The important thing to understand is that while people may not use the same labeling system (or more simply vocabulary) when participating in a folksonomy, they are classifying the same underlying real things. You can use this knowledge to create a map between individuals' and even whole departments' vocabularies. For example, let's say that every time Joe in Planning labels a document as “Strategic”, Bill in Marketing labels it as “CustomerCentric” 80% of the time, and Jane in Product Development labels it as “UserCentric” 60% of the time.

Now, let's say that Jane in Product Development wants to find all “UserCentric” documents. Well, you can give her all of the documents that she has herself labeled as “UserCentric”, but you can also show her with lesser confidence Joe's “Strategic” documents, and possibly Bill's “CustomerCentric” documents (depending on the specific correlation between Bill and Jane's classification systems). The trick is to sort things by relevance based on the correlation in the classification systems you have observed.

Think this sounds far-fetched, it's a super simple version of collaborative filtering, the essence of Amazon's much-vaunted customer recommendation system. There are few things needed to make such a system work. First, people need to have a fairly extensive body of shared items they have classified so that there is a basis to compute the correlations. You might accomplish this by feeding them some common company items, say corporate memos that receive wide distribution. Better yet, you could have them label blog postings on your internal intranet. Second, you might be concerned about the high number of one-to-one correlations you would need to compute (number of people classifying by number of labels per person in the worst case). This is much larger than if you came up with just one single, centralized classification system. Note that since people are using labels over multiple items, you are already gaining some economy over no classification system. Additionally, it is very likely that people who are in the same department will be more likely to share or almost share labeling systems as has been noted in past work on quantitative ethnography (Krippendorf's approach to Content Analysis in particular).

Recently, a number of people have considered folksonomy as either a least common denominator approach to classifying information that did not impart much value, a sort of fun way to discover what others might recommend from your own social network, or a lazy way of doing classification that fit what people did anyhow. Those who have defended folksonomy have talked about it as a way to measure the spread of your meme (1, 2, 3) or as an economic if messy necessity in the face of the impossible task of keeping a centrally defined taxonomy up to date. However, underlying these defenses has been the idea that you either want to spread your meme and make it dominant or needed to somehow pare down the folksonomy raw data to get something more closely approximating a standard, centralized system.

What I've tried to demonstrate here is a method that embraces diversity and merely tries to use the information embedded in its mere existence to bridge it. This approach has the economic advantage of deriving value from something people just seem to do naturally, label things, seemingly the initial impetus for its sudden burst on the scene.