Calculating similarity of values that contain a mix of text and numerics

I want to store name-value pairs in a multi-valued field, e.g. properties: "length 42", "height 12", "width 20" Is there a way of comparing similar documents with these part text/part numeric fields so that closer numeric values are scored higher?

Using large numbers of weighted tags to compare documents

term (which will usually only be at most once for each tag in each document)? Should I be looking at the DisMax query handler instead in order to apply boosts to tag values? Aidan