Note: translated with the help from ChatGLM3, minimalist revision only. The quality of translation may vary.
Spent the whole of May helping people to refine their theses.
In our educational system, no one has ever taught students how to write a thesis before they start writing it. This has led to many problems: students don’t know how to write, and advisors have a hard time correcting them; they keep revising and revising, and after a few months, still get arrested during defense. The vast majority of the problems with the theses are common, but they need to be taught one by one, which is time-consuming and labor-intensive, and the results are not very good. I have summarized some common problems for those who may need it, for self-examination and improvement.
In order to make it clear to students who have no experience in writing, the problems I have listed are all very surface-level, simple, and formulaic. They can all be quantitatively analyzed by running just a few lines of script. Please note: out of these phenomena does not mean that the thesis will pass, and vice versa does not necessarily mean that it is not an excellent thesis. I hope to inspire readers to think for themselves about why most problematic theses are like this.
Law 1: The frequency of the term “extremely” (
Alternative terms to “extremely” can certainly be used in scientific writings for “in limits”. But more often, the use of words like “extreme” “very” “significantly” and so on in a thesis is just to emphasize. In language, these words serve to express emotions and are subjective, providing little help in objectively describing things. However, the thesis, as the carrier of science, must remain objective. When writing these subjective descriptions, students need to be especially cautious, and should carefully check whether they have deviated from the basic requirement of maintaining objectivity. In general, when someone makes such a mistake, they tend to commit a string of similar errors, which can leave reviewers with an unforgettable experience of unprofessionalism.
Similarly, it is recommended that students check whether other forms of subjective evaluations, such as “widely applied” or “made significant contributions” appear in the thesis, or words to that effect. These are comments that only reviewers can say, as reviewing requires expert opinions. The authors of the thesis should discuss objective rules, not to advertising themselves; unless they provide a clear and objective basis, they also have no qualifications to talk about others.
The solution is to provide evidences. For example, “high efficiency” can be changed to “achieved a 98.5% efficiency in a (particular) test”, and “widely applied” can be changed to “applied in the fields of sentiment analysis [17], dialogue systems [18], and abstract generation [19].”
Law 2: The variance of the number of fields (
You are too lazy to deal with the references, so you copy the BiBTeX entries from DBLP, ACM DL, or the publisher’s website and paste them together, leaving the formatting of the reference section to the control of a random bst file. The end result is that the quality of the references varies: some fields are missing, and some have many fields, meaningless. Therefore, I say counting the number of fields in each reference, and calculating the variance, can roughly reflect the quality of the writing. The issues in the reference are truly intricated, and it takes endless time to clean them up. So it can reflect the author’s level of payment into the thesis, quite noteworthy. For the same conference, some only write the full conference name, some add the organizer in front, some add an abbreviation in the end, some use the year, and some use the session sequence. Already with an DOI, but some give an arXiv URL at the end. The title has mixed capitalization. Sometimes there are even a hundred different ways to write “NVIDIA”. Therefore, how carefully the paper is written, how much time is spent polishing, can be clearly seen from the references. During the degree defense, time is tight for the reviewers, they must be checking from the most weakness, you are impossible to fool around.
The solution is to put in the effort to go through it, which requires not being afraid of trouble and spending more time. The information provided by the publisher on the website can only be used as a reference. BiBTeX is just a formatting tool, the actual display always depends on the editing from author.
Law 3: The count of underscores (_) in the text (
There is no such punctuation in Chinese, nor in English, and no human language AFAIK has this character in the vocabulary. The only chance to bring it into a thesis, is if the author treated the thesis as a coding documentation, and adding code snippets into the text. This law is the most accurate one, with almost no exceptions.
The purpose of writing is to convey your academic views to others. To efficiently convey to others, the thesis must first abstract and condense the knowledge. One should never present a bunch of raw materials and expect the readers to understand it on their own, otherwise, what is the point of having an author? The process of abstraction means leaving behind the implementation details that are irrelevant to the narrative, and retaining the ideas and techniques that are eye-catching. Amongst the details, code is the most gory. Therefore, none of a good thesis will contain code, and good authors will never have any concept with an underscore symbol in the name.
The solution is to complete the process of abstraction and the process of naming abstract concepts. If the code, data, and theorem proofs need to be retained, they should be included in the appendix, and should not be considered part of the thesis.
