How to Write Good Research Articles

1
How to Write Good Research Articles

• Dr. Ashfaqur Rahman

2
Presentation Outline

• Publication Requirement
• Kinds of Scientific Publications
• Where to publish your work
• Plan your writing
• Structure of a Paper
• Case studies
• Q A

3
Publication Requirement

• BSc Degree (Minor)
• MPhil Degree / MS by Research / MS (Minor)
• PhD Degree
• Full time researcher / Academics

4
Kinds of Scientific Publications

• Thesis
• Aspects to be Assessed for a Thesis
• background knowledge
• original contributions (amount of work)
• methodology
• presentation (writing)
• Conference Publications
• Focus on a piece of work with limited discussion
• Journal Publications
• More complete (extensive) discussion
• Book chapters / Text books

5
Where to publish your work

• Journals
• Ranking of journals
• Review process of journals
• Publication cycle
• Conferences
• Ranking of conferences
• Review process of conferences
• N.B. a good journal / conference tends to have
  rigorous review process and long review time

6
SCI Journal Citation
7
Important journals conferences

• Database
• IEEE Trans on Knowledge and Data Engineering
• ACM Trans on Database Systems
• Intl Conf on VLDB
• Software Engineering
• IEEE Trans on Software Engineering
• ACM Trans on Software Eng. and Methodology
• IEEE Intl Conf on Software Engineering
• Distributed Systems
• IEEE Trans on Parallel and Distributed Systems
• ACM Trans on Computer Systems
• IEEE Intl Conf on Distributed Computing Systems
• Computer Networks
• IEEE/ACM Trans on Networking
• IEEE INFOCOM
• ACM Mobicom, etc.
• .

8
Plan your writing

• Ask two questions before starting
• What is new in your work?
• What are you going to write?
• Emphasize on the originality and significance of
  your work.
• Organize your thinking and decide the structure
  (outlines) of your paper.
• Stick on your central points throughout the whole
  paper and remove all unnecessary discussions.

9
Reader-oriented Writing

• Purpose of your writing disseminating your
  research results.
• Dont write if there is nothing to write
• Dont make a simple problem complicated to fool
  people
• Dont hide technical details
• Reader-oriented writing Write in a way that
  would lead readers to follow your thinking, NOT
  in the way of your thinking.
• Well-organize your thinking
• Give enough and clear explanation (never leave
  reader to guess)
• Try to present your idea in an accurate way (no
  ambiguous)
• Always think how readers would interpret your
  writing (assume youre a reader)
• Use simple/ plain English
• Purpose of technical writing express your idea
  correctly clearly.

10
Structure of a Paper

• Title
• Abstract
• Key words
• Introduction
• Background
• Related Work
• System Model Problem Statement
• Methods / Solutions
• Simulations / Experiments
• Conclusion
• Acknowledgement
• References
• Average number of pages of a journal paper
• Average number of pages of a conference paper

11
Choose a Right Title

• The title should be very specific, not too broad.
• The title should be substantially different from
  others.
• Topology control for multihop wireless
  networks, IEEE Trans. on Comm, 93.
• Topology control of multihop wireless networks
  using transmit power adjustment, infocom00.
• Distributed topology control for power efficient
  operation in multihop wireless networks,
  infocom01.
• Avoid general / big titles, e.g.,
• Research on data mining,
• Some research on job assignment in cluster
  computing,
• A new framework for distributed computing,

12
Write a concise Abstract

• The use of an abstract
• for search purpose.
• giving readers a paper-summary before getting
  into details.
• An abstract should tell
• the problem that the paper discusses.
• the work that has been done, or method being
  used.
• original findings / achievements.
• An abstract usually does NOT have
• reference numbers
• multiple paragraphs

13
Choose a right set of keywords

• The use of keywords
• database search,
• categorizing your work (for editors to choose
  reviewers).
• The keywords must be specific and, as a whole,
  represent the main topic of the paper.
• Avoid using the words that are not the main
  topic, such as calculus, simulations, etc.

14
Examples of an abstract / keywords
15
Organization of your Paper

• Top-down writing method
• Planning sections and subsections
• Sketching use a sentence to represent the points
  (paragraphs) in each subsections
• Writing details expand a sentence in the sketch
  into a paragraph
• Adjustment break / merge paragraphs, add / merge
  sections
• N.B. keep a logical flow from section to section,
  paragraph to paragraph, and sentence to sentence.

