1. People are not static
The issue here is that people’s interests change over time, the topics in which they are interested is by no means a fixed list. Their interests while working will relate to their profession, in their leisure time they may have a multitude of fairly unrelated hobbies or interests. However, even if the set of topics is changing over time, a large proportion of an individuals search queries will be clustered around specific topics. This problem does not rule out the ability to recognise increased (or decreased) relevance of results found in particular domains. Things like homonym disambiguation are still possible. For example, if a person has previously used the keyword “rowing” in queries and visited sites relating to boats, then chances are they will be thinking of the same context when they use the word in future, rather than domestic disputes. What this problem does mean is that any system designed to track and make inferences from attention will need to take the dynamic nature of people into consideration. One important point here is that pages about topics of interest appear in clusters in the information space, and one area of interest may be totally distinct from another. Attempts to “average” or linearise results over time without taking this into consideration are not likely to produce useful results.
Let me add one more thing to the ideas Danny Ayers provides for thinking about personalized search and attention. "Personal" is also social, our individual experience is through our communities we join online; so communities are not static or singular in any sense, because they are an amalgam of individuals who are not static.
The problems of folksonomy, the limits of agreement about the meaning of tags and terms, are underscored by Danny's point about the evolving character of the individual. Meanings are refined by the community rather than defined by the community. Personalized search results are intricately linked to the evolution of the online community, so the individual's attention to particular communities and information is essential to parsing available information to their interests.