Literature Reading in Three Steps——How to Read a Paper

        (Today I plan to officially start recording my computer learning process, and by the way, record the writing skills of CSDN. Since my MarkDown writing skills are not proficient, I still use the default article editor for the time being.)

        The first to be written is a paper reading " How to Read a Paper " by S. Keshav and David R.

introduction

        As a graduate student, how to read papers efficiently is a long-standing and difficult problem. When I think of myself as a sophomore and junior, I was still a teammate . Neither research nor research has improved. This article introduces a method called "three-pass" to read papers, a simple and efficient method, which is simply a three-step method .

Literature reading in three steps

        What is the three-step method? The key is that when we read a paper, we don't start from the beginning and go all the way to the last reference, which is inefficient and tedious.

        We should read the paper three times, each time to achieve our specific goals and build on the previous one for the next reading. The first goal is to roughly understand the gist of the thesis, the second goal is to understand and master the content of the thesis (but not its details), and the third goal is to understand the thesis in depth.

A round of reading - The first pass

        The first step is to do a quick scan of the article, "Overlook" the article ( summer reappearance ). The process takes about five to ten minutes and includes the following steps:

1. Carefully read the title, abstract and introduction
2. Read chapter and section headings (but not what follows them)
3. read the conclusion
4. Go through the citations , mentally ticking off the ones you've already read

        Ok, now that you have gone through the above 4 steps, you should have an approximate number now, now ask yourself the following questions:

• What type of document is this ? Is it a numerical experiment report? Or an analysis of existing systems?
• To which papers is this article related , and what theoretical basis is used to analyze these issues?
• Are the various assumptions in the article really reasonable ?
• What is the main contribution of the article ?
• Is the writing clear and well written ?

        OK, now ask yourself, do you still want to read this article. You can choose to just slip away, because you are not interested in this article. For this kind of papers that do not belong to your own professional field, it is enough to finish the First pass.

        By the way, this technique can also be reversed when you write a paper, because when you write a paper, you can also imagine that most readers or reviewers will only read it once. So we pay attention to coherent chapter and subsection titles, and write concise and comprehensive abstracts. Because a reviewer can't understand the main points of the article in one read, the article is likely to be rejected; if an ordinary reader can't get what you are writing within five minutes, then it is likely to give up reading.

Second round of reading - The second pass

        In the second pass, we have to read the paper carefully, but we still have to make some trade-offs, such as ignoring the proofs of theorems and steps in it (of course, if you can’t even read the theorems, it is recommended to read it separately). You can jot down key points in the margins as you read. This mainly includes the following steps:

1. Look carefully at the figures, graphs, and illustrations in the paper . Pay particular attention to the graph, are the axes labeled correctly? Are the results shown with error bars so that statistically significant conclusions can be drawn? Common mistakes like this can separate a shoddy paper from a really good one.
2. Mark relevant, unread references for further reading. This is a great way to get a sense of the background of your paper.

        This second reading usually takes an hour (for a relatively short paper), after this pass, we should be able to grasp most of the content of the paper, and we can summarize the main idea of ​​the paper to others, And apply relevant conclusions from the passage to support your description.

        But in fact, sometimes we don't understand the paper at the end of the second pass. Because the topic is new to me, there are a lot of unfamiliar terms and acronyms; or the author cites proofs or experimental techniques that we don't understand, so most of the content is incomprehensible; or the article itself writes It's not good, with unsubstantiated claims and a lot of forward-looking references; or it's just that you played the game all night yesterday, and the next day is completely exhausted when you go to work .

        Ok, now you have three options:

• Don't read it , the second reading is enough to get enough references to complete your research work.
• Read it again later , here refers to the Second pass again, maybe after I have figured out other knowledge, I can come back to understand the paper better.
• Start the third round of reading.

Three rounds of reading - The third pass

        Well, it’s all here. I believe you must be the person who was asked by your tutor to reproduce the paper, or for personal learning purposes, you need to understand the paper word by word, or you are a reviewer, but you are all reviewers . Contributor, do you still need to read this article ?

        To fully understand a paper, the most important thing is to read it the third time. The key to the third pass is to try to re-implement the paper and reproduce the experimental process under the same assumptions and experimental environment as the author. By comparing this reproduction with the actual paper, not only can the innovation of this paper be felt, but also its shortcomings can be recognized.

        This reading, you have to pay great attention to detail and have to test and question every assumption in every statement in the paper . Consider how you would come up with such an idea if you were writing the essay yourself. This method of combining reality with words on paper is helpful to understand the proof and writing techniques of articles. We can even learn such writing techniques and use them for our own use to create our own papers. During this process, we should also write down our views on future work . After all, we read a paper with a high probability because we want to improve and innovate on its basis.

        For novices, if it goes well, this level will take about four or five hours, and for veterans, it may only take a few hours. At the end of this pass, we should be able to reconstruct the entire structure of the paper from memory, and say the advantages and disadvantages.

        In particular, we should be able to use experimental or analytical techniques to identify implicit assumptions, missing related work, and potential problems in modified papers, and these are often things that can be written in your next paper .

do a literature review

        After solving the problem of how to read, let's analyze some actual situations.

        Generally speaking, no matter what type of article it is, a literature review is a trap that cannot escape. If you want to do your experiments well and write a good paper, then you must read a lot of literature to support your arguments. Next, on how to read a large amount of literature, the following suggestions can also be given:

1. Use Google Scholar or other academic search engines to find the three or five papers you want. Take a First pass on each paper , understand the current research situation, and read the related work of these articles. We'll find a general trend in recent research and even some ready-made literature reviews of the current state of research . If you can actually find such an eligible review, then your job is done, go read it!
2. Unfortunately, we were unable to find such a review, so we had to look elsewhere. We are going to find common citations in the references of these articles, or duplicate author names, which are often key papers and researchers in the field . Now we can download these papers, and follow the websites of these key researchers to search for their recently published papers. This will help us identify the top conferences in the field , as the best researchers usually publish at top conferences.
3. The third step is to visit the websites of these top conferences and browse the recent proceedings. Quickly browse and sift through some high-quality related work. So far, the literature you have found constitutes the first batch of materials for your literature investigation, and we can carry out the Second pass . Also, if these papers all cite a key paper that was not found before, we can repeat the second operation and iterate.

The original author's experience

 “I have used this method for the past 15 years to read conference proceedings, write reviews, do background research, and quickly review papers before discussion . Being able to estimate the time it will take to review a group of papers. Also, I can adjust the depth of the paper evaluation according to my needs and how much time I have."

        In addition, the author provides some other reading skills and research skills literature that researchers can learn.