16
Introduction the most difficult part

• Purpose of introduction
• Introducing readers to your problem / work.
• An introduction usually contains
• Brief background of the topic-area
• Existing work, which would lead to the importance
  / originality of your work
• Description of your problem
• Achievement / significance / brief-methodology of
  work

17
Related work and Reference list

• Proper selection of references
• Show your knowledge in the related area,
• Give credit to other researchers (reviewers are
  usually chosen from the references),
• Cite good quality work (particularly when citing
  your own work) and up to date work.
• Related work should
• Be organized to serve your topic,
• Emphasize on the significance / originality of
  your work (Introducing your work out).
• Format of references
• Consistent with the format, ordering, etc.
• Standard format of books / journal papers /
  conference papers, e.g,
• X. Jia, X.D. Hu and D.Z. Du, Multiwavelength
  Optical Networks, Kluwer Academic, 2002.
• J. Li, Yi Pan, and X. Jia, Analysis of Dynamic
  Location Management for PCS Networks, IEEE Trans
  on Vehicular Technology, Vol. 51, No. 5, Sep
  2002, pp.1109-1119.
• X. Jia, D. Li, X.Hu and D. Du, "Placement of
  Read-Write Web Proxies in the Internet", Proc of
  IEEE Intl. Conf. on Distributed Computing
  Systems, Phoenix, USA, Apr 2001, pp.687-690.
• Do NOT use non-standard abbrev.

18
Examples of reference lists
19
A Typical Review Form of a conference

• ACM International Conference on Image and Video
  Retrieval (ACM CIVR 07)
• Typical Review scenario
• Reviewer 1
• Rating
• Familiarity Please indicate your familiarity
  with the paper's subject matter
• This is my research area (4)
• Rating How would you rate the paper overall?
• Accept, but there are problems (4)
• Novelty How would you rate the novelty of
  the paper?
• New contribution (4)
• Impact Will this paper be important over
  time?
• Can't quite tell (3)

20
A Typical Review Form of a conference

• ACM International Conference on Image and Video
  Retrieval (ACM CIVR 07)
• Typical Review scenario
• Reviewer 1
• Rating
• Summary Please summarize the paper briefly
  This paper describes a method for recognizing the
  different dynamic textures present in an image
  sequence by directly comparing (without
  segmentation) the statistics (i.e., motion
  co-occurrence matrix) of the mixture with the
  statistics of the individual dynamic textures
  from a given known set of individual dynamic
  textures. A mathematical (geometric)
  justification of the proposed (linear) criterion
  is provided. Experiments have been carried out on
  synthetic mixture examples generated from samples
  of natural dynamic textures.
• Positives What are the most important
  reasons to accept this paper, in order of
  importance? Say whether the positives dominate
  the negatives (1-3 sentences) This is an
  interesting and difficult problem which has not
  been much investigated so far. The proposed
  solution is original and efficient. Satisfactory
  results are reported. The positives largely
  dominate the negatives.
• Negatives What are the most important
  reasons NOT to accept this paper, in order of
  importance? (e.g., the paper has serious
  technical mistakes, isn't novel, doesn't
  demonstrate its point by proofs, simulations or
  experiments, makes very unreasonable assumptions,
  etc.) If the overall conclusions are still likely
  to hold despite these flaws, please say so. Say
  whether the negatives dominate the positives.
  (1-3 sentences) Results on examples involving
  more than two different dynamic textures and on
  real videos including multiple dynamic textures
  would have been appreciated.

21
A Typical Review Form of a conference

• ACM International Conference on Image and Video
  Retrieval (ACM CIVR 07)
• Typical Review scenario
• Reviewer 1
• Rating
• Detailed Comments Please provide detailed
  feedback to the author(s) Page 1 MCM should be
  quoted as second-order texture measures. Page 2
  A formal definition of Vi should be given. The
  demonstration of Theorem 1 is not that clear to
  me (last part, formulae at the bottom of Pg.2 and
  top of Pg.3). The explanation should be improved.
  In case of several dynamic textures (more than
  two), how the number of textures could be
  determined? Is a block matching technique the
  best way to get reliable and accurate
  displacement fields on dynamic textures (for
  instance, is the brightness constancy constraint
  valid? Can the motion be always considered as
  nearly translational within a block?). This issue
  should be further discussed. Samples of computed
  displacement fields should be plotted.
  Experiments on real videos depicting multiple
  dynamic textures could be conducted as well. It
  is not that easy to precisely evaluate how
  similar the MCMs of Fig. 5 and 6 are. Differences
  of the MCMs (for each pair) should be plotted, or
  objective measures of the MCM differences should
  be provided. How is the ratio alpha of individual
  textures handled in the experiments? Is it
  estimated? Then, how is the mixture class with
  the true ratio selected? Or, are several
  predefined (or known) alpha values tested? Then,
  how the method would be applied to real cases
  without a priori information on the ratio alpha ?
  The reference to the work by Crivelli et al.,
  ICIP2006, on multiple dynamic textures could be
  added.

