UIUC-CS412 \Introduction to Data Mining" (Fall 2015) Final Exam Solution 180 minutes, 150 points 1. Short answer questions [30’]. (a) [3’] What is the advantage of cosine similarity over Euclidean distance in
...
UIUC-CS412 \Introduction to Data Mining" (Fall 2015) Final Exam Solution 180 minutes, 150 points 1. Short answer questions [30’]. (a) [3’] What is the advantage of cosine similarity over Euclidean distance in measuring the document similarity? (b) [3’] Consider a transaction database in which for each single item x, there is at least one transaction that only contains x. Can the number of closed frequent patterns be smaller than the number of frequent patterns? (Answer yes or no; no need to explain.) (c) [3’] In graph mining algorithm gSpan, what key steps were used to avoid redundant computation but not to miss any possible frequent graph? (d) [3’] In GSP, for a sequence pattern of length 100, what is the lower bound of the size of length- 1 candidates? How about that of length-2? And what is the number for all candidates? (Use the tightest lower bound estimation you can get.) (e) [3’] The larger k is, the more accurate the kNN classifier will be. (Choose from True/False, and briefly explain.) (f) [3’] Decision trees are superior to Naive Bayes classifiers since they give more interpretable results. (Choose from True/False, and briefly explain.) (g) [3’] A paper suggests running a single-link hierarchical clustering algorithm for a few iterations on the K-means clustering result. List one advantage of this approach over using K-Means alone. (h) [3’] In some cases, using PCA to reduce dimension can make classification much less precise than it would have been in the original feature space. Explain how this can happen. (i) [3’] In our guest lecture by Matt Ahrens, according to the speaker, what is the takeaway of the lecture, if there is only one thing? (j) [3’] In our guest lecture by Matt Ahrens, he explained that fraudulent advertisements are rather significant. What is the typical rate he reported? a) 1-3% b) 5-10% c) 15-20% d) 30-40%
[Show More]