Law 4: The proportion of western characters (
It is specifically required to write in Chinese. But poor writers love to mix things up in the language. Law 3 is to criticize the mixing of code snippets (which are not in human language), and Law 4 is to criticize the mixing of all foreign expressions (which are not in Chinese language). I can understood why you keep “AlphaGo” (阿尔法围棋) untranslated, but I really cannot when you keep “model-based RL” (基于模型的强化学习). This year, two of the students wrote like this: “降低cache的miss率” (reduce the miss rate of cache), the theses are total mix-up. Come on, you are about to graduate, but cannot even speak your native language fluent! It is normal if you do not know how to express a concept in Chinese. If the concept is important, you should put some time into finding a common translation; if it’s a new concept, you should put some thought into coming up with a professional, accurate, and recognizable name; if you’re too lazy to find a name, it only shows that you don’t give a fuck to the bullshit you wrote.
I have a suggestion for those who found themselves having the bad habit of mixing languages: For every western character used in the writing, you should donate a dollar to charity. It is not prohibited to write in western, but it should let you feel some pain in the ass to avoid abusing.
Law 5: The average length of mathematical symbols (
Einstein’s writing of equation:
The same equation, if relativity is found by our student:
Oh please, at the very least, always write text in text mode! Wrap in \text or \mathrm macros, otherwise it becomes
Alright, let’s just appreciate the difference from the masters. Why do their works become timeless classics, while we’re still struggling to pass the thesis defense? When doing work, we didn’t take the time to simplify: text is full of redundant words, the images are intricated, and even the formula are lengthy. The higher the complexity, the worse the academic aesthetics left.
Law 6: The maximum number of charts appeared on a single page (
When you can’t find enough pages, it’s always faster to spam charts. As soon as the students come to the experiment section, they pick up the profiler and take screenshots. Suddenly, a dozen of images in a row with dazzling blocks all over the screen weightened the thesis full and complete. As a result, the reviewers are fucked up: what the hell is that? Piles of colorful blocks, lacking accompanying text descriptions… Reading this is even more challenging than IQ test questionaires! Only the author himself and God knows the meaning, or actually more often, only the God knows. If the thesis is of decent quality, the reviewers just ignore the cryptic part to avoid making the meeting awkward. If not, the piled charts could be the last straw put on their tolerance.
If the value of
Secondly, it is important to evaluate the intuitiveness of charts and tables. The meaning of the charts and tables should be clear and easy to understand at a glance. When charts and tables are complex, it’s important to highlight the key points and guide the reader’s attention. Alternatively, provide detailed captions or reading guidelines.
Law 7: The number of screen shots (
In line with the previous Law: It is recommended to pull the “PrintScr” key out of your keyboard while writing – the Demon’s key. Due to the temptation of this key, students become lazy, and constantly take screenshots of software interfaces such as VCS, VS Code, or the command prompt, and insert them into the paper as figures. As a result, it often leads to issues such as excessive use of large or small fonts, mismatched fonts, inconsistent image styles, missing description text, lack of key points, and simple piling up of raw materials, etc. Although the issues may not be particularly serious, they will definitely leave a poor impression on readers. Whether the meaning of the charts is clear or not, they will first leave a sloppy impression.
Screenshots on the IDE, they will certainly be code! As per Law 3 and 4. Among all the captures, capturing the debugging logs from the terminal is the most untolerable. It’s still better to capture the code instead. Code, even not good for writing, still has well-defined syntax and semantics, while debugging logs are totally waste of paper and ink.
Law 8: The proportion of the page number on the last page of “Related Works” (
I always check this metric first, because it’s the most easy to: find the last page of the related works, stick a finger in, close the cover, and check the thickness of the two divisions. This only takes five seconds, but it constantly finds out who spam pages in the related work section. If the thesis writes too much in related works, it is always bound in poor quality, because writing related works is much more difficult than writing your own work! Just imagine, during the defense, the reviewers have to flip through your paper and listen to your presentation, dual-threaded in their minds. But on your side, you spend 20min introducing other’s works that are well-known, cliched, or unrelated before going into your own. The result must be terrible.
Survey of works requires a lot of devotions and academic expertises. Firstly, it’s necessary to gather and select important related work. Here we should not discriminate against institutions or publications, as the evaluation of specific work needs to be based on objective scientific assessment. However, exploratory work tends to (in a statistical sense) be published by leading institutions in important conferences and journals. Our students always cite papers from “Indian Institute of Engineering and Technology”, published on an open-access journal from Hindawi, while overlooking the fundamental works in the field. It is only showing themselves not yet on the cutting-edge of the field. It’s not easy to list a bunch of important related works while avoiding low-quality ones. This is a task that requires enduring academic expertise. Students who just start out is expected to fail short when trying to put too much pages in the related work section.