22
A Typical Review Form of a conference

• ACM International Conference on Image and Video
  Retrieval (ACM CIVR 07)
• Typical Review scenario
• Reviewer 2
• Rating
• Familiarity Please indicate your familiarity
  with the paper's subject matter
• Passing familiarity (2)
• Rating How would you rate the paper overall?
  
• I can't make up my mind (3)
• Novelty How would you rate the novelty of
  the paper?
• Incremental improvement (3)
• Impact Will this paper be important over
  time?
• Probably not (2)

23
A Typical Review Form of a conference

• ACM International Conference on Image and Video
  Retrieval (ACM CIVR 07)
• Typical Review scenario
• Reviewer 2
• Rating
• Summary Please summarize the paper briefly
  This paper presents a method of characterizing
  multiple temporal textures using present. Motion
  Co-occurrence Matrix (MCM). Some good experiments
  have been showed for the synthetic data.
• Positives What are the most important
  reasons to accept this paper, in order of
  importance? Say whether the positives dominate
  the negatives (1-3 sentences) MCM is well known
  method for characterizing temporal textures.
  Authors extend it for representing multiple
  temporal textures. Authors have analytically and
  experimentally verified the proposed method
• Negatives What are the most important
  reasons NOT to accept this paper, in order of
  importance? (e.g., the paper has serious
  technical mistakes, isn't novel, doesn't
  demonstrate its point by proofs, simulations or
  experiments, makes very unreasonable assumptions,
  etc.) If the overall conclusions are still likely
  to hold despite these flaws, please say so. Say
  whether the negatives dominate the positives.
  (1-3 sentences) The proposed method has only
  applied to toy problems (synthetic data). Not
  sure if it will work for real world problems.

24
A Typical Review Form of a Journal

• IEEE Transaction on Circuits and Systems for
  Video Technology
• Typical Review scenario
• Reviewer 1
• Rating
• Technical Content Tutorial 7
• Advance of Theory 7
• Advance of Application 6
• Presentation Clarity 5
• Organization 5
• Conciseness 6
• English 7
• Quality of References Completeness 8
• Missing Key References

25
A Typical Review Form of a Journal

• IEEE Transaction on Circuits and Systems for
  Video Technology
• Typical Review scenario
• Reviewer 1
• Detailed comments
• This paper describes an interesting approach to
  using block-based motion measure for temporal
  texture classification in video. The novelty of
  this work the author claimed is to achieve
  real-time performance and comparative
  classification results by using block-based
  motion description, compared with the
  state-of-art pixel-based motion approaches. The
  methodology described in this paper seems
  theoretically justified, and the analysis of
  experimental results is also sufficient enough.
  However, the structure of this paper seems a
  little unbalanced, since the author put more
  words on related works and experimental analysis,
  instead of approach itself (in section 3). I
  suggest that the author should make the
  literature review more concise and compact, thus
  section 2 may be more readable. The best way of
  revising section 2 is to classify the literatures
  into different classes. The other comments may
  include 1. It is better to numerically (e.g.
  using table) prove real-time" in addition to
  the comparison in section 4.3. You have mentioned
  in abstract that this approach is real-time",
  but did not have evidence to support this point.
  2. Could you explain in the abstract what the
  ratio of importance" means? And why this is
  very important in this scenario? 3. The author
  should simply the symbols used in section 3.1.
  For example, (eta_(x), eta_(y), eta_(t)) has
  appeared quite a few times in this section. It is
  better to make this paper more readable using
  simplified symbols. 4. In section 4.1, put a few
  words to describe the dataset you used in your
  experiments, i.e. the dataset in Ref. 26, e.g.
  the size, the video format, duration, and so on.
  5. Figure 69, please add End For" at the end
  of each For" loop.

