T-701-REM4 Research Methodology / Aðferðafræði Rannsókna



6. Scientific Papers: Contributions, Results, Data Presentation



Typical Structure of a Scientific Paper
Remember the Audience
Contribution Section
Evaluation & Results
Presentation of Results & Data
Data Presentation for Human Consumption
Next Project: Write Related Work Section













Typical Structure of a Scientific Paper
  Abstract This section is key - it's a mini-summary of your paper, intended to allow others to decide whether your work is relevant to their work (and whether they should read on)
  Introduction Overall context of the work, short summary of related work and a presentation of the motivation for the work - the problems that are to be addressed
  Related work Relatively dry discussion of prior work and how it is inadequate in addressing the problems that your idea addresses.
  Contribution Your idea. This is the topic of the paper. Describe it as clearly as you can.
  Evaluation How do you make sure your idea is a good one? How do you convince others that it's a great idea?
  Results Present the results so that they support the claims made throughout - and support the idea that your idea (the topic of the paper) is worth publication
  Discussion Optional section - sometimes things that you can't fit anywhere else, but cannot wait to tell the world about
  Conclusion This is the conclusion you draw from the work, as presented in the paper. Based on what has been said in this paper, what conclusions can you draw? This is often a semi-summary of the paper.
  References A structured list of publications that relate to the work described in the paper.














Remember the Audience
  Ask before you start your research This will determine your research context, experimental paradigm
and the emphasis or slant you choose for your work.
This is especially important if you are working in interdisciplinary research or on projects that can appeal to more than one scientific community.
  Ask before you start writing your paper Select the journal / conference first
Do a background search on papers recently published there, to verify that your background section and description of work fits into their context (less important for journals)
  Ask again when you do your background research It is good to remind oneself every now and then about who one wants to read the paper. A very good time to ask this question is right before starting to do background research - online search for related material.
  Ask yet again when you write up your contribution The audience determines what kind of background material is given and what you have to explain.















Contribution Section
  Contribution Section Describe your idea - not how it has been evaluated. You may reference related work, but keep it to a minimum and specific (i.e. no "review of what so-and-so did" in the Contribution section!).
  Method Isolate the key ideas
make sure you give them all the space they need.
Trim off superfluous ideas.
  - This can be painful: How can you trim off those little cool things that somehow you just know the world absolutely needs to know about? Trick: It is often a good idea to pretend you are going to write another paper, where you can put your shaved-off ideas.
  Do not cross-reference too much between sections It is a common difficulty to keep the discussion on the topic of the contribution in the Contribution section, without referencing the evaluation results. This should be avoided at all cost. The best papers are the ones where the results of the evaluation are not given away before the Results section. And each section can be read mostly without having to jump back and forth all the time.
  This is your stuff Make it look as good as it can look!













Evaluation & Results
  This section proves that your idea is great  
  Describe the evaluation method thoroughly but succinctly Nothing is more annoying than long-winded discussion of the evaluation method. Just the facts, ma'am!
  Evaluation method description: should have a one-to-one correspondence with the results section This means that any table, graph, or illustration in the results should have a directly corresponding statement/motivation/ discussion in the Evaluation section.
If you feel like you have to include tables whose existence are not discussed -- and hence not justified -- in the evaluation section, make sure they are either motivated by a surprise finding or else put them in a Discussion section.














Presentation of Results & Data
  Be clear Put yourself in the shoes of the reader. Try to imagine what is essential information to grasp the findings in as short a period as possible.
  Select the form for your data Tabular, line graph, bar graph, dot graph, pie chart, scatterplot
  Always summarize what the tables/graphs/etc. say with your own words, before drawing conclusions Do not say "As can be seen in Graph 1, my routing algorithm works best." Say, "As can be seen in the comparison between the three routing algorithms, A, B and C, the algorithms with partial information about network topology have an advantage in networks above a certain size (point X, Graph 1). Among these, my algorithm, C, gets the best result."
  Captions: Equally important as the graph Make sure your graph contains all the information necessary to interpret the graphics: Title, caption and legend should be written with the same care as the title of your paper and your abstract!!
  Two principles The clearest presentation - make the job easy on the reader
The biggest impact - try to make the point as strongly as you possibly can














Data Presentation for Human Consumption
  Scientific papers are for people People have limited capacity processors. Make sure you direct them to the most important points through the right data/information presentation methods.
  Select the right format Unidimensional temporal data: 2-D graph
Three-dimensional data: 3-D graph from various angles
Multidimensional temporal data: Many 2-D graphs
Dynamic multidimensional data: Illustrations, examples
  When not to use graphs When the graph contains too much or too little data
  Common mistakes in graphs Equating two-dimensional space with one-dimensional space
Forget to indicate that an axis does not start at zero
Using pie charts for open-ended scales
  9 tips to make your graphs great
(based on: source)
1. decide on a clear purpose
2 . convey an important message
3. draw attention to the message, not the source
4. experiment with various options and graph styles
5. use simple design for complex data
6. make the data 'speak'
7. adapt graph presentation to suit the data
6. ensure that the default visual perception process of the reader is easy and accurate
9. avoid ambiguity







(Graphs from: http://www.statcan.ca/english/edu/power/ch9/first9.htm)





Remember: Whether you like it or not, pople make inferences from your data. 
Make sure they make the right ones.











Next Project: Write Introduction & Related Work Sections
  When you rewrite your Title, Abstract, Introduction based on criticism Remember: If you get feedback from others, you should not judge the judge -- you are using their input to improve your paper! You should thank them and try to figure out what they mean, why they say what they say, in a way that you can use to improve your paper!
  Write a complete literature review section Related work and References sections. Due: This Friday, Oct. 27.
  Remember: keep everything a bit shorter than a typical paper!