What is a good system for generating BibTeX reference keys?

Question

When using a reference manager (or even when tracking references on a handwritten draft), it is common to have a key for each entry when referring to items, which the computer (or you in a final form) replaces with a properly styled reference later (in the style of the journal).

E.g., in BibTeX, which I am most familiar with, writing \cite{Box2015a} might be replaced with "[1]" or with "(Box, O. 2015)" or with "(Box, O. 2015, Meaningful reference key format)" depending on style -- the key in question being the Box2015a part.

Having a good format for the key seems useful, since one would spend a lot of time only seeing the key, rather than the full reference. Simply the author name and date, and a letter to break conflicts doesn't seem ideal.

Was Box2015a Box's work on Meaningful Reference Key formats? Or was it his work on Bicycle Speed Dependency on Weather? Getting that wrong would be embarrassing, and also accidental plagiarism (since credit was not given to the right paper).

What is a better format for reference keys?

Answer 1

Probably, the only general rule to follow is that the key should be something that well characterizes the document. Anything more specific will depend a lot on the topics you are writing about, and vary from reference to reference:

• Is the title very unique, or rather generic? In the former case, a shortened version of the title could be integrated into the key. (As a random example: The title The triangle processor and normal vector shader: a VLSI system for high performance graphics might be shortened to something like TriangleProcHighPerf.)
• Does the publication describe a product or technique that has a name (which you might also use in your text)? If so, that name could become a part of the key.
• Is the publication connected to a recognizeable author name, or do you rarely see the same author name twice in the literature you deal with, in particular with respect to specific approaches? In the latter case, author names may just be arbitrary strings that do not help you remember anything particular, while in the former case, you might think about including the author name the work is associated with the name in the key.
• Is the year in any way special for the work? For example, is it an exceptionally early example of a supposedly modern invention, or is it the variant that has become known as "the 2011-version" of a particular approach? If so, the year could reasonably be a part of the key, otherwise, it seems superfluous.
• Can the publication be categorized? For instance, you may want to indicate in the key whether something is a concept draft, a user study of a concept presented elsewhere, a survey of several techniques, or a design rationale for a given concept.
• Are there various versions of essentially the same work published by different publishers? Different layouts and presentation forms (monochrome vs. color, ...) may have different strengths, so you may end up wanting to specifically refer to (w.l.o.g.) the Springer version and the IEEE version of some work that for some reason was published twice. In that case, including the publisher name in the key might be reasonable.

I hope this helps to get the idea - I do not see a reason for a uniform key format here; instead, this case-specific format highlights the peculiarities of each referenced work and therefore seems to help best to remember which reference points to what work in my experience.

For me personally, the above system generally leads to keys that never include an author name or publication year, and almost always a concept name, otherwise some fragments of the paper title. Depending on your topics, you may well end up with different preferences.

I have never had a problem of key collisions while using this approach; if anything (not that it would actually cause any problem), I may have ended up with several different keys in cases where I created bibliography entries for the same publication several times rather than copying the first one to later works.

To summarize:

• An advantage of this system is that you do not need to remember any information that is not descriptive for the content of the paper (publication year, author name, publication venue, ...) to understand the reference.
• A disadvantage may be that the key cannot be generated automatically, though I use one .bib file per project and my workflow is usually decide to add reference -> search for reference in JabRef to check whether it is there -> add if it does not exist -> Ctrl+K to get insertion-ready \cite-command in the clipboard, where automatic generation of keys is a non-issue.

Answer 2

I'm using (BibTeX) reference keys that consist of three parts:

• the last name of the first author
• the last two digits of the publication year
• the first meaningful word from the title

The idea of "meaningful words" is a bit vague, of course. Usually, that's the first adjectiv, verb (except "to be"), or noun in the title. For example, I would use box15meaningful to refer to a paper entitled "On Meaningful Reference Key Formats" but box15bicycle to refer to "Is Bicycle Speed Dependent on Weather?".

(I think there is a term for "first meaningful word from the title" as the concept is used or was used in catalogues of libraries, but I forgot the term).

From my perspective, the main benefits of this system are:

1. Given the full reference, I can predict the citation key. For example, when I'm looking at the printed paper and want to cite it, I know the citation key without looking it up in the BibTeX file or using any software tool to look it up for me.

2. Given the citation key, I have some idea which paper it is, because the key includes a meaningful word from the title.

3. The keys are reasonably short and don't clutter up my text too much.

4. The citation keys can be used as part of file names, so I can also name pdfs like that if I happen to have a paper available as pdf on my hard drive. For example, I would have box15meaningful.pdf and box15bicycle.pdf.

5. The citation keys can be used as part of URLs, so I can name websites about my own papers like that. For example, if I would be Box, I might have a website like http://my-university.edu/~box/publications/box15bicycle/ where you could donwload the raw data used for my research on bicycle speed.

