A look at two Papers

• These are from KDD 2017.
  • This conference had an acceptance rate of
    • 8.6% for research papers
    • 9.2% for applied papers.
• Clustering Individual Transactional Data for Masses of Users
  • This was in the research track.
  • What does the abstract do?
    • Intoduce the problem.
    • Introduce the algorithm (solution)
    • State that an experiment was successful.
    • State that an application using the algorithm has been developed.
• What does the intoroduction do?
  • Big picture
    • I sure would like a citation justifiying the 5GB of data per year.
    • Which is stated as Gb, but I bet they mean GB.
  • Definition of focus.
    • Not Data Science
    • Not clustering
    • "In this paper we focus on the problem of performing transactional clustering for a large number of dierent datasets."
      • Note that they even narrow the scope further.
      • "Given a collection of transactions, transactional clustering consists in discovering groups of homogeneous transactions which share many common items [30]. "
      • Notice in this paragraph, they continue to narrow the problem.
        • Large number of users.
        • Large amount of data for each user.
        • Need to perform the algorithm for each user.
        • No ability to tune parameters.
    • A discussion of the state of the art.
      • They discuss what has been done before.
        • This is the place that you show that you know what you are talking about.
        • Or at least that you have done your homework
      • xmeans is presented
        • how it solves part of the problem but not the other part.
      • As are two other clustering algorithms.
      • There are many citations in this section.
    • Finally, thier solution is introduced in some detail.
      • Note, this is not a scientific presentation at this point.
      • It is still overview.
    • From the introduction, you may need to go do some background work.
      • Do you know all of the terms?
      • Do you know the algorithms they are comparing this to?
      • Do you understand the problem.
  • This paper has section 2, "Related Work".
    • It begins a more detailed discussion of the problem and how others have attempted to solve it.
    • If you are missing background, this is the palce to pick it up.
    • You might find a different solution to your problem.
    • This section is not always present.
      • See X-means: Extending K-means with Efficient Estimation of the Number of Clusters.
  • Section three
    • This begins a detailed discussion of the problem.
    • Mathematical definitions are given/made.
      • Note the use of math here. (Set theory)
  • In section four they present the algorithm.
    • This is an examination of the algorithm in deatil.
    • And should provide sufficient detail to implement the algorithm if the user chooses.
    • This becomes a tough read
      • Individual sentences are "dense"
      • Large amount of detail.
    • You might skip this section on first read.
  • Section 5 describes experimental results.
    • Again, sufficient information should be present to allow readers to recreate the experiment.
    • They used artificial data.
      • This data was from previous experiments.
      • This allows some comparison with previously reported reuslts.
  • The Real Case study section
  • Slat section, Conclusions
    • Usually includes future direction.
• "Not All Passes Are Created Equal:" Objectively Measuring the Risk and Reward of Passes in Soccer from Tracking Data
  • This was in the applied track.
  • Look for similarities and differences in the paper.