I’ve been thinking my fair bit about tagging, and have come to the conclusion that there are two fundamentally different tagging strategies, depending on whether your goal is to categorize your information, or to make it easy to retrieve a particular item. I’ll use the example of bookmarks here, but the same essentially applies to any collection of information.

The “categorize” strategy is essentially the one proposed by Ian. The goal of this strategy is to be able to quickly retrieve all bookmarks belonging to a particular category. It consists of two steps. In the first step, you make a list of tags that you will use, corresponding to your categories. In the second step, you use tags from this list to tag your bookmarks as you add them to your collection. From time to time you might decide to add another tag to your list, if you’ve found that tag to be useful.

For example, your tags might be microsoft, windows, linux, gentoo, debian, kernel, cplusplus. If you’d want to see all bookmarks about kernels, whether they be about the Linux kernel or the Windows XP kernel, you’d use the kernel tag to find these. Or you might use the combination of kernel and linux to find only bookmarks about the Linux kernel.

The second, alternative strategy I call the “retrieval” strategy. Its goal is to quickly be able to retrieve one particular bookmark. The idea is that if you remember approximately the content of one particular website, you quickly find the right bookmark. For example, you would use the tags gentoo, kernel, compilation, and udev to tag a page about compiling the kernel with Udev support in Gentoo Linux.

The fundamental difference between the two strategies now emerges. If you were using the “categorize” strategy, you would be likely to also tag the Gentoo kernel page with linux, since it clearly falls into that category, but you would not use udev, since you will be very likely to never have another bookmark about Udev, and you want to avoid categories containing single items. On the other hand, if you are using the “retrieval” strategy and you want to find that particular page you remember about compiling Udev into the gentoo kernel, then the tag gentoo makes the tag linux obsolete. Indeed, since Gentoo implies Linux (unless your collection contains bookmarks about a that particular kind of penguin), the bookmarks tagged gentoo are a subset of those tagged linux, and adding the tag linux to your search does not alter your results. The udev tag, though, will be very helpful since it will narrow down your results to the wanted bookmark.

To summarize, the “categorize” strategy asks for relatively broad tag categories (such as linux or windows, but not compiling or udev), otherwise your tag list would quickly be a cluttered mess. On the other hand, for the “retrieval” strategy the tags should be as specific as possible. This will clutter up my tag list, you say? Never mind, if you don’t use the tags to browse through your bookmarks.

I am curious if there is a way to reconcile the two strategies, to get the “best of both worlds”.

(As an aside, note that the “categorize” strategy is not much different to conventional bookmark handling in a web browser: First you create your bookmark folder hierarchy, then you add bookmarks to these folders as you surf the web. The difference is that with folders you can only make sets and subsets, but you cannot have intersections of sets, since if a bookmark belongs to a particular folder, it cannot also belong to a folder which is not a subset or a superset of the first folder. Another difference between between the two approaches is that there is no tagging system I know of that allows you to define subset-relationships between tasks. For example, you would define gentoo as a sub-tag of linux, so that anything you tag with gentoo would also be tagged with linux. Similarly, you would define linux as a sub-tag of the operating systems tag, and so on.)