Reading a computer science research paper can seem like a daunting
task at first. However, it's important to understand that most papers
follower a particular pattern for their structure (to allow other
computer scientists to find information quickly). Here are some tips
for reading a CS paper:
-
Papers are like ogres in
that they have layers. Don't try to read and understand every word
the first time.
-
Don't necessarily read the paper from start to finish. It's okay
to skip around a bit. Take breaks between iterations of reading.
-
Papers generally follow a structure of:
-
Introduction
-
Background/related work (i.e., current state of the art)
-
Proposed approach
-
Evaluation
-
Discussion and Related work (if it wasn't second)
-
Conclusion
Each section will likely start with a brief summary of what will be
discussed next.
-
First, focus on finding answers to the following high-level questions.
CS papers are often written to answer these questions (a version of
the Heilmeier Catechism)
-
What problem is the paper trying to solve?
-
How is the problem currently solved? What are the limitations
of the current solution?
-
What is the key insight that
will make the proposed approach work.
-
What is one- or two-sentence summary of the new approach?
-
Who cares? What impact will success have?
-
How is the work evaluated? What are the metrics for success?
-
Was the new approach successful?
Note that these answers are often found in the introduction section of the
paper. If you can't find the answers there, they are likely near the
beginning/end of each of the subsequent sections.
-
Next, dive a bit more into the technical details. Can you figure out
what the authors are actually doing? Try to understand what the
evaluation is doing. Can you follow the analysis?
-
Be critical when reading. Look for logic that doesn't follow and unanswered
questions. Research doesn't necessarily provide perfect answers,
so there will likely be weaknesses in papers. These weaknesses can actually
become the motivation for a future project/publication (see question 2
of the Heilmeier Chatechism).