I suggested below that we think about catalog - or other search system - features in terms of rank, relate, and recommend. Things, you notice, tend to come in threes ....
In bibliographic systems we have created explicit relationships and we have used controlled data to create relationships - names, subjects, places, titles. One of the interesting things about current catalog discussion is that we are trying to mobilize more of this structure in the user interface. Faceted browsing approaches, for example, take advantage of this structure. FRBR is an approach to relating the members of a work.
Relationships of another sort are becoming more interesting also. These depend on behaviors: user choices or intentions. People who borrowed or bought this also borrowed or bought that, for example. Google's pagerank is another example.
Myself and Thom Hickey ended up looking for Huck Finn in various systems the other day. Many people know the Adventures of Huckleberry Finn as, simply, Huck Finn. The question was this: would systems 'relate' Huck Finn and the Adventures of Huckleberry Finn for a user? We did not have a great deal of luck (that said, we did spend a lot of time doing this!).
We had an illuminating experience on Amazon though. An Amazon search for Huck Finn finds a variety of items with Huck Finn in the title, as other systems do. However, at the top of the search are three versions of the Adventures of Huckleberry Finn. These are in a special area labelled: 'Customers who searched for huck finn ultimately chose:'.
This is a very nice example of a relationship mined from actual user behaviors. Amazon is relating the initial search query with peoples' purchase choices. It is interesting that they feature this so prominently in results.
Again, this is just one feature among the many that comprise Amazon's rich texture of suggestion.