CS270 -- Research Paper as a Class Project

Rich Wolski --- Winter, 2009

this page last updated: Sun Dec 14 18:00:14 PST 2008

Notes on Writing a Research Paper for CS270

Clearly it is not reasonable to expect to craft a complete research paper for publoication during the course of this class (althoigh it is certainly possible). Thus the expectation we will have is that you have written a complete paper as if you were to publish it without having completed all of the experiments or analysis necessary to make the paper complete. There should be at least one set of preliminary results and place holders for experiemnets, mathematics, etc. that would be needed to make the paper truly a publishable work.

With this in mind, it is important to review what goes into a good research paper. A good paper has an introduction, an exposition of the specific problem at hand, related and previous work that has examined the problem, a description of the new methodology that you will be applying to the problem, a description of how you will evaluate that methodlolgy, the results of that evaluation, and a conclusion. At a course level, you can think of these as subsections in research paper

However, one rarely sees papers that are decomposed strictly in this way, even among the papers that are considered to be the most significant in the field. While the organization may not reflect these components in a one-to-one manner, all good research papers include them in one way or another. Yours must as well.

There are other stylistic considerations that can also impact the quality of your paper. For example, the introduction should cover the complete paper in synopsis form. That is, by reading the introduction, the reader should understand the entire body of the paper. The rest of the paper, then, re-iterates the introduction, but substantiates its content with background material.

For example, in the introduction, you might include a sentence that says

Our results indicate that file system performance can be increased by as much as 38% due to the new method of caching block data that we have developed. This statement is, in fact, a statement fromt he results section. The results section, however, must substantiate this claim by providing detail that convinces the reader that it is likely to be true.

Another stylistic conundrum often arises with respect to the inclusion of related work. Typically, related work must both describe what work has taken place before your work and why that work has failed to indaequately solve the problem at hand. Thus you are tempted to put the related work section after the problem statement section. Often, though, a good way to expose this inadequacy while also making your method clearer is to contrast your method with previous methods in the related work section. This inclination mandates that the related work section appears after the description of your method. You may also want to contrast your work and previous work based on the quality of your resuts, which then dictates that the related work section appear after the results section. However, the way in which others have addressed the problem is also helpful in introducing the problem, or your methodology thereby indicating that the related work section appear early in the paper.

In truth, all placements of this section are acceptable as long as they are put into the proper context. Authors just tend to struggle with how to frame this context properly.

For this class, our expectation is that a full set of results will not be available and that, instead, you will include preliminary results and placeholders for missing results. You can either leave these results out, and describe how you would go about gathering them, or make up "phoney" results (that you indicate as such) that you think will eventually arise. Thus, the conclusions you draw will be based on a combination of preliminary results are your expectation of what a full set of results would show.