The grouping problem
Consider books in an academic citation database. Serial publications – journals, book series – possess profile pages. What if books would have one too? Would each edition of a book have its own separate profile page, or would all editions be combined into one?
There is always a discrepancy between the technical definition of an entity and the somewhat more expansive view that is useful to a user. Books and their editions illustrate this: each edition is technically a separate entity with its own identifier, but a user would really like to see them together. But there are many more examples:
- A journal and its translation into another language are clearly two different entities. However, since most metrics are based on citations, the impact of these journals gets diluted when some authors cite the original and others cite the translation.
- Organisations that merged with another organisation: attribution will be correct through inter-organisation relationships, but it would be more meaningful to group them on one profile page. Removing the ceased-to-exist organisation altogether is too drastic.
- Articles and their errata: an erratum is a work in its own right, but it plays a different role. Citation counts for errata are not meaningful. The article profile page should group the article and the erratum together.
In general, it is comfortable to believe that profile pages always map one-to-one with our entities. But that is not what the world of research looks like. The right concept here is counting units. Profile pages should be at the counting unit level. Most counting units are singletons, but sometimes they include something extra – something that deserves both its individual identity and its grouped treatment.
The indication problem
A related issue at the other end of the model: indications of occurrences.
It often happens that occurrences of entities in textual content cannot be identified as neat substrings. The best-known example is the affiliation string: "Departments of Mathematics and Computer Science, University of Utrecht" contains three organisation names, but they cannot be bracketed off using inline XML tags. They can, however, be annotated.
This is not just a feature of affiliations. References are a prime example. In arts and humanities journals, in law journals, multiple bibliographic references are interspersed within the text, not presented as a neat numbered list. No data left behind: in future we should capture the whole reference string as an indication of occurrences, rather than forcing it into a normalised bibliographic record that loses contextual information.
Even author bylines can behave this way. A document authored by "Bill and Melinda Gates" does not contain "Bill Gates" as a substring.
For this situation, we need a formal level just above the occurrences: the indication of occurrences. The mapping of a work to another entity is in general only a mapping from the "indication" to zero or more registered entities – not a series of one-to-one mappings between occurrences and entities.
The vast majority of cases are not like this, and hence this level is often overlooked. That was a mistake in legacy systems, and one we should not replicate in any new data model or new profile pages.