About the model: Question 7
7. The aforementioned book on data modelling seemed to imply that models require each entity to have one and only one name. Does that mean that data modelling would preclude recognizing the fact that entities are often known by several variant names and that searchers should be successful no matter what variant name they search on? Or does the rule pertain only to the machine-readable identifier (not the human-readable identifier(s) linked to that machine-readable identifier)? Is there a “best-practices” way to model what we used to call a “see-reference” in RDF?
“No, it’s just a technical constraint of the relational model. To make persistent identifiers work, one thing is that there’s more than one per entity, but you have to resolve them using relational paradigms such as lookup tables. For example ; herbert,_frank f_herbert(1920-)both are PI’s to the same author, but in a relational model this is ; table : “author” columns : id, name ——————– 200 | Frank Herbert table : “author_pi”, columns : id, pi, author_id, primary ——————– 1 | herbert,_frank | 200 | false 2 | f_herbert(1920-) | 200 | trueIt’s the “author_pi” table (lookup table) that binds the PI’s together with the “author_id” field which points to the “author” table. this binding is expressed with SQL. (Not sure you know about SQL and relational models, so I’m writing as if you don’t. Just tell me if you do.)
The before mentioned constraint is that any id of any table must (well, should) be unique, not that PI’s can’t be unique. Basically, identifiers of a table should be unique, but the columns with their various properties need not”–September 11, 2007 email from Alexander Johannesen.