Transcript PowerPoint

Discussion Class 10
Cluster Analysis
1
Question 1: Background
(a) Define cluster analysis. What is the relationship to
automatic classification?
(b) Why might cluster analysis be useful for information
discovery?
(c) What characteristics of data are suitable or unsuitable for
clustering?
(d) Define hierarchical cluster analysis.
2
Question 2: Similarity matrix
(a) What is a similarity measure?
(b) What is a similarity matrix?
(c) Suppose that you are clustering documents based on cooccurrence of citations. Suggest a similarity measure that you
might use.
(d) Explain the ideas behind the inverted file algorithm for
calculating a similarity matrix.
3
Question 3: Hierarchical methods
(a) What do the single link and complete link methods have in
common?
(b) Explain the concept of intercluster similarity.
(c) Explain the concept behind the single link method of cluster
analysis. What are its advantages? What are its disadvantages?
(d) Explain the concept behind the complete link method of
cluster analysis. What are its advantages? What are its
disadvantages?
4
Question 4: Non-hierarchical methods
(a) Under what circumstances might non-hierarchical
cluster analysis by valuable?
(b) Define single pass methods. What problems do
you see with them?
(c) Define reallocation methods. What problems do
you see with them?
5
Question 5: Validation
(a) What is the objective of cluster validation?
(b) Discuss clustering tendency, overlap test, nearest neighbor
test, and density test.
6
Question 6: Document retrieval
Describe and discuss the use of cluster analysis in:
(a) top-down search
(b) bottom-up search
(c) nearest neighbor cluster
7
Question 7: Final question
Dubes and Jain ask the question:
How does one determine whether the results of a clustering
method truly characterize the data?
What is your opinion?
8