is to remove the obvious spam, so that web-developer-india one goes away. And then we look at the tags for those related documents, some other documents, do some processing and statistical analysis and hand you your tags. Since www.foreignpolicy.com has a lot of economic related things on it, it's not too surprising that there are economics-related terms in there. There's only seven documents in the index that contain the word "Ghandi", so your text is being matched on the other concepts in it, like peace and Iraq."

Good explanation, and a very interesting service. If you're a blogger or just interested in automated tagging (aka "keywording") of articles, check out Tagyu and see what you think.

Oh, and this article when referenced by URL produces the following tags:

tagging del.icio.us tags

and when the text of the article is pasted into the Tagyu box, it produces the following tags:

tagging del.icio.us tags tag tools

Where's the "tagyu" tag, Adam?

What do you think? Are they accurate classifications?