Originally published at DEV.
When I first began, it felt incredibly overwhelming because I lacked context and dived right in. This post aims to share my insights to simplify the journey for others just starting.
TLDR;
Pass 1: abstract, keywords, introduction, conclusion
Pass 2: methodology, experiments, results MOU Pass 3: Summarize, identify key strengths, weaknesses and possible directions of future work.
The Process
In the initial stages, divide your reading into a three-pass process:
Pass 1
Skim through the entire paper. Read the abstract, keywords, introduction and conclusion. Make notes of:
- technical terms that are new to you
- key contributions and claims made by the paper
- questions
Before you begin Pass 2, study about the technical terms you need help understanding. This will clear a lot of the questions you have in Pass 1.
Pass 2
With a newfound understanding of the technical terms, now look at the methodology or algorithms(if any) mentioned by the paper. Next, move on to the experimental setup and results. Makes notes of:
- technical terms that are new to you
- any new vital contributions you see to add to the list from the first pass
- do you have answers to your questions from the first pass?
- new questions
Pass 3
These are some questions you can ask to summarize your understanding:
- What is the problem being solved by the authors?
- What is the proposed methodology?
- How were the experiments structured?
- How do the results correlate?
- What is the possible direction for future work?
Where the learning really begins
One of the best ways to build more context about the problem being solved is to read the section on related work. But beware, it is easy to get lost in the rabbit hole (however, a fruitful diversion in this case) because there is a wealth of knowledge. This helps build an understanding of where the paper falls in the current realm of research.
Another essential technique to apply after reading the paper is think about the possible areas in which the work presented by the paper can be further extended. More often than not, published works can be incremental updates to work done in a previous paper. Identify possible shortcomings of an approach might give you a direction for the following line of your work!
Devote time towards exploring the codebase of the published work (if available). Running your experiments and trying to reproduce the paper results is an excellent way to get more insights.
This method is a subjective process and I would love to hear from you if you have ideas on improving this process.