26
A Typical Review Form of a Journal

• IEEE Transaction on Circuits and Systems for
  Video Technology
• Typical Review scenario
• Reviewer 2
• Rating
• Technical Content Tutorial 7
• Advance of Theory 5
• Advance of Application 5
• Presentation Clarity 5
• Organization 5 Conciseness 4
• English 5
• Quality of References Completeness 5
• Missing Key References

27
A Typical Review Form of a Journal

• IEEE Transaction on Circuits and Systems for
  Video Technology
• Typical Review scenario
• Reviewer 1
• Detailed comments
• This To my understanding, this paper proposes to
  characterize motion texture in terms of
  block-based motion co-occurrence matrix. This
  proposed approach is strongly motivated by paper
  4. The strength of this paper can be summarized
  in two aspects 1)Block based motion extraction
  which contributes to computational efficiency.
  The authors also argue that block motion can be
  easily obtained due to the currently available
  compression techniques such as MPEG-4 and H.26X,
  which is sound. 2)Explicitly integrating spatial
  and temporal information of motion texture. In
  addition, the proposed method has been evaluated
  with sufficient experiments and analysis.
  However, the novelty of this paper is limited in
  terms of quality of a regular journal paper. I am
  also concerned about the following issues
  1)Sections Introduction and Related work can be
  more succinct. For example, do not try to attack
  almost every aspect of pixel-based motion
  estimation (e.g. optical flow). As to Section
  Related Work, 2)Can you elaborate the claim C
  the sentence after words Section 4.4 in Page
  8? How does the effectiveness fade significantly?
  3)Paragraphs 2 and 3 in Page 9 are redundant,
  since both temporal texture analysis and texture
  synthesis are not very relevant in the context of
  this paper. It is fair enough to mention where it
  has been surveyed for the sake of completeness.
  4)Section 3.2 on separating spatial domain and
  temporal domain is obscure. How does genetic
  algorithm work for feature selection? Why do
  those three matrice form a minimal set? 5)Authors
  argue that the time-space ratio is explicitly
  controlled. However, it seems that how to select
  the ratio is really empirically as stated in
  Section 4.4 of Page 23, which is an obstacle for
  real-time applications. 6)In Figure 5 of Page 17,
  why does the performance drop while k increases?
  7)In Section 4.3, computational complexity is
  evaluated approximately. Is the evaluation based
  on the assumption that every approach uses full
  search for motion estimation? Even the proposed
  approach seems to be much faster than existing
  ones, I wonder how fast the proposed approach is
  in terms of CPU time. Is it practical to achieve
  real-time performance? 8)The pseudo-codes of
  STCN, SFTR, and SSTF would better appear in
  Appendix. 9)As to the Table 2 of Page 24, the
  authors conclude that the proposed approach is
  comparable with existing ones in terms of
  classification accuracy. However, it also can be
  observed that the performance of the proposed
  method using block motion (i.e.97.62) does not
  outperform that of SFTR using Normal flow (i.e.
  99.21). Is the pay-off for computational
  complexity? It is also observed that block-based
  motion estimation only favors the proposed
  approach. Does it mean the dataset favors the
  proposed approach given empirically tuned
  parameters? Why should authors revert the trend
  as claimed in the second sentence of the last
  paragraph, Page 24? In addition, STCN is supposed
  to improve the method of SFTR. However, it is
  observed that it is not true at all for the given
  dataset. 10)The figures (e.g. 3.37) discussed in
  the last paragraph of Page 25 do not match the
  figures in Table 2 (Page 24). 11)A comparison
  also needs to be conducted on Dyntex database as
  shown in Table 2. 12)References are very verbose
  and redundant. In addition, the authors need to
  pay attention to the format and spelling. For
  example, there is a typo in 4 casual should be
  causal.

28
A Typical Review Form of a Journal

• IEEE Transaction on Circuits and Systems for
  Video Technology
• Typical Review scenario
• Reviewer 3
• Rating
• Technical Content Tutorial 7
• Advance of Theory 6
• Advance of Application 8
• Presentation Clarity 8
• Organization 7
• Conciseness 8
• English 8
• Quality of References Completeness 8
• Missing Key References

