Wednesday, November 15, 2006

Key References

Carlos Castillo asks for comments on his Key References Proposal.
The key references section of a paper points to the most similar previous articles on the same topic that were extended, improved, challenged, or built upon by the paper. Key references allow the author of a research article to highlight the most closely related previous work in the specific topic of the paper. Key references are the natural complement of keywords.
An interesting idea. If it is used (and taken seriously) by many authors it might help automated search systems identify important papers in the field.

On the other hand many journals require keywords and AMS classifications although I have rarely seen this information put to good use. For humans a well-written "Previous Work" section will have much greater value than just a list of references.

Key References won't become popular unless a major publisher requires them in their journal or conference articles. Would Key References play a useful role or just become one more thing authors have to do to get their papers published?


  1. I can see it now. Some arrogant paper-writer cites all of his papers as "key references."

  2. To me it sounds like a classical "why hasn't anyone thought of it before" idea. Keywords (and subject numbers) are useless but this could really work great, and is less work than keywords (you already cite the references, just tag the key ones).

    I wouldn't worry too much about anon 1's concerns, the cap on the number of key references (I would use 5 rather than the suggested 2-3) should take care of that nicely (very few people would fill it with 100% their own papers, and those would be taken care of by academic group dynamics).

  3. Here's a good reason to not do this: as anon #1's comment demonstrates, there's a very real and dangerous possibility that the idea of a "key" reference will be conflated with that of an "important paper". As I see it, a key ref. is one that's closely related to the article making the citation, not necessarily a great work.
    In fact, I would argue that we need fewer references in general, not more. There are too many "See also [11, 14, 8, 44, 33, 21, 22, 23, 24] for related work" type citations.

    Automatic systems already do a good job of extracting similarity between articles based on co-citation (and common references), we don't need authors to tell us more.

  4. I can think of reasons why this would be a bad idea: like keywords, it "freezes" the meaning of a paper into the time it was written. For example, would early papers on the PCP theorem mention "approximation algorithms"?

    I think collaborative filtering would work better: ask readers to contribute footnotes, so when getting an article, you would see
    "grad students who struggled through this paper, found the following other papers also useful ...."

  5. Keywords are useless, I can't imagine why key references are better than a well-written related work section. I can also easily imagine more than 3-4 works that are all highly relevant. And what is more relevant: the first paper on the topic, or the most recent one that the current paper directly improves on?

  6. This comment has been removed by a blog administrator.

  7. Hi everybody, thank you for your comments.

    > I would argue that we need fewer references in general, not more

    The key references are a small subset of the references. Today, depending on the area, papers have many references. "Key references" allow the author of a paper to highlight just a few of them.

    > I can't imagine why key references are better than a well-written related work section.

    They are not a replacement for the related work section, but can be seen as a very brief summary of it.

  8. What's the point? It's not that hard to read the introduction to a paper. Heck, that's about all I read 90% of the time.

    If the introduction of the paper is well written, it will include citations that develop motivations and applications, as well as citations for earlier work on a similar problem.

    Well drawn contrasts between the present work and earlier works are the norm- often in the form of the well-loved "table of parameters".