Summarizing the selected related works is a highly scientific work in itself. If done well, its value is no less than any research point in the thesis. Students don’t have so much time and energy to go through all the related works, or they haven’t been trained to summarize properly. So most of the time, they simply enumerate the works. Among those who have mastered it well, some can write like this:
** 2 Related Works **
2.1 Researches on A
A is important. There are many works on A.
Doe et al. proposed foo, and it is zesty.
Roe et al. proposed bar, and it is nasty.
Bloggs et al. proposed buz, and it is schwifty.
However, prior work sucks. Our work rocks.
2.2 Researches on B
B is also important…
Such writing methods at least ensure that the related works are listed with a beginning and an end, and that they are centered around the main work. It’s commendable that some students can write in this level, but there are many who can’t even manage to do that, and only list several works that are not even related to the main topic. If review experts have the intention to torment the author and make him/her look silly, they could ask questions like: “What’s the key technique behind this work?” “What’s the difference between this work and that one?” “How is this work related to the main topic?” After asking such questions one by one, the author would probably be so embarrassed. After listing so many related works, some of which the author may not have even opened before, it’s not uncommon for people to just copy and paste the BiBTeX entries directly from Google Scholar into the paper without really understanding what they mean.
The three questions I listed correspond to three essential elements of related works: key scientific concepts or discoveries, taxonomy, and relevance to the main topic. A good summary should achieve the following three goals: explain the key concepts or discoveries of each related work in a single sentence; organize related works into multiple categories, with clear boundaries; and ensure that the logical relationship between each category is clear and the narrative thread is evident, closely related to the main topic. This requires the author to have strong scientific abstraction and concise language abilities, which need a thorough drafting and refinement process to produce a coherent and clear writing.
For most students, summarizing three research points in the first chapter of a thesis is already a difficult task. So, I especially advise students not to introduce related works lightly in the second chapter, unless there is a special understanding, as it challenges their writing abilities. It’s better to specifically mention related works where they are needed and provide targeted explanations during the process of discussing each research point. Otherwise, students without scientific writing training might easily turn the related work chapter into a paragraph that doesn’t make much sense, just to spamming words. Since it’s a spam to meet the word count requirement, students may feel compelled to include more and more related work, even if it doesn’t fit well with the main topic. Lengthy content usually implies a lower quality of the thesis, indicating that there must be something wrong with the thesis.
Law 9: The length of the longest consecutive block of experimental content (
Another area where spaming is heavily used is in the experimental section. After all, compared to writing the thesis, conducting experiments seems more comfortable. Adding unnecessary metrics, creating bar charts for several related works and compare them, all done with reused matplotlib scripts. Alternatively, you can list the data in a table, until not fit on the page. Additionally, there are many students who want to include three research points but can’t find enough space for them, so they extract the experiments and write them as the fifth chapter, the third research point in the thesis. As a separate chapter, the continuous experimental section is at least 20 pages in thickness, so as not to appear thin. The harm caused by listing experiments together with related work is just as serious. Experiments were originally used to verify research content, but now some research content has lost its experimental support. A bunch of disorganized experimental data is piled up together, creating irrelevant and meaningless content once again. This law proposes indices specifically designed for checking this common bad practice.
The underlying mindset can be attributed to two main factors.
The first factor is the prevalence of empiricism, which views experiments as the equal of scientific research and thereby ignores the value of establishing and communicating scientific abstractions. As an important infrastructure for cutting-edge research in physics, the Beijing Electron-Positron Collider has already generated over 10PB of experimental data. Many important scientific discoveries originally come from these raw data. Printing and compiling these data into a book will definitely become an invaluable work of science, and every physicist should have a set, right? Nope. Because the bandwidth of human communication using symbol language is only about 39bps, and the maximum amount of symbol data a person can accept within a 100-year lifespan is only 15GB. It is impossible for anyone to read through these data in their lifetime, let alone discover anything from it. What has true scientific value is not the data presented in graphs or tables, but the abstract scientific concepts verified by the data. The role of the experiments is only to verify the scientific ideas. Experiments themselves can be said to be not even first-class citizens, and in many cases, second-class. In scientific research, it is often the abstract scientific ideas that are proposed first, and then experiments are designed specifically to verify their authenticity.
