A lot of interesting thoughts going on in this thread.
First of all, I think the approach taken depends on the type of photography you do as well as the amount of images you catalog. A wedding photographer will probably have quite different needs than say an entomologist or a forensic photographer. That being said, I think it is better to have too many keywords than too few.
Now, I don't use LR for cataloging so you will have to take what follows further down and adapt it, if possible. Frankly, I think the cataloging abilities in LR are severely lacking. I am also an RSP owner and will use LR for its RC abilities, which I like a lot, but I will (for the time being anyway) be cataloging with IMatch and printing with QImage.
Regarding keywording, I use a hierarchical system of categories with many "roots". Categories are converted to IPTC keywords (via scripts), but not all categories in a categories-path become IPTC keywords. Other categories get converted to other IPTC fields than keywords, such as location tags. However, for the purpose of a keywording strategy, lets just assume that keywords and categories are the same.
The root categories I have set up are: Events, Locations, Subject Type, What, Who, Workflow States and some others. Some of these roots only have one level below them, an example being Workflow States, which simple has the child categories New, Reviewed, Edited and Final. Subject Types is the same, with e.g. Abstract, Action, Still Life, Landscapes, etc. Others are quite deep, such as the Locations root which goes from continent all the way down to local area, for example:
-Locations
----Europe
------Spain
--------Andalucia
----------Granada
------------Alhambra
...
The What root is the Universal IPTC category structure provided by IMatch, to which I add missing entries.
-What
----Animals
------Birds
------Insects and Spiders
--------Bees
--------Wasps
------Mammals
----Nature & Environment
------Beaches
------Deserts
------Water
--------Canals
...
One of the most complex ones to work out is the Who root, which includes people by name. I chose to create the hierarchy starting with the people's relationship to me, and then in order from oldest to youngest unless they are part of the immediate family. Just below the root, I have Unknown, Colleagues, Family, Friends. In the following example, each person's name has been replaced by their relationship to makes the example clear.
-Who
--Family
----Me
------Wife
------Daughter
--------Husband
--------Son
----------Wife
----Brother
------Wife
------Daughter
------Pet
--------Dog
--------Cat
----Mother
------Father
------Mother
------Sister
--------Husband
--------Daughter
------Brother
--------Wife
--------Daughter
----Father
...
Something to keep in mind is that IMatch allows me to copy categories "by reference", so I can have the same category in several paths, but assigning one of them assigns them all. For example, under What, I have "Cars" under both the path "Transportation/Automotive/Cars" and "Sports/Auto Racing/Cars".
Now, when it comes time to start assigning categories to the images, here's a method that's worked well for me, especially when there are a lot of images to categorize (and this is where the "Workflow States" category comes in handy).
-During import, all images automatically get assigned the "Workflow States/New" category.
-Do a search to show only images with the category "Workflow States/New"
-Start at the top of the list of images returned by the search and select the first image.
-Choose a category/keyword this image belongs to but do not assign it yet.
-Go down the list of "new" images and select all other images that also fit this category/keyword (e.g. using Ctrl-Click).
-Now assign the category/keyword to all the selected images in one go.
-Deselect all images except the first one.
-Choose the next category/keyword that the first image belongs to.
-Go down the list of "new" images and select all other images that also fit this category/keyword (e.g. using Ctrl-Click).
-Now assign the category/keyword to all the selected images in one go.
-Deselect all images except the first one.
-Repeat until there are no more categories/keywords left to assign to the first image.
-Select the first image and all other images that also have no more keywords to be assigned to them.
-Assign them the "Workflow States/Reviewed" (or Final).
-Refresh the search (for "new" images)
-Select the first image and repeat the process.
You will notice that as you get down the list (which gets shorter as you work), there are less and less keywords to be added and the process becomes quicker and quicker. Psychologically, I find this method easy to use because, while the process is slow in the beginning, it speeds up rather quickly and you feel like progress is being made. Usually, by the time I've tagged 10 or so images, it is real fast from there.
Another thing I do during this process is that I do not look at the image and think of a keyword, but instead start at the top of my keywords list and see it they fit the image. Not until I'm at the end of the list do I look at the image and assign any missing keywords.
/Kris