Classification - does it work?

A key KM competency is understanding and applying meta-data. Assigning keywords and tags, extracting concepts, finding and making distinctions, displaying and working with categories are foundational information activities.

Bill Ives has recently posted on this subject at Portals and KM, asking some key questions. Here are some additional reflections:

- Can you fully exploit Knowledge Assets in the form of documents without categorizing them?

Always cringe at the thought of exploiting knowledge assets, seems so explicit, static, repository and information paradigm orientated. If you do go down this route, then meta-data is indeed a key affordance, it needs to be added and reviewed with great care, there must be applications that make use of the added work and value, e.g. as navigation and search enhancements. Search alone just does not cut it.

- How is contextual content best understood and presented?

My experience is patterns show the way forward here. Make it imperative to define the applicable context, give explicit examples of areas which do not apply, list the exceptions and state the risks. The best way to add context is to assist with making person to person connections. Provide the opportunity to ask questions, test & validate, experiment and learn from the experience of others that have been down this path before. It is not the content but the connection that matters in the end.

- Is automatic categorizing working for some (or all) of your content?

Entirely automated classification is dangerous and ineffective - just look at Google search returns!. Where this helps is as a bootstrap tool, suggesting possible tags, keywords and concepts, but there needs to be supervision and manual intervention and discrimination to select best fitting descriptors. No machine can understand or make sense of the meaning behind assigned categories, this is an emergent property of the audience and the group working within the domain.

- Does taxonomy or navigated search have a place in your organization? Is it available and used?

Certainly. making distinctions, assigning meaningful names to key concepts, testing boundaries and common understandings are key to KM work. It does not matter that there is a strict hierarchy, what is important is to be able 'see' the related concepts, appreciate the boundaries, make the connections, share and grasp the core meaning around the classes. Any living and useful categorization needs to emergent, negotiated, reviewed, revisited and continually tested in everyday conversation.

- When is key-word, free-text or Google style searching good enough?

Almost never. Although keyword search is useful and becoming ubiquitous it is insufficient. We need an easy way to explore, learn, share, refine and repeat our individual searches. Notification when a new search yields different results, perhaps via RSS feeds, an ability to converse with others around the results, easily explore related concepts, receive pointers to people in addition to objects, are enhancements that spring to mind.

- Are you clear about which approach is needed and for what circumstances?

Interesting question. As KM folks we need an array of tools and practices, we should be always looking for social connections, ways to enable dialog, promote learning and encourage sharing. The information science approaches are useful, but they lack the social affordances that we need to make knowledge flow, to allow validation, to promote experimentation.

May I suggest (keyword) searching for information is not always the best way to approach the problem. What is needed is a forum to hold conversations around distinctions, ways to find experts, tools to build relationships and ask questions. This is a very different approach from searching archives as it promotes awareness, learning and provides solutions in context.

Comments

[Jamie]

Automatic categorisation is not automatic it requires expert resources. I have already reviewed Verity and Stratefy and am looking for other alternatives as I was very frustrated with the unpredictable nature of the results from these engines

Has anyone had any real success with aut-cat and not paid a fortune for it.

I have just read an interesting view of automatic classification on http://stream.framfab.com that seams to reflect what you say and my frustration.

http://stream.framfab.com/index.php?/weblog/comments/automatic_classification/

What about folksonomies as an alternative for metadata