The second factor is that theses are mistaken for Proof of Works (PoW): The experiments during my postgraduate studies were the direct proofs of the work I did, which I listed as my work report to my boss. This is to misunderstand one’s own identity in scientific research. Graduate students are not laborers. Scientific exploration is not a form of labor for hire, and the effort put into scientific work usually does not yield a corresponding reward (or else it would be considered as “paper spamming”), the workload itself has limited significance. Papers should serve as the carrier of creative ideas, rather than a record of work hours for an employer - the latter will only be cared about by your employer and no one else. Your advisor, reviewers, examiners, and potential readers of your thesis, are they your employers? What do they care about and what do they want to see? Before writing a thesis, it is essential to first understand this point.
Of course, students will also ask: “You know nothing about my situ! My work is just too trivial to collect three research points, so what else can I do?” Here, I must remind those who haven’t graduated yet that you should avoid ending up in such a situation unintentionally. When writing a thesis, it is essential to plan in advance. Before starting the research work, layout the rough content of three research points within the same field, and carry out one or two research projects under the guidance of your advisor specifically for the purpose of this layout. As soon as you complete a valuable research work, it is easy to extend three research points, such as: using the research work itself as the second point; searching for the defects of the research work and further improving it, which is the third point; summarizing the ideas of the two research works, proposing a scientific, concise metaphysic (model, paradigm, principle, style), which is the first point. The metaphysic part can be completed entirely during the process of writing the thesis.
In the worst case, you must also adhere to the principle of truthfulness. Do list the works you have done, and use the experiments to verify each. Do not spam experiments to stuff the chapters.
Law 10: The minimum distance between titles (
Unlike all previous laws, the tenth law is a recommend, so it proposes the only positively relevant index. It is not easy to write an thesis with a clear structure and explicit clues. However, there is a quick and effective way: Force yourself to clearly list out the clues for every sentence, and write them right under the corresponding titles in all levels. Let’s take related work as an example:
2 Related works
This chapter introduces the important related work in a certain field, so that readers can understand the content later. From one aspect, the research work in a certain field can be roughly divided into categories of A, B, and C; from another aspect, these works can be divided into sub-aspects, foo, bar, buz. This chapter will introduce these aspects in a certain order. The author suggests that readers who are interested in a certain field can read [18] for a deeper understanding.
2.1 A
One of the more intuitive ways to solve problems in a certain field is to adopt method A. Method A has characteristics of blah, blah, and blah, and has been widely used in various fields such as blah [19], blah [20], blah [21], etc.
John [22] proposed Afoo, and… but it still lacks…
Addressing this concern, Jane [23] proposed Abar…
……
2.2 B
A is effective on… However, … As a result, B is in turn extensively researched since 2006. Differs with A, B is specialized on D, E, and F…
……
Using this method, readers can obtain a clear guide to the later content from the title at each point, which will significantly improve the reading experience. More importantly, consciously following this fixed writing method can force you to organize the narrative of the thesis. As long as it can be written out, it can ensure that the thesis is structured, organized, and reasonable. This writing method brings the characteristic that there is a guiding text below each title, without any two titles being adjacent. Therefore, I use the minimum distance between titles to describe this characteristic.
Conclusion
A thesis paper is an unique material in a student’s life, representing their academic level when they obtain their degree, and is always available for reference in the database. Although my doctorate thesis has received high praise from everyone, there are still some errors in my thesis that need to be corrected; Having my thesis published and proofread it again makes me feel very regretful. I suggest that students should treat writing a thesis as a major event in their lives and put as much passion and effort into it as possible.
In this post, I simply discussed the process of writing a thesis from a critical perspective. Relatively, it is still not so constructive. I hope that after reading these contents, students can reflect on them and understand the basic requirements of the thesis, and establish a basic aesthetic judgment of the thesis. Recognizing one’s own shortcomings is the basis for progress; understanding that writing a thesis requires many restrictions and requires a lot of hard work to establish the correct mindset.
Last month, I was invited by Li to give a report on the experience of writing a thesis and defending it to graduates, which included more constructive feedback. Because the time was urgent at that time, many of the content was not fully developed or even incomplete. Writing a thesis is an important task that requires long-term planning. The most important aspect of writing a thesis is to organize and plan well for all the research work, and this requires cautious planning from the beginning of the research. Therefore, I think that such reports shouldn’t be given to graduates in May, but should instead be given to incoming freshmen in September. I plan to complete the content before September and share it then.