6. Given two BibTeX files that both use this scheme, I can merge them by merging entries with the same key, because the same paper will always get the same reference key.

Answer 3

This is almost a non-answer. The reason is that I use a similar format to what you find inadequate: "FirstAuthorLastName:Year". In my field we use Harvard style referencing so this is almost what appears in the text. The colon is not a key ingredient, it is just that I use the form "tab:xxx", "fig:xxx" and "eq:xxx", where "xxx" is the unique name I want for the object, for labelling floats and equations. My point is that for me being short is a necessity since I do not want unnecessarily long BibTeX keys or labels hanging around the document. I tried for a while to add number of authors, for example "Smith+4:2005" to distinguish from single authored "Smith:2005" but that ended up being to tiresome to set up I also shortened the multiauthors to "Smith+:2005" for a while. I use JabRef (no promotion intended) and have now simply resorted to specify that the preferred automatic key generation is "FirstAuthorLastName:Year" where JabRef will add "a", "b", "c" etc. where similarities appear.

So why do I work with the shorter form. First of all, I know what material I reference. I also know the material in the field. I later also double check the references, as they appear in the reference list, certainly before I submit a manuscript. So the key point here is the trade off between adding a lot of information to a label with lower degree of "mis-referencing" and short forms with potential risk of more misses. In the end you use whatever suits you but with time you probably end up simplifying.

Having written both papers and very long reports/books, I have never found this to be a big issue. I can understand that it becomes a problem if you for some reason are using a lot of references with which you are unfamiliar. So "better" is what you find best. If I understand correctly from your profile you aim to get into a PhD and I am sure your database will grow in a specific direction during the PhD, you will become intimately familiar with that set of literature, and you will perhaps change the way you BibTeX key label your references.

So better is what works best for you at the moment. In the long term, the simpler the better in my experience.

Answer 4

The mapping from "full citation information" to "citation key" is, in many ways, a hash function, and usually are by construction easy to apply but hard to invert. We do this because we don't want to have to type 50 to 100 characters each time we cite a paper.

The easiest way, frequently, to reconstruct the data from the hash key, is by looking up it up in a dictionary/table. If your bibliography database is big enough this makes it something much suited for software than for your brain.

So my workflow does the following:

• I use JabRef as my primary bibliographic data manager, as well as to keep track of my growing collection of PDFs.
• The citation key format is ABCDEF1234? The first six characters are formed by the names of the authors (following some rule), followed by four digit year, and followed by disambiguation suffix.
• I use Vim as my editor of choice.
• I export my citation database from JabRef using a custom-written export filter to become a Vim completefunc, which I load through my .vimrc everytime I edit a TeX file. This allows me to use it two ways:
  • I can type the start of the citation key, say Won, followed by <ctlr>-X <ctrl>-U and it will show me a popup list of all entries with key starting with "Won" which includes, in my case, all of my first authored papers with at most two authors. Highlighting selected entries in the list will show a "preview screen" showing the bibliographic information about that entry. I configure mine to only show the full title of the article, but it is easy enough to includes also publisher info etc.
  • Seeing an existing citation key in the document, bringing my cursor to the end of the key and hitting again the combo <ctrl>-X <ctrl>-U the list now has just one element, but the preview window still comes up showing me the bibliographic information.
• If I need to insert a citation to an article whose complete author list I cannot reconstruct in my head (and hence cannot know even the start of the citation key), I can either browse through the full list provided by the completion function in Vim, or just search in JabRef.

---

For illustrations:

Before invoking the previewer: Note the \cite{...} string in the middle of the window.

Putting the cursor on Alinac1999 brings up a pop-up menu (turns out Serge Alinhac as at least three papers in 1999) and a preview pane on top.

Answer 5

My system of choice is to let my reference manager, or some external database, handle the key generation. Every time I need to insert a citation, I go to the external tool or database, look up the reference (thus avoiding most of the risk of using the wrong key), and copy the citation key into my document. It sounds impractical, but actually I've found the process to be pretty smooth and not that inconvenient. The few keys that I use the most often in any given paper, I wind up remembering anyway.

I use Mendeley as a reference manager, which presents metadata next to a view of the PDF of the paper. So when I look up some information in a paper, it's easy to copy the citation key directly from Mendeley into my document. I imagine there are many other reference managers that have this same feature.

Answer 6

I primarily use Jabref and LyX. Hence, the need to encode every bit of meaningful information in the key does not arise.

Hence, I have a simple mechanism of key naming, which is <FirstAuthorLastName>:<Year>:<JournalAbbreviation>

If it is a book, then I replace the last part with Book. I have set up Jabref that it creates the first two parts automatically when importing. Since the journal abbreviations are not standard (yet), have not automated that.

Since LyX is used to write the document and insert entries, I can do a search to find out which articles need to be cited and the key is not hugely important ...