29
A Typical Review Form of a Journal

• IEEE Transaction on Circuits and Systems for
  Video Technology
• Typical Review scenario
• Reviewer 1
• Detailed comments
• The authors present an approach to temporal
  texture characterization. In this paper, the
  block-based motion measure is used to reduce the
  computational complexity while preserving the
  classification accuracy. Basically, the paper is
  well structured and well written, but some of the
  key issues still need to be addressed, such as
  1) The proposed method used the block-based
  motion vectors. One of its advantage is
  block-based motion vectors readily available for
  MPEG and H.26x video. It is noted that
  block-based motion vectors are simply estimated
  for video coding and compression. So, the
  block-based motion vectors are not reliable for
  motion analysis. "the approximation of true
  motion by the block baed motion vectors .. is
  even stronger" at p4, l12 is not convincing. And
  "therefore, highly likely that the true
  displacement vector will also ..., the criteria
  used by all video coding motion vector estimation
  technique" at p4, l13 from the bottom also has
  the same problem. Video coding just find the
  "best-match" block for compression, it considers
  little about the true displacement vector. That
  makes the underlying assumption of using
  block-based motion vector to character temporal
  textures not obviously reasonable. 2) The
  experiments are implemented on MPEG video data
  and can get real time performance. If the testing
  data is not MPEG video, and need to estimate the
  block-based motion vectors beforehand. What will
  be the processing time, it would be better if the
  author could give a detailed discussion. The
  authors need to give more explaination for some
  technique details, such as 1) p3, l10, "One
  obvious alternative for real time motion
  estimation is to estimate the normal flow", this
  is not "obvious" for other people. Please either
  provide a reference or justify it. 2) p3, l11
  from the bottom, "Any significant reduction in
  the feature extraction... of normal flow vectors,
  however, cannot be effective". Why they cannnot
  be effective, please give the reason in a more
  clearer way, not just give a statement. 3) p10,
  Figure 2, what difference between (a) image frame
  sequenceand (b) normal flow sequence. It would be
  better if the author gives a clearer figure. 4)
  p11, eq.(1). "any arbitrary image processing
  element is assumed independent conditionally", it
  is not obviously reasonable. Please justify such
  an assumption. 5) p12, eq.(5), what do q' and q''
  mean respectively? Any two adjacent frames? 6)
  p13, Table 1, what is the correspondence between
  temporal cliques and spatial cliques? 7) p13, l8
  from the bottom. "Clearly, (0,0,-1),(1,0,0),(0,1,
  0) is a minimal set fulfilling the above
  requirements". Please justify it as it is not
  obviously clear. 8) p14, eq.(8), what do p and q
  means? It is better if the terms used in this
  equation are consistent with previous ones. 9)
  p14, l1 from the bottom, "that are then fused
  into a single distance measure in the second
  stage using weighted Euclidian distance". The
  discussion on weighted Euclidian distance can not
  be found throughout the paper, please add
  relevant content in the revised manuscript. 10)
  p15, eq.(11), v and v' means what? eq.(12) v1 and
  v2 means what? If they have the same meaning, it
  is better to use consistent terms. 11) p17, "All
  the experiments were conducted on MATLAB 6.5.1".
  How to evaluate the performance of processing
  time pratically. It would be better to analyze
  the computational complexity both qualitative and
  quantitatively.

30
A Typical Review Form of a Journal

• IEEE Transaction on Circuits and Systems for
  Video Technology
• Typical Review scenario
• Associate Editor Resubmit after Major Revision
  for Review
• Detailed comments
• The paper -- A temporal texture characterization
  technique using block-based approximated motion
  measure has been reviewed. The paper presents a
  technical approach for temporal texture
  classification in video using block-based motion
  measure. Two major aspects from the paper include
  the block based motion feature extraction with
  certain computation efficiency and the
  exploration of the integrating spatial and
  temporal domain information explicitly. Through
  the review process, some key comments from three
  reviewers can be summarized as follows 1. The
  level of contribution to the temporal texture
  based video classification is limited which
  challenges the novelty in this paper. 2. The
  block based motion estimation and compensation in
  video codec such as h.264 explore the statical
  redundancy rather the true motion estimation.
  Therefore, there is an issue to use block-based
  motion vector from coded video bitstream for
  motion analysis. 3. The sections of Introduction
  and Related work are verbose and far from the
  point of concise and compact for readers. 4.
  References are very verbose and redundant. A
  compact reference list will give reader a focused
  area with some benefits. 5. There are some
  redundant paragraphs which are not necessary to
  be repeated in the paper 6. The pseudo-codes of
  STCN, SFTR, and SSTF would better appear in
  Appendix. 7. Section 3.2 on separating spatial
  domain and temporal domain is obscure. How does
  genetic algorithm work for feature selection? Why
  do those three matrices form a minimal set? 8.
  The methodology described in this paper seems
  theoretically justified, and the analysis of
  experimental results is also sufficient enough 9.
  It is important to describe the advantage and
  disadvantage in terms of your methodology and
  performance which can let readers have their own
  judgement rather make a misleading to them. Based
  on the reviewers comments, I recommend the
  authors to resubmit the paper after major
  revision by addressing all points raised from the
  reviewers. Key references that must be included

31
Writing Tips carry you to a long way

• Reader-oriented writing (good organization,
  logical flow, etc).
• Standard and consistent formatting (professional
  and neat looking).
• Learning from other peoples writing.