Practical Metadata

Jim Rapoza of eWeek has a great article on how desktop search is not as effective as his metadata-enhanced MP3 Collection. To quote, "Because of the way that music ripping and creation encourages proper tagging, it's very easy for me to find and sort music based on, for example, artist and musical style."
Thus, my favorite example of metadata (music metadata) strikes again and continues to make headway into the public consciousness...

Here is an improved attempt at a metadata definition:

Metadata – an external description of a distinct data resource. Common usages for metadata include providing the context of the data resource, managing its lifecycle and extending it to new uses. An example of metadata is the external description of an audio file specifying the artist that created it, when it was created, the length or play time, and the genre of music it belongs to. The purpose of metadata is to manage and improve the use of data as a strategic asset.

I believe this is an improved version that fixes the "data about data" problem, avoids the laundry list problem while giving some usage examples and closes with a clear statement of purpose. Next steps will be to apply the definition and drill down into the common usages I state in the definition: data asset context, data asset lifecycle management, and data asset extension (or repurposing). Many of these will be tested in a new contract we just won to provide data management support (more on that after the press release).

A good article on XML.com discussing the benefit of adding metadata tags in the content of news stories.
While, in general, I agree with the author's approach. There are some difficulties and some push back in content tagging. In fact, I have personally been in numerous briefings where content tagging was bashed as a pipe dream, overly burdensome for content creators, not worth the cost and several other criticisms. That is why I have favored a hybrid strategy where you combine an entity extractor for universal attributes and focus the content creators on more domain specific tagging.
That brings us to the difficulty of knowing what to tag. As entity extractors improve we will see them evolve beyond who, what, when and where to ontology-driven extractors that can focus on domain-specific knowledge.
I believe that with sites like del.icio.us we are seeing the public accept tagging as a natural activity and therefore the search engine vendors should be keenly aware of this and begin exploiting it to improve the precision of results. To even spur the common adoption of tagging content creation tools should do a better job of quantifying the utility of adding tagging to a user by providing some kind of statistical side panel on related documents as you add more tags. In other words, something that immediately allows the end-user to see the benefit that adding content metadata has on the ability for consumers to find their information once they publish it in the organization.

An important W3C Technical Architecture Group (TAG) finding has just come out entitled "Metadata in URIs". For those who have heard some of my presentations, you know that I am in favor of "semantic identifiers" in support of the notion of "semantic chains". My justification for both ideas is based upon the understanding that "semantics are infinite" and that applications need "just enough semantics" to get the job done. Given that it is crucial to deliver semantics in small quantities at the closest point to the user in conjunction with a link back to a larger pool of metadata. The document has a great example about URIs used in the public that I feel highlights the point well. Highly recommended reading...

I met Lieutenant General Steven Boutelle in August 2004 when I was speaking at the Army Knowledge Management conference in florida. He had read and liked my book on the semantic web. He was kind enough to actually kick-off my talk by exhorting the packed audience to listen carfefully to what I had to say. An article came out today where he discusses his current priorities and metadata ranks right up there! It is clear from the article that he both understands metadata (specifically his comments about authoritative data sources) and how critical it is in an SOA environment. Makes me proud to be a former Army Intelligence Officer... ;^)

In my opinion, the best and most intuitive example of metadata is digital music. I recently wrote a song about the unique name of my company, Oberon Associates, Inc. Here is the Dublin Core attributes of the song:

Title: O' Oberon
Creator (author): Michael C. Daconta
Subject or keywords: song,"Oberon Associates, Inc.",company,tribute
Description: A song about the company I work for, Oberon Associates, Inc., and the varying semantics of the word "Oberon". Specifically there are 6 references to the word Oberon and 2 literary references.
Date: 2006-05-03
Type: audio/mpeg
DCMI Type: [None]CollectionDatasetEventImageInteractive ResourcePhysical ObjectServiceSoftwareSoundText
Format: audio/mpeg 2446489 bytes
Identifier: http://www.daconta.us/Media/oberon-song.mp3
Language: en
Coverage: http://purl.org/dc/dcmitype/Sound
Rights management: (c) 2006 Michael C. Daconta
(Note: if you want the lyrics or chords to the song, you can get see them www.daconta.us).

You could enter these into dcdot to create an RDF version of the metadata. This song has 6 references to the semantics of Oberon and 2 literary references. In later posts, I will explore those semantics by creating richer metadata descriptions. Without the metadata, this song's value is almost counter-productive because people are very sensitive to their music. Music is powerful because it elicits emotion. Try randomly blaring a piece of music to someone you know nothing about and see what kind of reaction you get. Most of the time it will be negative. People strongly rely on music metadata (genre, artist, decade, etc.) to guide them to the right music. Metadata is all about guiding you to the "right" resource.