CS 5974: Writing in Computer Science
Fall Semester, 2022

Jingyi's Annoying Papers

1. NikolopoulosEtAl_AISTATS_2021.pdf
   • Not defining the problems to be solved
   • There is no clear definition of the problem to be solved in this article, and there are many limitations listed that were not stated at the start of the article.
2. EscobarEtAl_medRxiv_2020.pdf
   • Inadequate description of the proposed methods
   • This article uses a combined machine learning approach to predict the risk for each individual. However, it is not clear in the article which feature of the data were used to make the predictions and which features contributed more in the predictions.
3. FangEtAl_preprint_2020.pdf
   • Lack of precise definition
   • There're so many parameters and notations in this work! It is difficult to remember all the notation and their meanings.

Reza's Annoying Papers

1. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding in the first subsection of section 3 named "model architecture" the authors skip a lot of detail referring to another famous paper. However they fail to cover some of the detailed aspects of the attention and output links which are not covered in their referenced paper. This made a lot of trouble when I was trying to imagine the data flow within the matrixes impossible so I had to look for code bases and some illustrative representation of the model on websites and blog posts.
2. The "Entrez Help" and "NCBI Help Manual" books for using their entrez API Not technically papers, but books that document how to use the popular NCBI API. They fail to show the overall database design (easily shown with a UML diagram) and the relationships between some of the tables. The API parameter descriptions are also missing and are only available when running help on the software itself, usually without any mention of possible variations of these parameters. Furthermore, the texts are designed for descriptions for the search and usage of the website and with little to no overlap with the API fields.
3. Unfortunately I could not find the original paper since I had forgotten its title even after going through a lot of papers (I guess it being annoying didn't help in me preserving it both mentally and physically) I recall the paper was on using deep learning on NLP and their model had multiple sections. For the input, weights, and outputs of each section they used an arbitrary letter e.g. X_i for the first input S_i for the second input and so on. The notation descriptions were defined within the text only (no table) which made things like trying to understand their custom cost function (which included around 18-20 primary and 10 secondary symboles) similar to decoding an enigma with a table that I had to write on a piece of paper myself.

Xiao's Annoying Papers

1. https://doi.org/10.1111/j.1365-294X.2006.03167.x I generally like this paper, but it has a tendency of using long sentences and complex clauses, which prevents me from capturing the key points quickly sometimes.
2. https://link.springer.com/article/10.1186/1471-2148-9-259 This is a good paper and has been cited many times, nevertheless I have some complaints about its writing. There are terms that are not well explained but straightly put in the text, and readers without the background will need some time to dig into the meaning. For example, the term "Tarsiiformes" appears 2 times in the paper, and each time it appears, the authors actually discuss only the tarsiers without mentioning Tarsiiformes again. In fact, tarsiers are the only living members under that infraorder Tarsiiformes, but this relation between Tarsiiformes and tarsiers has never been introduced in the paper.
3. https://www.sciencedirect.com/science/article/pii/S0277379117308211 I am not really sure if this is an annoying aspect of this paper, but I do want to draw some attention to this topic: in this paper, I noticed that the content of the different Result subsections is not well balanced. For example, Section 4.3 (Craniofacial shape) is very short and does not provide as much discussion as other subsections. I understand that different results have their own significance or importance, but putting a very short section between several long sections still seems a bit strange to me.

Yoonjin's Annoying Papers

1. *Sexton Et al., Three-Dimensional Folding and Functional Organization Principles of the Drosophila Genome, Cell, 2012 *did not have provide clear definitions of the terminologies that are used in the paper. For example, In figure 3C, the author used "gene expression" without defining the method to calculate the gene expression. They also did not explain the reason why they decided to use median value to compare epigenetic labels not mean value.
2. *Lo and Park, Modeling the spread of the Zika virus using topological data analysis, PLoS ONE, 2018 *decided to use models and terminologies before they define a later section in the paper such as "mosquito occurrence". The result section of this paper did not provide clear direction or reasoning on how the result can help track future contagions.
3. *Chin and Bouffanais, Spatial super-spreaders and super-susceptibles in human movement networks, Scientific Reports, 2020 *contains a lot of useful information, covers many fields, and is well written in general, but I wish it was structured differently. Since the article has many sections and various information, I get lost track while reading and had to read the related section multiple times.

Badhan's Annoying Papers

One thing that has been in my mind for a long time is that some papers have their result section before the Material and Method section and just after the Introduction. It is the style of some journal papers that I saw. The *GCNG* paper has this style. The matter is not annoying though. Sometimes I jump to the method section after reading the introduction and then return to the result section. Is this the way to read this sort of paper? Again, while writing, how do I keep a good flow of my writing in this style?

1. *GCNG*
   • This paper has so much information with many sections, it was hard to keep track of the main goals.
   • In page 5, they provided a list of AUROC/AUPRC ratios, which can easily be added as a table instead of listing it in a paragraph which is not a good idea for data representation.
   • The result section was so long that, the last two subsections: inferring causal interactions, and, functional gene assignments felt like excess information and couldn't keep a good reading flow of this section.
2. *Evolutionary adaptation*
   • Very brief background/introduction (just 2 paragraphs before narrating the aim)
   • No figure for the workflow
   • No figure for the models described in the result section
3. *LEMON*
   • Very short introduction/background to give a better overview.
   • Problem(s) to be solved is not clearly defined in the introduction section. They just mentioned what they did.